A method and system for accurately diagnosing a fault of a hybrid grid-connected power generation system of a medium and far sea wind power generation, and a medium

Abstract

The present application relates to the technical field of power system automation, and more particularly to a method and system for accurately diagnosing faults in a mid-ocean wind power hybrid grid-connected system, and a medium, the method comprising: obtaining three-phase electrical quantity signals of the mid-ocean wind power hybrid grid-connected system under different fault states, and constructing a fault sample set according to the three-phase electrical quantity signals and corresponding fault categories; performing multi-domain feature extraction and three-phase feature fusion on the three-phase electrical quantity signals in the fault sample set to obtain comprehensive fault features; training a fault classification model based on the comprehensive fault features and the fault categories; obtaining three-phase electrical quantity signals to be diagnosed, obtaining comprehensive fault features to be diagnosed in the manner of multi-domain feature extraction and three-phase feature fusion, and outputting a fault diagnosis result using the fault classification model. Through the present application, the problem of mutual coupling of electrical quantity features of different types of faults in the mid-ocean wind power hybrid grid-connected system, which makes it difficult to accurately and uniformly identify the faults, is effectively solved.

Description

Technical Field

[0001] This invention relates to the field of power system automation technology, and in particular to a method, system and medium for accurate fault diagnosis of hybrid grid-connected wind power in offshore and medium-sized cities. Background Technology

[0002] The COSCO offshore wind power hybrid grid-connected system connects offshore wind power to the onshore power grid through different transmission channels. Faults in the system's converter equipment and grid-side faults can cause changes in three-phase electrical quantities. Among existing technologies, CN113341345A extracts the capacitor voltage characteristics of modular multilevel converter submodules and uses random forests to identify open-circuit faults, primarily targeting single converter equipment; CN114325236A extracts the spectral characteristics of distribution network fault recordings and uses random forests to identify faults, primarily targeting general distribution network faults.

[0003] However, COSCO offshore wind power hybrid grid-connected systems are characterized by multi-channel parallel operation and the combined effect of multiple types of faults. The characteristics of different faults in the three-phase electrical quantities are easily coupled with each other, and existing technologies are unable to fully characterize the differences of faults and achieve unified and accurate identification.

[0004] The information disclosed in this background section is intended only to enhance the understanding of the general background of this disclosure and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0005] This invention provides a method, system, and medium for accurate fault diagnosis of hybrid grid-connected wind power in offshore and medium-sized cities, which can effectively solve the problems in the background art.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for accurate fault diagnosis of hybrid grid-connected wind power in offshore and medium-sized cities, the method comprising: Acquire the three-phase electrical quantity signals of the COSCO offshore wind power hybrid grid-connected system under different fault states, and construct a fault sample set based on the three-phase electrical quantity signals and the corresponding fault categories; Multi-domain feature extraction and three-phase feature fusion are performed on the three-phase electrical quantity signals in the fault sample set to obtain comprehensive fault features; A fault classification model is trained based on the comprehensive fault characteristics and the fault categories. The three-phase electrical quantity signals to be diagnosed are acquired, and the comprehensive fault features to be diagnosed are obtained according to the multi-domain feature extraction and the three-phase feature fusion method. The fault diagnosis results are then output using the fault classification model.

[0007] Furthermore, the offshore wind power hybrid grid-connected system includes an offshore wind farm power collection system, a flexible DC grid-connected channel, a low-frequency AC grid-connected channel, a power frequency AC grid-connected channel, a common coupling point, and an onshore power grid. The flexible DC grid-connected channel, the low-frequency AC grid-connected channel, and the power frequency AC grid-connected channel are connected in parallel between the offshore wind farm power collection system and the common coupling point, and the common coupling point is connected to the onshore power grid.

[0008] Furthermore, the flexible DC grid-connected channel is equipped with a modular multilevel converter, and the low-frequency AC grid-connected channel is equipped with a modular multilevel matrix converter.

[0009] Furthermore, the different fault states include the bridge arm open circuit fault of the modular multilevel matrix converter, the single-phase ground fault and the three-phase ground fault of the COSCO offshore wind power hybrid grid-connected system.

[0010] Furthermore, the three-phase electrical quantity signal is a three-phase current signal.

