Multi-converter fault diagnosis method and device of multi-scale rolling correlation feature vector

Abstract

The invention relates to a multi-converter fault diagnosis method and device for a multi-scale rolling correlation feature vector, and the method comprises the steps: enabling a known PWM signal controlled by a power electronic controller to generate i triangular wave signals through a triangular wave algorithm, and carrying out the rolling window type correlation coefficient operation of the i triangular wave signals and a DC bus current ripple signal, the correlation coefficient obtained through operation is further processed through a mean filtering algorithm, so that the waveform of the correlation coefficient is easier to distinguish by the convolutional neural network; according to the device, a bus current ripple signal is firstly collected, then the signal is conditioned, and then correlation coefficient operation is carried out in a DSP (Digital Signal Processor). According to the method, through an improved correlation coefficient algorithm, the calculation speed is improved, so that the fault diagnosis time is shortened; the PWM signal is subjected to linearization processing, so that the correlation coefficient is clearer and smoother, the accuracy of fault diagnosis is higher, and the problems of complex characteristic parameters, model dependence and high fault diagnosis cost in the related electric energy network fault diagnosis technology are solved.

Description

Technical Field

[0001] This invention belongs to the field of DC bus-level fault diagnosis technology, specifically relating to a bus-level multi-converter fault diagnosis method and device based on multi-scale rolling correlation feature vectors, applicable to ring-grid DC microgrids and interconnected power network systems in energy routers or energy routers. Background Technology

[0002] Microgrid fault diagnosis systems can quickly detect and locate faulty components in a microgrid. Their ultimate goal is to ensure the stable and safe operation of the microgrid, guaranteeing the reliability of the output power and facilitating rapid repair after a fault occurs. Therefore, developing a rapid fault diagnosis method for faulty components in a power network system is of great significance, as it not only ensures power quality but also guarantees the security of the power system network.

[0003] Currently, there are some studies on fault diagnosis methods and devices for microgrids, but all have certain shortcomings. Existing technology discloses a fault diagnosis method for islanded DC microgrids based on wavelet sliding window energy, which extracts effective features while reducing the number of wavelet decomposition levels, and achieves fault diagnosis through an islanded DC microgrid fault diagnosis model. However, this method requires "4-level wavelet sliding window energy," resulting in complex feature parameters. Existing technology also discloses a microgrid fault diagnosis method based on whale algorithm-optimized extreme learning machine, which establishes a fault diagnosis model by optimizing the input weights and hidden layer thresholds of the extreme learning machine using the whale algorithm. This method depends on the accuracy of the model, thus affecting the effectiveness and reliability of fault diagnosis. Existing technology also discloses a DC-DC converter fault detection method and device, proposing to collect current ripple signals and PWM signals for incremental correlation coefficient calculation for fault diagnosis. This method requires a PWM signal acquisition circuit, which has high costs; furthermore, the feature parameters of the incremental correlation coefficient algorithm are quite complex.

[0004] Therefore, there is an urgent need to develop a multi-scale rolling correlation feature vector multi-transformer fault diagnosis method and device to effectively solve the above problems. Summary of the Invention

[0005] The purpose of this invention is to provide a bus-level multi-converter fault diagnosis method and apparatus based on multi-scale rolling correlation feature vectors, so as to solve the problems of complex feature parameters, model dependence and high fault diagnosis cost in the prior art.

[0006] The objective of this invention is achieved through the following technical solution:

[0007] A multi-transformer fault diagnosis method using multi-scale rolling correlation feature vectors includes the following steps:

[0008] First, establish an interconnected power network model consisting of n1 photovoltaic power generation units and / or n2 wind power generation units, n3 energy storage devices, n4 loads, and a DC bus. The photovoltaic power generation units are connected to the DC bus via a unidirectional DC-DC module, the wind power generation units are connected to the DC bus via a unidirectional AC-DC module, the energy storage devices are connected to the DC bus via a bidirectional DC-DC module, and the loads are connected to the DC bus via a unidirectional DC / DC or unidirectional DC / AC module.

[0009] Secondly, the power network model is run and a fault experiment is conducted. The triangular wave algorithm generates i triangular wave signals from the known i PWM signals controlled by the power electronic controller in the above network model. At the same time, the bus current ripple signal is compared with the i triangular wave signals by a rolling window correlation coefficient calculation. The calculated correlation coefficients are then filtered.

[0010] Normal correlation coefficient samples and fault-handled correlation coefficient samples are taken as input data. At the same time, the normal data and fault data in the samples are labeled as output data and input into the convolutional neural network for training to obtain a trained convolutional neural network model. Test data is input into the trained convolutional neural network for testing. Finally, the actual output of the convolutional neural network is compared with the predicted output to determine whether the circuit has failed.

[0011] Furthermore, the specific steps include:

[0012] A. Establish an interconnected power network model;

[0013] B. Run the power network model and conduct fault experiments;

[0014] C. Acquire the bus current ripple signal and set the current ripple signal to Y;

[0015] D. Generate i triangular wave signals X1, X2, ..., X from the known i PWM signals controlled by the power electronic controller in the above power network model. i The triangular wave algorithm is defined as follows:

[0016]

[0017] Where k is the number of sampling points; A i S is the number of pulse width time samples in one cycle of the i-th PWM signal; i This represents the total number of samples in one cycle of the i-th PWM signal; n = 1, 2, 3...;

[0018] E. Perform rolling window correlation coefficient calculation on the bus current ripple signal and i triangular wave signals respectively, and calculate the correlation coefficient in real time.

[0019] F. The real-time correlation coefficient calculated in step E is further processed using a mean filtering algorithm;

[0020] G. Take the normal correlation coefficient samples and the correlation coefficient samples after fault handling as input data, and at the same time, configure the labels of the normal data and fault data in the samples as output data, and input them into the convolutional neural network for training to obtain the trained convolutional neural network model.

[0021] H. Use the correlation coefficient samples after filtering in step F as test data and input them into the trained convolutional neural network for testing.

[0022] I. Based on the output of the convolutional neural network, determine whether the circuit has malfunctioned. If there is a malfunction, cut off the power supply through the protection circuit to stop the circuit from operating. If there is no malfunction, shift the data acquisition window forward by one sampling point and return to step E.

[0023] Furthermore, the network topology of the interconnected power network model is a ring-grid DC microgrid, and the PWM signal of each module is individually controlled by its corresponding power electronic controller.

[0024] Furthermore, the interconnected power network model is the internal infrastructure of an energy router or power router, and the PWM signals of all modules are centrally controlled by a power electronic controller.

[0025] Further, in step D, the slope of the rising and falling phases of the triangular wave signal is determined based on the duty cycle of the PWM signal, and the triangular wave signal is further obtained.

[0026] Further, step E specifically includes the following steps:

[0027] First, the nth (n = 1, 2, 3...) sampling period is collected with C... i N is the number of sample values ​​to be sampled, X i (n) ={X i (m),X i (m+1),...,X i (m+C i N-1)} and Y (n) ={Y(m),Y(m+1),...,Y(m+C)} i N-1)}, then according to the correlation coefficient algorithm formula:

[0028]

[0029] Calculate the correlation coefficient for the nth sampling period;

[0030] Among them, C i Here, is the sampling coefficient, N is the number of samples in one cycle of the triangular wave signal, and m is the initial sampling point of the nth sampling cycle.

[0031] The sample value of the correlation coefficient in the (n+1)th sampling period is

[0032] X i (n+1) ={X i (m+1),X i (m+2),...,X i (m+C i N)} and Y (n+1) ={Y(m+1),Y(m+2),...,Y(m+C)} i N)},

[0033] Then, based on the relevant algorithm formula:

[0034]

[0035] Calculate the correlation coefficient for the (n+1)th sampling period.

[0036] Further, step F specifically includes the following steps:

[0037] Take a sample of correlation coefficients for one sampling period

[0038] Will Add them together and then take the average:

[0039]

[0040] The processed correlation coefficient was obtained;

[0041] Take the correlation coefficient sample for the next sampling period.

[0042] Will Add them together and then take the average:

[0043]

[0044] The correlation coefficient after processing is obtained.

[0045] A detection device for a multi-converter fault diagnosis method that implements the above-mentioned multi-scale rolling correlation feature vector is based on a ring-grid DC microgrid network topology of an interconnected power network model. In the ring-grid DC microgrid, the PWM signal of each module is individually controlled by its corresponding power electronic controller. The detection device includes a bus current ripple acquisition module, a signal conditioning module, a bus DSP signal processing module, multiple protection circuit modules, multiple power electronic converter modules, and multiple DSP signal processing modules. Each converter module provides a PWM signal corresponding to a DSP signal processing module. The bus current ripple signal acquisition module is connected to the signal conditioning module, and after being conditioned by the signal conditioning module, it is connected to the bus DSP signal processing module. The bus DSP signal processing module communicates serially with other DSP signal processing modules. Each DSP signal processing module is connected to the corresponding converter module through the protection circuit module.

[0046] A detection device for implementing the above-mentioned multi-scale rolling correlation feature vector bus-level multi-converter fault diagnosis method is based on an interconnected power network model as the internal infrastructure of an energy router or power router. The detection device includes a bus current ripple acquisition module, a signal conditioning module, multiple protection circuit modules, multiple power electronic converter modules, and a DSP signal processing module. The PWM signals given to all converter modules are controlled by a DSP signal processing circuit. The bus current ripple signal acquisition circuit is connected to the signal conditioning circuit, and after being conditioned by the signal conditioning module, it is connected to the DSP signal processing circuit. The DSP signal processing circuit is connected to each converter module through the protection circuit.

[0047] Furthermore, after the bus current ripple signal is acquired by the Hall current sensor, it is amplified by the inverting amplifier and then input into the signal conditioning module for signal conditioning. The PWM signal is converted into a triangular wave signal in the DSP signal processing module and the correlation coefficient is calculated with the conditioned bus current ripple signal.