[0011] Furthermore, the multi-domain feature extraction includes performing time-domain feature extraction, frequency-domain feature extraction, and waveform feature extraction on each phase current signal in the three-phase current signals to obtain corresponding phase fault features.

[0012] Furthermore, the three-phase feature fusion includes splicing the fault features of each phase according to the phase sequence to obtain the comprehensive fault features.

[0013] Furthermore, the fault classification model is a random forest classification model, which is used to classify the bridge arm open circuit fault, the single-phase ground short circuit fault and the three-phase ground short circuit fault according to the comprehensive fault characteristics, and to determine the faulty bridge arm position in the classification results corresponding to the bridge arm open circuit fault.

[0014] A precise fault diagnosis system for hybrid grid-connected offshore wind power, the system comprising: The fault sample construction module acquires the three-phase electrical quantity signals of the COSCO offshore wind power hybrid grid-connected system under different fault states, and constructs a fault sample set based on the three-phase electrical quantity signals and the corresponding fault categories. The multi-domain feature fusion module performs multi-domain feature extraction and three-phase feature fusion on the three-phase electrical quantity signals in the fault sample set to obtain comprehensive fault features; The classification model training module trains a fault classification model based on comprehensive fault features and fault categories. The fault diagnosis output module acquires the three-phase electrical quantity signals to be diagnosed, obtains the comprehensive fault characteristics to be diagnosed by multi-domain feature extraction and three-phase feature fusion, and outputs the fault diagnosis results using the fault classification model.

[0015] A computer-readable storage medium storing a computer program, the computer program including program instructions, which, when executed by a processor, can implement the aforementioned method for accurate fault diagnosis of hybrid grid-connected offshore wind power.

[0016] The technical solution of this invention can achieve the following technical effects: By collecting three-phase electrical signals under different fault conditions in the COSCO-COSCO offshore wind power hybrid grid-connected system, the system sequentially completes fault sample construction, multi-domain feature extraction, three-phase feature fusion, and fault classification model training. The trained fault classification model is then used to identify the fault in the signal to be diagnosed, effectively solving the problem that the electrical characteristics of different types of faults in the COSCO-COSCO offshore wind power hybrid grid-connected system are coupled with each other and difficult to accurately and uniformly identify.

[0017] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 A flowchart illustrating the precise fault diagnosis method for COSCO's hybrid grid-connected wind power system; Figure 2 This is a diagram showing the overall structure of a COSCO offshore wind power hybrid grid-connected system. Figure 3 Flowchart for training and diagnosing a random forest classification model; Figure 4 This is a schematic diagram of frequency domain characteristics; Figure 5 This is a schematic diagram of the fault diagnosis classification results. Detailed Implementation

[0020] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0022] Example 1; like Figure 1 and Figure 2 As shown, this application provides a method for accurate fault diagnosis of hybrid grid-connected wind power in offshore and medium-sized cities. The method includes: S10: Obtain the three-phase electrical quantity signals of the COSCO offshore wind power hybrid grid-connected system under different fault states, and construct a fault sample set based on the three-phase electrical quantity signals and the corresponding fault categories; S20: Perform multi-domain feature extraction and three-phase feature fusion on the three-phase electrical quantity signals in the fault sample set to obtain comprehensive fault characteristics; S30: Training a fault classification model based on comprehensive fault features and fault categories; S40: Acquire the three-phase electrical quantity signals to be diagnosed, obtain the comprehensive fault characteristics to be diagnosed by multi-domain feature extraction and three-phase feature fusion, and output the fault diagnosis results using the fault classification model.

[0023] Specifically, this embodiment provides a fault diagnosis method for a hybrid grid-connected offshore wind power system. The hybrid grid-connected offshore wind power system includes an offshore wind farm collection system, a flexible DC grid-connected channel, a low-frequency AC grid-connected channel, a power frequency AC grid-connected channel, a common coupling point, and an onshore power grid. The three types of grid-connected channels are connected in parallel between the offshore wind farm collection system and the common coupling point, and are connected to the onshore power grid via the common coupling point. The flexible DC grid-connected channel uses a modular multilevel converter, and the low-frequency AC grid-connected channel uses a modular multilevel matrix converter. A simulation model of the COSCO offshore wind power hybrid grid-connected system was established, and the following faults were set: open circuit fault of the bridge arm of the modular multilevel matrix converter, single-phase ground fault and three-phase ground fault on the receiving end side. The modular multilevel matrix converter has nine bridge arms, and the open circuit state of each bridge arm was set in sequence to form the operating conditions corresponding to different fault locations. Under various fault conditions, the three-phase voltage and three-phase current signals on the receiving end side are collected and the corresponding fault categories are recorded. The three-phase current signal is selected as the fault diagnosis input. The three-phase current signal under a single fault condition is combined with the corresponding fault category to form a fault sample. A fault sample set is constructed from multiple fault samples. The bridge arm open circuit fault category is used to characterize the location of the faulty bridge arm, and the ground short circuit fault category is used to characterize a single-phase ground short circuit or a three-phase ground short circuit. For each phase current sampling sequence, time-domain features, frequency-domain features, and waveform features are extracted. Time-domain features include mean, variance, maximum value, minimum value, RMS value, kurtosis, and skewness. Frequency-domain features are obtained through Fast Fourier Transform and include fundamental amplitude, total harmonic distortion rate, and harmonic amplitude standard deviation. Waveform features include waveform change rate and zero-crossing rate. The waveform change rate is calculated based on the changes in adjacent sampling points, and the zero-crossing rate is calculated based on the number of sign changes between adjacent points in the sampling sequence. The time-domain features, frequency-domain features, and waveform features corresponding to the current of each phase are combined in a fixed order to form twelve-dimensional phase fault features; then, three sets of phase fault features are spliced ​​together in the order of phase A, phase B, and phase C to form thirty-six-dimensional comprehensive fault features; each comprehensive fault feature and its corresponding fault category together constitute a classification sample. The fault sample set is divided into a training set and a test set. The comprehensive fault features in the training set are used as input and the corresponding fault category is used as output to train a random forest classification model. During training, samples are randomly drawn from the training set to form a sub-training set for each decision tree. Some features are randomly selected from the comprehensive fault features as candidate splitting features. The node splitting features are determined according to the information gain, and multiple decision trees are recursively generated. Each decision tree outputs the classification result, and the random forest determines the final fault category based on majority voting. The diagnostic results of the random forest classification model were tested using a test set. The model output categories were compared with the actual fault categories to verify the model's ability to identify bridge arm open circuit faults, single-phase ground faults, and three-phase ground faults. In actual diagnosis, the three-phase current signals to be diagnosed are collected at the receiving end. The time-domain features, frequency-domain features, and waveform features of each phase are calculated according to the same feature extraction order as in the training phase. The signals are then spliced ​​together according to the three-phase phase sequence to obtain the comprehensive fault features to be diagnosed. The comprehensive fault features to be diagnosed are input into the trained random forest classification model. After classification by multiple decision trees and majority voting, the fault diagnosis result is output. When the output result is an open circuit fault in the bridge arm, the location of the corresponding faulty bridge arm is determined at the same time.

[0024] The technical solution of this invention collects three-phase electrical quantity signals under different fault states in the COSCO-COSCO-Ocean Wind Power Hybrid Grid-connected System, and sequentially completes fault sample construction, multi-domain feature extraction, three-phase feature fusion, and fault classification model training. The trained fault classification model is then used to identify the fault in the signal to be diagnosed, effectively solving the problem that the electrical quantity characteristics of different types of faults in the COSCO-Ocean Wind Power Hybrid Grid-connected System are mutually coupled and difficult to accurately and uniformly identify.

[0025] Furthermore, the COSCO offshore wind power hybrid grid-connected system includes an offshore wind farm power collection system, a flexible DC grid-connected channel, a low-frequency AC grid-connected channel, a power frequency AC grid-connected channel, a common coupling point, and an onshore power grid. The flexible DC grid-connected channel, the low-frequency AC grid-connected channel, and the power frequency AC grid-connected channel are connected in parallel between the offshore wind farm power collection system and the common coupling point, which is connected to the onshore power grid.