[0048] Compared with the prior art, the beneficial effects of the present invention are:

[0049] This invention uses fewer feature parameters to determine whether a circuit has failed, overcoming the problem of complex feature parameters and saving fault diagnosis costs. It is independent of the accuracy of the model; the improved correlation coefficient algorithm increases computation speed, thereby reducing fault diagnosis time; the mean filtering algorithm improves the identifiability of the correlation coefficient and the accuracy of circuit fault diagnosis, enabling technicians to maintain faulty circuits promptly; furthermore, energy routers or interconnected power network systems within energy routers represent a typical microgrid architecture; therefore, this DC bus-level fault diagnosis method is equally applicable to both types of networks. Attached Figure Description

[0050] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1 This is a flowchart of the fault diagnosis method;

[0052] Figure 2 This is a structural diagram of the fault detection device described in Example 2;

[0053] Figure 3 This is a structural diagram of the fault detection device described in Example 3;

[0054] Figure 4 This is a circuit diagram for acquiring bus current ripple signals.

[0055] Figure 5 Diagram of the DSP core module;

[0056] Figure 6 This is a photovoltaic simulation diagram;

[0057] Figure 7 This is a comparison chart of normal and faulty data.

[0058] Figure 8 A graph showing the accuracy of fault diagnosis;

[0059] Figure 9 This is a fault response time diagram. Detailed Implementation

[0060] The present invention will be further described below with reference to embodiments:

[0061] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present invention, and not all of the structures.

[0062] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0063] As can be seen from the typical architecture of a DC microgrid, the DC bus can serve as the input and / or output terminal of a DC-DC module, the output terminal of an AC-DC module, and the input terminal of a DC-AC module. When the DC bus is used as an input terminal, current flows from the DC bus into the module; when the DC bus is used as an output terminal, current flows from the module into the DC bus. Each module's connection to the DC bus can be macroscopically considered as a node, and the current at each node must satisfy the Kirchhoff Current-Clearing Principle (KCL). This is the relationship between the DC bus current and the input or output current of each module. Furthermore, the magnitude of the input or output current required by each module is related to the duty cycle of the PWM signal of the corresponding switching transistor. This further demonstrates that the DC bus current is related to the duty cycle of the PWM signal of the switching transistor in each module.

[0064] like Figure 1 As shown, the multi-transformer fault diagnosis method of the multi-scale rolling correlation feature vector of the present invention includes the following steps:

[0065] 1. Run the power network model;

[0066] 2. Acquire the bus current ripple signal and set the current ripple signal to Y;

[0067] 3. Using the triangular wave algorithm, generate i triangular wave signals X1, X2, ..., X from the known i PWM signals controlled by the power electronic controller in the power network system. i The triangular wave algorithm is defined as follows:

[0068]

[0069] Where k is the number of sampling points; A i S is the number of pulse width time samples in one cycle of the i-th PWM signal; i This represents the total number of samples in one cycle of the i-th PWM signal; n = 1, 2, 3...;

[0070] 4. Data acquisition window collects C i With N sampling points, the bus current ripple signal is compared with i triangular wave signals using a rolling window correlation coefficient calculation to obtain the correlation coefficient in real time.

[0071] 5. The real-time correlation coefficient calculated in step 4 is further processed using a mean filtering algorithm;

[0072] 6. Take the normal correlation coefficient samples and the correlation coefficient samples after fault handling as input data, and assign labels to the normal data and fault data in the samples as output data. Input them into the convolutional neural network for training to obtain the trained convolutional neural network model.

[0073] 7. Use the correlation coefficient samples after filtering in step 5 as test data and input them into the trained convolutional neural network for testing.

[0074] 8. Determine whether the circuit has malfunctioned based on the output of the convolutional neural network.

[0075] If a fault occurs, the power supply will be cut off through the protection circuit to stop the circuit from operating and protect the circuit.

[0076] If there is no fault, the data acquisition window moves forward one sampling point and returns to step 4. The fixed length is C. i The data acquisition window of N is shifted forward by one sampling point, and this additional sampling point is combined with the previously acquired (C) i With N-1 sampling points, calculate the expected change between the previous and next two periods, and then substitute it into the correlation coefficient algorithm to calculate the correlation coefficient for the new period.

[0077] In step 4, the formula for calculating the correlation coefficient is as follows:

[0078] First, the nth (n = 1, 2, 3...) sampling period is collected with C... i N(C i The variable is defined in this invention as the sampling coefficient, where N is the number of samples in one cycle of the triangular wave signal, and X is the sample value of the number of samples. i (n) ={X i (m),X i (m+1),...,X i (m+C i N-1)} and Y (n) ={Y(m),Y(m+1),...,Y(m+C)} i (N-1)} (m is the initial sampling point of the nth sampling period), and then according to the correlation coefficient algorithm formula:

[0079]

[0080] Calculate the correlation coefficient for the nth sampling period;

[0081] From the nth to the (n+1th)th sampling period, rolling sampling is implemented to collect sample values. The number of sampling points remains unchanged, but the data acquisition window is shifted forward by one sampling point. The sample value of the correlation coefficient in the (n+1th)th sampling period is X. i (n+1) ={X i (m+1),X i (m+2),...,X i (m+C i N)} and Y (n+1)={Y(m+1),Y(m+2),...,Y(m+C)} i N)}, then according to the formula:

[0082]

[0083] E (n+1) [X i ] = E (n) [X i ]+ΔE (n+1)-(n) [X i ]

[0084] E (n+1) [Y] = E (n) [Y]+ΔE (n+1)-(n) [Y]

[0085]

[0086] E (n+1) [Y 2 ] = E (n) [Y 2 ]+ΔE (n+1)-(n) [Y 2 ]

[0087] E (n+1) [X i Y]=E (n) [X i Y]+ΔE (n+1)-(n) [X i Y]

[0088] The improved correlation coefficient calculation formula for the (n+1)th sampling period is as follows:

[0089]

[0090] E (n) [X i ]、E (n) [Y]、E (n) [X i 2 ]、E (n) [Y 2 ] and E (n) [X i Y] are respectively X i Y, X i 2 Y 2 and X i Y in the nth acquisition cycle uses C i N is the expected number of samples; E (n+1) [X i ]、E (n+1) [Y]、E(n+1) [X i 2 ]、E (n+1) [Y 2 ] and E (n+1) [X i Y] are respectively X i Y, X i 2 Y 2 and X i Y in the (n+1)th acquisition cycle uses C i N is the expected number of samples; ΔE (n+1)-(n) [X i ]、ΔE (n+1)-(n) [Y]、 ΔE (n+1)-(n) [Y 2 ]、ΔE (n+1)-(n) [X i Y] are respectively X i Y Y 2 X i Y is sampled from the nth to the (n+1th)th period with C i N is the expected change in the number of samples.

[0091] This invention presents a multi-converter fault diagnosis method using multi-scale rolling correlation feature vectors. It utilizes Simulink in MATLAB as the fault diagnosis platform and establishes an interconnected power network model. A triangular wave algorithm generates *i* triangular wave signals from the known PWM signal controlled by the power electronic controller. By performing a rolling window correlation coefficient calculation between these *i* triangular wave signals and the DC bus current ripple signal, the obtained correlation coefficients are further processed using a mean filtering algorithm, making the correlation coefficient waveforms easier for convolutional neural networks to distinguish. The fault diagnosis method provided by this invention improves the calculation speed through an improved correlation coefficient algorithm, thereby reducing fault diagnosis time. Furthermore, by linearizing the PWM signal, the correlation coefficients become clearer and smoother, resulting in higher fault diagnosis accuracy. This solves the problems of complex feature parameters, model dependence, and high fault diagnosis costs in related power network fault diagnosis technologies.

[0092] like Figure 2 As shown, the detection device for the multi-converter fault diagnosis method using multi-scale rolling correlation feature vectors of the present invention is based on a ring-grid DC microgrid network topology of an interconnected power network model. In the ring-grid DC microgrid, the PWM signal of each module is individually controlled by its corresponding power electronic controller. The detection device includes a bus current ripple acquisition module, a signal conditioning module, a bus DSP signal processing module, multiple protection circuit modules, multiple power electronic converter modules, and multiple DSP signal processing modules.

[0093] Each converter module provides a PWM signal corresponding to a DSP signal processing module; the bus current ripple signal acquisition module is connected to the signal conditioning module, and after being conditioned by the signal conditioning module, it is connected to the bus DSP signal processing module. The bus DSP signal processing module communicates serially with other DSP signal processing modules; each DSP signal processing module is connected to the corresponding converter module through a protection circuit module.

[0094] like Figure 3 As shown, the detection device for the multi-converter fault diagnosis method using multi-scale rolling correlation feature vectors of the present invention can also be based on an interconnected power network model as the internal infrastructure of an energy router or power router. The detection device includes a bus current ripple acquisition module, a signal conditioning module, multiple protection circuit modules, multiple power electronic converter modules, and a DSP signal processing module.

[0095] The PWM signals given to all converter modules are controlled by a DSP signal processing circuit; the bus current ripple signal acquisition circuit is connected to the signal conditioning circuit, and after being conditioned by the signal conditioning module, it is connected to the DSP signal processing circuit. The DSP signal processing circuit is connected to each converter module through a protection circuit.

[0096] like Figure 4 As shown, the bus current ripple signal acquisition circuit diagram shows that after the bus current ripple signal is acquired by the Hall current sensor, it is amplified by the inverting amplifier and then input into the signal conditioning module for signal conditioning.

[0097] like Figure 5 As shown in the diagram, the DSP core module converts the PWM signal into a triangular wave signal in the DSP signal processing module and performs correlation coefficient calculation with the conditioned bus current ripple signal.

[0098] like Figure 6 As shown in the photovoltaic simulation diagram, two photovoltaic simulation sources are each connected to a boost circuit. The output terminals of the two boost circuits are connected to establish a DC bus. Then, the battery is connected to a bidirectional DC / DC converter and connected to the DC bus.

[0099] like Figure 7The diagram shows a comparison of normal and fault data. It is divided into three parts: the first part is the DC bus ripple signal waveform after removing the DC component; the second part is the waveform of correlation coefficient 1, obtained by performing correlation coefficient calculation between the triangular wave signal 1 and the DC bus ripple signal after removing the DC component, followed by mean filtering; the third part is the waveform of correlation coefficient 2, obtained by performing correlation coefficient calculation between the triangular wave signal 2 and the DC bus ripple signal after removing the DC component, followed by mean filtering. The diagram also marks the fault point corresponding to the time of fault occurrence and the point where the protection circuit trips after the fault is detected.