[0026] As a preferred embodiment of the above, the COSCO offshore wind power hybrid grid-connected system includes an offshore wind farm power collection system, a flexible DC grid-connected channel, a low-frequency AC grid-connected channel, a power frequency AC grid-connected channel, a common coupling point, and an onshore power grid; The offshore wind farm power collection system is used to collect the electrical energy output from multiple offshore wind turbines and connects to the sending end of the flexible DC grid connection channel, low-frequency AC grid connection channel and power frequency AC grid connection channel respectively; the three types of grid connection channels run in parallel and transmit the electrical energy output from the offshore wind farm power collection system to the receiving end respectively. The receiving ends of flexible DC grid-connected channels, low-frequency AC grid-connected channels, and power frequency AC grid-connected channels are connected to the same receiving end bus, and a common coupling point is formed at the receiving end bus. The common coupling point is connected to the onshore power grid, so that the electrical energy transmitted by the three types of grid-connected channels is collected and sent to the onshore power grid. During normal operation, the three types of grid-connected channels jointly undertake the transmission of offshore wind power and achieve power aggregation at the receiving end through a common coupling point. When a fault occurs within the grid-connected channel or on the onshore power grid side, the voltage and current changes generated by each fault can be transmitted to the common coupling point through the corresponding grid-connected channel, providing a system basis for subsequent acquisition of three-phase electrical signals and fault diagnosis.

[0027] Furthermore, the flexible DC grid-connected channel is equipped with a modular multilevel converter, and the low-frequency AC grid-connected channel is equipped with a modular multilevel matrix converter.

[0028] As a preferred embodiment of the above, the flexible DC grid-connected channel is equipped with a modular multilevel converter, and the low-frequency AC grid-connected channel is equipped with a modular multilevel matrix converter. The modular multilevel converter includes three phase units, each phase unit includes an upper bridge arm and a lower bridge arm, forming a total of six bridge arms; each bridge arm consists of multiple sub-modules connected in series and bridge arm inductors. By controlling the activation state of each sub-module, a multilevel AC voltage is formed to realize the conversion between AC power and DC power, and to complete the power transmission of the flexible DC grid-connected channel. The modular multilevel matrix converter is used to connect two three-phase AC systems with different frequencies. The input side includes three-phase AC terminals a, b, and c, and the output side includes three-phase AC terminals r, s, and t. Each input AC terminal is connected to the three output AC terminals through a bridge arm, forming nine bridge arms. Each bridge arm consists of multiple power units with the same structure connected in series with the bridge arm inductor. By controlling the switching state of each power unit, the output voltage of the bridge arm is adjusted to realize the direct conversion between power frequency AC energy and low frequency AC energy. The power unit can adopt a full-bridge sub-module, which includes an H-bridge switching circuit and a DC support capacitor. During operation, the output state of each full-bridge sub-module is controlled by pulse width modulation, so that each bridge arm forms a multi-level stepped wave that approximates the reference voltage, and realizes the voltage and power regulation of the low-frequency AC grid-connected channel.

[0029] Furthermore, the different fault states include bridge arm open circuit faults in modular multilevel matrix converters, single-phase ground faults and three-phase ground faults in COSCO offshore wind power hybrid grid-connected systems.

[0030] As a preferred embodiment of the above, modular multilevel matrix converter bridge arm open circuit fault, single-phase ground short circuit fault and three-phase ground short circuit fault models are established in the COSCO offshore wind power hybrid grid-connected system, and the system is run under each fault state to obtain the corresponding three-phase electrical quantity signals. Nine bridge arms are formed between the three-phase input terminals a, b, c and the three-phase output terminals r, s, t of the modular multilevel matrix converter. Bridge arm open-circuit faults are set according to the single-bridge-arm fault mode. Each time, a target bridge arm is selected, causing the power devices in the target bridge arm to lose their normal switching capability, resulting in the bridge arm output voltage deviating from the control command and forming an open-circuit state. The remaining bridge arms continue to operate normally. Open-circuit faults are set for the nine bridge arms in sequence to form nine types of bridge arm open-circuit fault samples, and the location of the bridge arm where the open circuit occurs is used as the fault category label. A single-phase ground fault is set in an AC transmission line to create a short circuit connection between any phase conductor and ground. After the fault occurs, the voltage of the faulted phase drops, the system generates negative sequence and zero sequence components, and the three-phase current exhibits amplitude imbalance, phase shift, and waveform distortion. The three-phase electrical quantity signals during the fault are collected at the receiving end and marked as a single-phase ground fault. A three-phase ground fault is set at the receiving end, so that the three-phase conductors are simultaneously short-circuited to the ground; after the fault occurs, the three-phase voltages at the receiving end drop simultaneously, and the three-phase currents increase synchronously and remain relatively symmetrical; the three-phase electrical quantity signals during the fault are collected at the receiving end and marked as a three-phase ground fault. Nine types of bridge arm open circuit faults, single-phase ground short circuit faults, and three-phase ground short circuit faults were simulated multiple times. The three-phase electrical quantity signals and fault categories corresponding to each fault state were saved for subsequent construction of fault sample sets.

[0031] To elaborate further, three-phase electrical quantity signals are three-phase current signals.

[0032] As a preferred embodiment of the above, under each fault condition, the currents of phase A, phase B and phase C are simultaneously collected at the current measurement point at the receiving end to obtain a three-phase current sampling sequence; the three-phase currents use the same sampling start point, sampling frequency and sampling duration to ensure that the sampling points of each phase correspond in time. Each fault operation corresponds to a set of three-phase current signals, which are stored in association with the current fault category. When an open-circuit fault occurs in a bridge arm, the three-phase currents exhibit different degrees of distortion, imbalance, and phase shift. When a single-phase ground fault occurs, the three-phase currents exhibit amplitude and phase imbalance. When a three-phase ground fault occurs, the amplitudes of the three-phase currents increase synchronously and remain relatively symmetrical. These differences enable the three-phase currents to serve as diagnostic inputs for different fault states. The validity of the collected results is checked, and missing or obviously abnormal sampling sequences are removed. The three-phase current data are saved in a fixed order of phase A, phase B and phase C for subsequent phase-by-phase multi-domain feature extraction.

[0033] Furthermore, multi-domain feature extraction includes performing time-domain feature extraction, frequency-domain feature extraction, and waveform feature extraction on each phase current signal in the three-phase current signal to obtain the corresponding phase fault features.

[0034] As a preferred embodiment of the above, the current signals of phase A, phase B and phase C are processed using the same sampling window, and the current signals of each phase are independently extracted in the time domain, frequency domain and waveform. For any phase current sampling sequence, the mean, variance, maximum value, minimum value, effective value, kurtosis and skewness are calculated to form a seven-dimensional time-domain feature, which is used to characterize the current amplitude level, dispersion, energy magnitude and probability distribution. Let the sampling sequence of any phase current be... x [ n The number of sampling points is N Discrete frequency index is k The discrete Fourier transform is expressed as: ; The current sampling sequence is decomposed into even and odd sequences, and expressed by the Fast Fourier Transform as follows: ; in, x [ n [The following is a list of phase current signals and their values, which are not translated as they are not part of the main text.] n The sampled values ​​at each sampling time point; X [ k[The current sampling sequence after discrete Fourier transform is at the [number]th [time]. k The complex spectrum values ​​corresponding to each discrete frequency point; N This represents the total number of current sampling points contained within a sampling window. n This is the time-domain sampling point number, with a value ranging from 0 to N-1; k This is the discrete frequency point index, with a value range from 0 to N-1; r This is the sampling point number of the subsequence after parity decomposition of the current sampling sequence, with a value range from 0 to N / 2-1; j The imaginary unit; e It is a natural constant; is the rotation factor of the discrete Fourier transform, where ; This is the rotation factor used in the Fast Fourier Transform to combine the spectra of odd and even sequences; The rotation factor used when performing a discrete Fourier transform on a subsequence of length N / 2; x [2 r ] represents the 2rth sample value in the original current sampling sequence, i.e., the even-numbered sample value; x [2 r+ [1] is the 2r+1th sampled value in the original current sampling sequence, i.e., the odd-numbered sampled value; A [ k [] represents the spectral value of the subsequence composed of even-numbered sampled values ​​at the k-th discrete frequency point; B [ k [] represents the spectral value of the subsequence composed of odd-numbered sampled values ​​at the k-th discrete frequency point; X [ k + N [2] is the k+N / 2th spectral value corresponding to the second half of the original current sampling sequence after fast Fourier transform; the fundamental amplitude, total harmonic distortion rate, and harmonic amplitude standard deviation are extracted from the transformed spectrum to form a three-dimensional frequency domain feature; the fundamental amplitude is taken as the amplitude corresponding to the fundamental frequency point, the total harmonic distortion rate is determined according to the degree of distortion of the harmonic components relative to the fundamental component, and the harmonic amplitude standard deviation is determined according to the degree of dispersion of each harmonic amplitude; For any phase current sampling sequence, the waveform change rate and zero-crossing rate are calculated to form a two-dimensional waveform feature. The waveform change rate is determined based on the amplitude change between adjacent sampling points and is used to characterize the degree of change of the current waveform. The zero-crossing rate is determined based on the number of sign changes between adjacent sampling points and is used to characterize the frequency of the current waveform crossing zero. By combining time-domain features, frequency-domain features, and waveform features in that order, twelve-dimensional phase fault features corresponding to any phase current signal are obtained. The same processing method is used to obtain phase A fault features, phase B fault features, and phase C fault features, providing input for subsequent three-phase feature fusion.