[0100] like Figure 8 As shown in the figure, the accuracy of CNN fault diagnosis can reach over 95% after CNN training.

[0101] like Figure 9 As shown in the CNN fault response time graph, it can be seen that the fault point detected by the CNN is about 64 points after the actual fault point, that is, the CNN can detect the fault after about 32us.

[0102] This invention generates i triangular wave signals by linearizing i known PWM signals controlled by a power electronic controller in a power grid, and performs correlation coefficient calculations between the bus current ripple signal and the i triangular wave signals respectively.

[0103] Example 1

[0104] A multi-scale rolling correlation feature vector multi-converter fault diagnosis method uses Simulink in MATLAB as the fault diagnosis platform. First, an interconnected power network model is established, consisting of n1 photovoltaic power generation units and / or n2 wind power generation units, n3 energy storage devices, n4 loads, and a DC bus. The photovoltaic power generation units are connected to the DC bus through a unidirectional DC-DC module, the wind power generation units are connected to the DC bus through a unidirectional AC-DC module, the energy storage devices are connected to the DC bus through a bidirectional DC-DC module, and the loads are connected to the DC bus through a unidirectional DC / DC or unidirectional DC / AC module.

[0105] Secondly, the power network model is run and a fault experiment is conducted. The triangular wave algorithm generates i triangular wave signals from the known i PWM signals controlled by the power electronic controller in the above network model. At the same time, the bus current ripple signal is compared with the i triangular wave signals by a rolling window correlation coefficient calculation. The calculated correlation coefficients are then filtered.

[0106] Normal correlation coefficient samples and fault-handled correlation coefficient samples are taken as input data. At the same time, the normal data and fault data in the samples are labeled as output data and input into the convolutional neural network for training to obtain a trained convolutional neural network model. Test data is input into the trained convolutional neural network for testing. Finally, the actual output of the convolutional neural network is compared with the predicted output to determine whether the circuit has failed.

[0107] Specifically, the following steps are included:

[0108] 1. Run the power network model;

[0109] 2. Acquire the bus current ripple signal and set the current ripple signal to Y;

[0110] 3. Generate i triangular wave signals from the known i PWM signals controlled by the power electronic controller in the power network system. The triangular wave algorithm is defined as follows:

[0111] Let A i S is the number of pulse width time samples in one cycle of the i-th PWM signal. i This represents the number of samples in one cycle of the i-th PWM signal. The rising phase of the triangular wave corresponds to the waveform of the PWM signal during its pulse width time, and the falling phase of the triangular wave corresponds to the waveform of the PWM signal during its non-pulse width time.

[0112] During the rising phase, if the amplitude of the triangular wave is set to 1, then its slope is...

[0113] During the descent phase, if we assume the minimum value of the triangular wave is 0, then its slope is...

[0114] Therefore, the triangular wave algorithm is as follows:

[0115] First cycle:

[0116] Second cycle:

[0117] The nth period:

[0118] Where k is the number of sampling points; n = 1, 2, 3...

[0119] 4. In the real-time correlation coefficient calculation, the correlation coefficient calculation formula is as follows:

[0120]

[0121] COV(X i ,Y)=E (n) [X i Y]-E(n) [X i E (n) [Y] (2)

[0122] D[X i ] = COV(X i ,X i ) = E (n) [X i 2 ]-(E (n) [X i ]) 2 (3)

[0123] D[Y]=COV(Y,Y)=E (n) [Y 2 ]-(E (n) [Y]) 2 (4)

[0124] in For X i The correlation coefficient between Y and Y in the nth sampling period;

[0125] The difference between the sampling period and the PWM period is that: there are no overlapping sampling points between two adjacent periods of the PWM period; there are (N-1) overlapping sampling points between two adjacent periods of the sampling period, and the latter period is represented by the former period being shifted forward by one sampling point.

[0126] COV(X i (,Y) is X i The covariance of Y, D[X] i ] and D[Y] are X i The variance of Y;

[0127] X i For sample {X i (1),X i (2).......X i The set of (m).......};

[0128] Y is the set of samples {Y(1),Y(2)......Y(m)......};

[0129] E (n) [X i ]、E (n) [Y]、E (n) [X i 2 ]、E (n) [Y 2 ] and E (n) [X i Y] are respectively X i Y, Xi 2 Y 2 and X i Y in the nth sampling period uses C i N is the expected number of samples. The E (n) [X i The sample value of ] is X i (n) ={X i (m),X i (m+1),...,X i (m+C i The calculation formula for N-1)} is as follows:

[0130]

[0131] Where m is the initial sampling point of the nth sampling period.

[0132] Because the frequencies of the PWM signals controlled by the power electronic controllers are not exactly the same, the number of sampling points collected during correlation calculations is also different and should be considered a variable. Therefore, in this patent, C is the number of samples N per cycle of the triangular wave signal. i times (C) i Let i be a variable, and i corresponds to the i-th triangular wave signal generated by the i-th PWM signal. The correlation coefficient is calculated by taking the number of samples in one sampling period.

[0133] E (n+1) [X i ]、E (n+1) [Y]、E (n+1) [X i 2 ]、E (n+1) [Y 2 ] and E (n+1) [X i Y] are respectively X i Y, X i 2 Y 2 and X i Y in the (n+1)th acquisition cycle uses C i N is the expected number of samples. From the nth to the (n+1)th sampling period, rolling sampling is implemented to collect sample values. The number of sampling points remains unchanged, and the data acquisition window is shifted forward by one sampling point. The E... (n+1) [X i The sample value of ] is X i (n+1) ={X i (m+1),X i (m+2),...,X i (m+C iThe calculation formula for N)} is as follows:

[0134]

[0135] Combining equations (1) to (4), we obtain the formula for calculating the correlation coefficient in the nth sampling period:

[0136]

[0137] The formula for calculating the correlation coefficient in the (n+1)th sampling period is:

[0138]

[0139] In formula (8):

[0140] E (n+1) [X i ] = E (n) [X i ]+ΔE (n+1)-(n) [X i (9)

[0141] Where ΔE (n+1)-(n) [X i [Sample X] i From the nth to the (n+1th)th sampling period, C i N is the expected change in the number of samples.

[0142] Combining equations (5), (6), and (9), we get:

[0143]

[0144] Thus, in the (n+1)th sampling period, the mathematical expectation E (n+1) [X i It can be calculated by combining equations (13) and (14) in a simpler way.

[0145] Similarly:

[0146] E (n) The sample value of [Y] is Y (n) ={Y(m),Y(m+1),...,Y(m+C)} i The calculation formula for N-1)} is as follows:

[0147]

[0148] In the nth to n+1th sampling periods, rolling sampling is implemented to collect sample values, E (n+1) The sample value of [Y] is Y (n+1) ={Y(m+1),Y(m+2),...,Y(m+C)} iThe calculation formula for N)} is as follows:

[0149]

[0150] E (n+1) [Y] = E (n) [Y]+ΔE (n+1)-(n) [Y] (13)

[0151] Sample Y is sampled from the nth to the (n+1)th sampling period with C i N is the mathematical expectation of the change in the number of samples, ΔE. (n+1)-(n) [Y] is obtained by combining formulas (11), (12), and (13).

[0152]

[0153] Thus, in the (n+1)th sampling period, the mathematical expectation E (n+1) [Y] can be calculated by combining the simpler formulas (13) and (14).

[0154] The sample value is The calculation formula is as follows:

[0155]

[0156] From the nth to the (n+1th)th sampling period, rolling sampling is implemented to collect sample values. The sample value is The calculation formula is as follows:

[0157]

[0158]

[0159] Formula (15), formula (16), and formula (17) are combined to obtain

[0160] sample From the nth to the (n+1th)th sampling period, C i The formula for calculating the expected change of N, where N is the number of samples, is:

[0161]

[0162] Thus, in the (n+1)th sampling period, the mathematical expectation is... It can be calculated by combining the simpler formulas (17) and (18).

[0163] E (n) [Y 2 The sample value of ] is Y 2(n) ={Y 2(m),Y 2 (m+1),...,Y 2 (m+C i The calculation formula for N-1)} is as follows:

[0164]

[0165] In the nth to n+1th sampling periods, rolling sampling is implemented to collect sample values, E (n+1) [Y 2 The sample value of ] is Y 2(n+1) ={Y 2 (m+1),Y 2 (m+2),...,Y 2 (m+C i The calculation formula for N)} is as follows:

[0166]

[0167] E (n+1) [Y 2 ] = E (n) [Y 2 ]+ΔE (n+1)-(n) [Y 2 ] (twenty one)

[0168] Formula (15), formula (16), and formula (17) are combined to obtain

[0169] Sample Y 2 From the nth to the (n+1th)th sampling period, C i The formula for calculating the expected change of N, where N is the number of samples, is:

[0170]

[0171] Thus, in the (n+1)th sampling period, the mathematical expectation E (n+1) [Y 2 It can be calculated by combining the simpler formulas (21) and (22).

[0172] E (n) [X i The sampled value of Y is X i Y (n) ={X i Y(m),X i Y(m+1),...,X i Y(m+C i The calculation formula for N-1)} is as follows:

[0173]

[0174] In the nth to n+1th sampling periods, rolling sampling is implemented to collect sample values, E (n+1) [X i The sampled value of Y is X i Y (n +1) ={X i Y(m+1),X i Y(m+2),...,X i Y(m+C i The calculation formula for N)} is as follows:

[0175]

[0176] E (n+1) [X i Y]=E (n) [X i Y]+ΔE (n+1)-(n) [X i Y] (25)

[0177] Formula (23), formula (24), and formula (25) are combined to obtain

[0178] Sample X i Y is sampled from the nth to the (n+1th)th period with C i The formula for calculating the expected change of N, where N is the number of samples, is:

[0179]

[0180] Thus, in the (n+1)th sampling period, the mathematical expectation E (n+1) [X i Y] can be calculated by combining the simpler formulas (25) and (26).