[0035] Furthermore, three-phase characteristic fusion involves splicing together the fault characteristics of each phase according to the phase sequence to obtain a comprehensive fault characteristic.

[0036] As a preferred embodiment of the above, based on the obtained phase A fault characteristics, phase B fault characteristics and phase C fault characteristics; each phase fault characteristic is arranged in the order of mean, variance, maximum value, minimum value, effective value, kurtosis, skewness, fundamental amplitude, total harmonic distortion rate, harmonic amplitude standard deviation, waveform change rate and zero crossing rate to form a twelve-dimensional phase fault characteristic; According to the phase sequence of phases A, B, and C, the fault characteristics of the three phases are spliced ​​together in sequence to form a 36-dimensional comprehensive fault characteristic. The first 12 dimensions of the comprehensive fault characteristic correspond to phase A, the middle 12 dimensions correspond to phase B, and the last 12 dimensions correspond to phase C, thus simultaneously preserving the multi-domain characteristics of the current of each phase and the difference information between the three phases. The training, testing, and actual diagnosis of the fault classification model all use the same feature arrangement order; for each fault sample, the generated comprehensive fault features are associated with and stored with the corresponding fault category; for the three-phase current signal to be diagnosed, the comprehensive fault features to be diagnosed are generated in the same order to ensure that each feature corresponds one-to-one with the input position of the fault classification model.

[0037] Furthermore, such as Figure 3 As shown, the fault classification model is a random forest classification model. The random forest classification model is used to classify bridge arm open circuit faults, single-phase ground short circuit faults and three-phase ground short circuit faults based on comprehensive fault characteristics, and to determine the location of the faulty bridge arm in the classification results corresponding to the bridge arm open circuit fault.

[0038] As a preferred embodiment of the above, the open-circuit faults of the nine arms of the modular multilevel matrix converter are set as independent fault categories, and the single-phase ground short-circuit fault and the three-phase ground short-circuit fault are set as independent fault categories respectively; each training sample consists of comprehensive fault features and corresponding fault categories, so that the category label of the open-circuit fault of the bridge arm simultaneously represents the fault type and the location of the faulty bridge arm. A random forest classification model is used for training; the random forest consists of multiple decision trees constructed based on classification and regression tree algorithms; a subset of samples is randomly selected from the training set to form a sub-training set for the i-th decision tree. D i ( i = 1, 2, 3,…, n), and randomly select some features from the comprehensive fault features as candidate splitting features; Sub-training set D i Information entropy is calculated as follows: ; in, H ( D i (for the sub-training set) D i Information entropy C j For the j-th fault category, P ( C j ) represents the fault category C j In the sub-training set D i The probability of occurrence in; Fault Category C j The probability of its occurrence is calculated as follows: ; Among them, count( C j (for the sub-training set) D i The middle belongs to the fault category C j The number of samples, | D i | For the child training set D i The total number of samples; Candidate Fault Characteristics A The corresponding information gain is calculated as follows: ; Among them, Gain ( D i A) Using candidate fault features A Divide the training set D i The information gain obtained H ( D i |A) Based on candidate fault characteristics A Conditional information entropy after partitioning; Conditional information entropy is calculated as follows: ; in, X Candidate fault characteristics A The value of , P ( X ) represents the value X The probability of its occurrence, P (C j | X The candidate fault feature takes the value of ) X The fault category is C j The conditional probability; At each node, calculate the information gain of each candidate fault feature, select the candidate fault feature with the largest information gain as the splitting feature, and recursively split the nodes of the resulting subsets until the stopping condition is met, thus completing the construction of the decision tree; repeat the above process to form multiple mutually different decision trees; During diagnosis, the comprehensive fault features to be diagnosed are input into each decision tree, and each decision tree outputs a fault category. The random forest counts the frequency of occurrence of each fault category and uses the fault category with the highest frequency as the final diagnosis result. The voting process is as follows: ; in, T ( i Let be the fault category output by the i-th decision tree. n For the number of decision trees, C j To obtain the fault category with the highest voting frequency; When the final diagnosis result is a single-phase ground fault or a three-phase ground fault, the corresponding ground fault type is output; when the final diagnosis result is an open-circuit fault of the bridge arm, the location of the bridge arm where the open-circuit fault occurred is output according to the correspondence between the fault category label and the bridge arm location.