[0181] Therefore, the formula for calculating the correlation coefficient in the (n+1)th sampling period can be further written as the following improved formula for calculating the correlation coefficient:

[0182]

[0183] Knowing the correlation coefficient for the current period, when calculating the correlation coefficient for the next period, substituting equations (10)(14)(18)(22)(26) into the improved correlation coefficient calculation formula will greatly speed up the calculation.

[0184] Example 2

[0185] A detection device for a multi-converter fault diagnosis method based on multi-scale rolling correlation feature vectors, wherein the network topology of the interconnected power network model is a ring-type DC microgrid.

[0186] The interconnected power network model consists of n1 photovoltaic power generation units and / or n2 wind power generation units, n3 energy storage devices, n4 loads, and a DC bus. The photovoltaic power generation units are connected to the DC bus via unidirectional DC-DC modules, the wind power generation units are connected to the DC bus via unidirectional AC-DC modules, the energy storage devices are connected to the DC bus via bidirectional DC-DC modules, and the loads are connected to the DC bus via unidirectional DC / DC or unidirectional DC / AC modules.

[0187] In this detection device, the PWM signal of each module is individually controlled by its corresponding power electronic controller. The detection device includes a bus current ripple acquisition module, a signal conditioning module, a bus DSP signal processing module, multiple protection circuit modules, multiple power electronic converter modules, and multiple DSP signal processing modules. Each converter module provides a PWM signal corresponding to a DSP signal processing module; the bus current ripple signal acquisition module is connected to the signal conditioning module, and after conditioning, it is connected to the bus DSP signal processing module. The bus DSP signal processing module communicates serially with other DSP signal processing modules; each DSP signal processing module is connected to its corresponding converter module via a protection circuit module.

[0188] Example 3

[0189] A detection device for a multi-converter fault diagnosis method based on multi-scale rolling correlation feature vectors can also be used as the internal infrastructure of an energy router or power router based on an interconnected power network model.

[0190] The interconnected power network model consists of n1 photovoltaic power generation units and / or n2 wind power generation units, n3 energy storage devices, n4 loads, and a DC bus. The photovoltaic power generation units are connected to the DC bus via unidirectional DC-DC modules, the wind power generation units are connected to the DC bus via unidirectional AC-DC modules, the energy storage devices are connected to the DC bus via bidirectional DC-DC modules, and the loads are connected to the DC bus via unidirectional DC / DC or unidirectional DC / AC modules.

[0191] The detection device includes a bus current ripple acquisition module, a signal conditioning module, multiple protection circuit modules, multiple power electronic converter modules, and a DSP signal processing module. The PWM signals supplied to all converter modules are controlled by a DSP signal processing circuit. The bus current ripple signal acquisition circuit is connected to the signal conditioning circuit, and after conditioning by the signal conditioning module, it is connected to the DSP signal processing circuit. The DSP signal processing circuit is then connected to each converter module via protection circuits.

[0192] Note that the above description is merely a preferred embodiment of the present invention and the technical principles employed. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in detail through the above embodiments, the present invention is not limited to the above embodiments, and may include many other equivalent embodiments without departing from the concept of the present invention, the scope of which is determined by the scope of the appended claims.

Claims

1. A multi-transformer fault diagnosis method using multi-scale rolling correlation feature vectors, characterized in that, Includes the following steps: First, establish an interconnected power network model consisting of n1 photovoltaic power generation units and / or n2 wind power generation units, n3 energy storage devices, n4 loads, and a DC bus. The photovoltaic power generation units are connected to the DC bus via a unidirectional DC-DC module, the wind power generation units are connected to the DC bus via a unidirectional AC-DC module, the energy storage devices are connected to the DC bus via a bidirectional DC-DC module, and the loads are connected to the DC bus via a unidirectional DC / DC or unidirectional DC / AC module. Secondly, the power network model is run and a fault experiment is conducted. The triangular wave algorithm generates i triangular wave signals from the known i PWM signals controlled by the power electronic controller in the above network model. At the same time, the bus current ripple signal is compared with the i triangular wave signals by a rolling window correlation coefficient calculation. The calculated correlation coefficients are then filtered. Normal correlation coefficient samples and fault-handled correlation coefficient samples are taken as input data. At the same time, the normal data and fault data in the samples are labeled as output data and input into the convolutional neural network for training to obtain a trained convolutional neural network model. Test data is input into the trained convolutional neural network for testing. Finally, the actual output of the convolutional neural network is compared with the predicted output to determine whether the circuit has failed.

2. The multi-transformer fault diagnosis method for multi-scale rolling correlation feature vectors according to claim 1, characterized in that, Specifically, the following steps are included: A. Establish an interconnected power network model; B. Run the power network model and conduct fault experiments; C. Acquire the bus current ripple signal and set the current ripple signal to Y; D. Generate i triangular wave signals X1, X2, ..., X from the known i PWM signals controlled by the power electronic controller in the above power network model. i The triangular wave algorithm is defined as follows: Where k is the number of sampling points; A i S is the number of pulse width time samples in one cycle of the i-th PWM signal; i This represents the total number of samples in one cycle of the i-th PWM signal; n = 1, 2, 3...; E. Perform rolling window correlation coefficient calculation on the bus current ripple signal and i triangular wave signals respectively, and calculate the correlation coefficient in real time. F. The real-time correlation coefficient calculated in step E is further processed using a mean filtering algorithm; G. Take the normal correlation coefficient samples and the correlation coefficient samples after fault handling as input data, and at the same time, configure the labels of the normal data and fault data in the samples as output data, and input them into the convolutional neural network for training to obtain the trained convolutional neural network model. H. Use the correlation coefficient samples after filtering in step F as test data and input them into the trained convolutional neural network for testing. I. Based on the output of the convolutional neural network, determine whether the circuit has malfunctioned. If there is a malfunction, cut off the power supply through the protection circuit to stop the circuit from operating. If there is no malfunction, shift the data acquisition window forward by one sampling point and return to step E.

3. The multi-transformer fault diagnosis method for multi-scale rolling correlation feature vectors according to claim 1, characterized in that: The interconnected power network model has a ring-type DC microgrid topology, and the PWM signal of each module is individually controlled by its corresponding power electronic controller.

4. The multi-transformer fault diagnosis method for multi-scale rolling correlation feature vectors according to claim 1, characterized in that: The interconnected power network model is the internal infrastructure of an energy router or power router, and the PWM signals of all modules are centrally controlled by a power electronic controller.

5. The multi-transformer fault diagnosis method for multi-scale rolling correlation feature vectors according to claim 1, characterized in that: Step D: Determine the slope of the rising and falling phases of the triangular wave signal based on the duty cycle of the PWM signal, and further derive the triangular wave signal.

6. The multi-transformer fault diagnosis method for multi-scale rolling correlation feature vectors according to claim 1, characterized in that, Step E specifically includes the following steps: First, the nth (n = 1, 2, 3...) sampling period is collected with C... i N is the number of sample values ​​to be sampled, X i (n) ={X i (m),X i (m+1),...,X i (m+C i N-1)} and Y (n) ={Y(m),Y(m+1),...,Y(m+C)} i N-1)}, then according to the correlation coefficient algorithm formula: Calculate the correlation coefficient for the nth sampling period; Among them, C i Here, is the sampling coefficient, N is the number of samples in one cycle of the triangular wave signal, and m is the initial sampling point of the nth sampling cycle. The sample value of the correlation coefficient in the (n+1)th sampling period is X i (n+1) = {X i (m + 1), X i (m + 2),..., X i (m + C i N)} and Y (n+1) = {Y(m + 1), Y(m + 2),..., Y(m + C i N)}, Then, based on the relevant algorithm formula: Calculate the correlation coefficient for the (n+1)th sampling period.

7. The multi-transformer fault diagnosis method for multi-scale rolling correlation feature vectors according to claim 1, characterized in that, Step F specifically includes the following steps: Take a sample of correlation coefficients for one sampling period. Will Add them together and then take the average: The processed correlation coefficient was obtained; Take the correlation coefficient sample for the next sampling period. Will Add them together and then take the average: The correlation coefficient after processing is obtained.

8. A detection device for implementing the multi-scale rolling correlation feature vector fault diagnosis method of claim 1, characterized in that: The network topology based on the interconnected power network model is a ring-grid DC microgrid. In the ring-grid DC microgrid, the PWM signal of each module is individually controlled by its corresponding power electronic controller. The detection device includes a bus current ripple acquisition module, a signal conditioning module, a bus DSP signal processing module, multiple protection circuit modules, multiple power electronic converter modules, and multiple DSP signal processing modules. Each converter module provides a PWM signal corresponding to a DSP signal processing module. The bus current ripple signal acquisition module is connected to the signal conditioning module, and after being conditioned by the signal conditioning module, it is connected to the bus DSP signal processing module. The bus DSP signal processing module communicates serially with other DSP signal processing modules. Each DSP signal processing module is connected to the corresponding converter module through the protection circuit module.

9. A detection device for implementing the multi-scale rolling correlation feature vector fault diagnosis method of claim 1, characterized in that: The interconnected power network model serves as the internal infrastructure of the energy router or power router. The detection device includes a bus current ripple acquisition module, a signal conditioning module, multiple protection circuit modules, multiple power electronic converter modules, and a DSP signal processing module. The PWM signal given to all converter modules is controlled by a DSP signal processing circuit. The bus current ripple signal acquisition circuit is connected to the signal conditioning circuit, and after being conditioned by the signal conditioning module, it is connected to the DSP signal processing circuit. The DSP signal processing circuit is connected to each converter module through the protection circuit.

10. The detection device for a multi-transformer fault diagnosis method using multi-scale rolling correlation feature vectors according to claims 8 and 9, characterized in that: After the bus current ripple signal is acquired by the Hall current sensor, it is amplified by the inverting amplifier and then input into the signal conditioning module for signal conditioning. The PWM signal is converted into a triangular wave signal in the DSP signal processing module and the correlation coefficient is calculated with the conditioned bus current ripple signal.

Images