[0039] Example 2; Based on the same inventive concept as the method for accurate fault diagnosis of hybrid grid-connected offshore wind power in the foregoing embodiments, the present invention also provides a system for accurate fault diagnosis of hybrid grid-connected offshore wind power, the system comprising: The fault sample construction module acquires the three-phase electrical quantity signals of the COSCO offshore wind power hybrid grid-connected system under different fault states, and constructs a fault sample set based on the three-phase electrical quantity signals and the corresponding fault categories. The multi-domain feature fusion module performs multi-domain feature extraction and three-phase feature fusion on the three-phase electrical quantity signals in the fault sample set to obtain comprehensive fault features; The classification model training module trains a fault classification model based on comprehensive fault features and fault categories. The fault diagnosis output module acquires the three-phase electrical quantity signals to be diagnosed, obtains the comprehensive fault characteristics to be diagnosed by multi-domain feature extraction and three-phase feature fusion, and outputs the fault diagnosis results using the fault classification model.

[0040] The adjustment system described above in this invention can effectively realize a method for accurate diagnosis of faults in hybrid grid connection of offshore wind power, and the technical effects it can achieve are as described in the above embodiments, and will not be repeated here.

[0041] Example 3; Based on the same inventive concept as the method for accurate fault diagnosis of hybrid offshore wind power grid connection in the foregoing embodiments, the present invention also provides a computer-readable storage medium storing a computer program, the computer program including program instructions, which, when executed by a processor, can realize the method for accurate fault diagnosis of hybrid offshore wind power grid connection.

[0042] Example 4; A simulation model of a modular multilevel matrix converter was built in the MATLAB / Simulink environment, and the correctness and accuracy of the fault diagnosis method introduced above were effectively verified. Based on the built simulation model, the corresponding results were analyzed. The COSCO-COSCO offshore wind power hybrid grid-connected system includes three grid-connected channels, namely a flexible DC channel, a power frequency AC channel, and a low frequency AC channel. Based on the simulation model of the COSCO-COSCO offshore wind power hybrid grid-connected system, simulation verification was performed on open circuit faults, single-phase ground faults, and three-phase ground faults of the modular multilevel matrix converter, and voltage and current signal data under the corresponding operating conditions were collected to obtain the characteristic data corresponding to different fault types. When open-circuit faults occur in different bridge arms, the three-phase currents will exhibit varying degrees of distortion and imbalance. Due to the strong coupling relationship between the bridge arms, an open circuit in a single bridge arm not only affects the current transmission capacity of the corresponding phase of the faulty bridge arm but also causes changes in the operating state of the remaining bridge arms, resulting in the overall three-phase current deviating from the normal operating state. As the system re-establishes the power transmission path, the current waveform gradually enters a new steady-state operating state. The degree of current distortion, amplitude changes, and phase shifts corresponding to open-circuit faults in different bridge arms are significantly different, thus providing effective characteristic information for subsequent fault location and identification. Both single-phase and three-phase ground faults cause the system current to deviate from normal operating conditions, but their manifestations differ significantly. Single-phase ground faults are asymmetrical faults, generating negative-sequence and zero-sequence components after the fault occurs, leading to a significant imbalance in the amplitude and phase of the three-phase currents, and distortion in some phase currents. In contrast, three-phase ground faults are symmetrical faults, with the amplitude of the three-phase currents at the receiving end increasing rapidly after the fault occurs and remaining at a high level during the fault's duration, while maintaining good symmetry among the three phases. Once the fault is cleared, the current gradually returns to normal operating conditions. Therefore, single-phase ground faults are mainly characterized by imbalanced three-phase current amplitudes and waveform distortion, while three-phase ground faults are characterized by simultaneous large abrupt changes in the three-phase currents during the fault period while maintaining good symmetry. These two types of faults exhibit significantly different time-domain response characteristics, providing a reliable data foundation for subsequent random forest models to achieve fault identification and classification. (1) Fast Fourier Transform Results Fast Fourier Transform was performed on the fault samples collected from the COSCO offshore wind power hybrid grid-connected system to map the time domain data to the frequency domain for feature analysis; Figure 4 The fundamental amplitude, total harmonic distortion (THD), harmonic amplitude standard deviation, and three-dimensional frequency domain characteristic distribution of each sample are displayed. It can be seen that different fault types exhibit significant differences in the frequency domain. The fundamental amplitude, harmonic distribution, and THD of the samples of open-circuit faults, single-phase ground faults, and three-phase ground faults in modular multilevel matrix converters have different characteristics. These frequency domain differences can effectively reflect the impact of various faults on the system current, providing a reliable basis for the subsequent construction of comprehensive feature vectors and random forest classification models. (2) Fault diagnosis results of random forest open path In the fault diagnosis method for hybrid grid-connected offshore wind power systems based on multi-feature fusion and random forest proposed in this invention, a multi-dimensional feature vector containing time-domain features, frequency-domain features, and waveform features is constructed. Simultaneously, a random forest classification model is combined to achieve accurate identification of different faults in the hybrid grid-connected offshore wind power system. Experimental results are as follows: Figure 5 As shown, after PCA dimensionality reduction of the extracted high-dimensional features, different fault types exhibit obvious clustering distribution characteristics in the feature space; samples of the same type of fault are densely distributed in the principal component space, while different categories have good separation, indicating that the constructed feature system can effectively identify the differences between different fault types, providing a good basis for the classification model.

[0043] Although this application has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made thereto without departing from the spirit and scope of this application. Accordingly, this specification and drawings are merely exemplary illustrations of the application as defined herein, and are to be considered as covering any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Thus, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A method for accurate fault diagnosis of hybrid grid-connected wind power in medium and offshore areas, characterized in that, The method includes: Acquire the three-phase electrical quantity signals of the COSCO offshore wind power hybrid grid-connected system under different fault states, and construct a fault sample set based on the three-phase electrical quantity signals and the corresponding fault categories; Multi-domain feature extraction and three-phase feature fusion are performed on the three-phase electrical quantity signals in the fault sample set to obtain comprehensive fault features; A fault classification model is trained based on the comprehensive fault characteristics and the fault categories. The three-phase electrical quantity signals to be diagnosed are acquired, and the comprehensive fault features to be diagnosed are obtained according to the multi-domain feature extraction and the three-phase feature fusion method. The fault diagnosis results are then output using the fault classification model.

2. The method for accurate fault diagnosis of hybrid grid-connected offshore wind power as described in claim 1, characterized in that, The offshore wind power hybrid grid-connected system includes an offshore wind farm power collection system, a flexible DC grid-connected channel, a low-frequency AC grid-connected channel, a power frequency AC grid-connected channel, a common coupling point, and an onshore power grid. The flexible DC grid-connected channel, the low-frequency AC grid-connected channel, and the power frequency AC grid-connected channel are connected in parallel between the offshore wind farm power collection system and the common coupling point, and the common coupling point is connected to the onshore power grid.

3. The method for accurate fault diagnosis of hybrid grid-connected offshore wind power according to claim 2, characterized in that, The flexible DC grid-connected channel is equipped with a modular multilevel converter, and the low-frequency AC grid-connected channel is equipped with a modular multilevel matrix converter.

4. The method for accurate fault diagnosis of hybrid grid-connected offshore wind power as described in claim 3, characterized in that, The different fault states include the bridge arm open circuit fault of the modular multilevel matrix converter, the single-phase grounding short circuit fault and the three-phase grounding short circuit fault of the COSCO offshore wind power hybrid grid-connected system.

5. The method for accurate fault diagnosis of hybrid grid-connected offshore wind power as described in claim 1, characterized in that, The three-phase electrical quantity signal is a three-phase current signal.

6. The method for accurate fault diagnosis of hybrid grid-connected offshore wind power as described in claim 5, characterized in that, The multi-domain feature extraction includes performing time-domain feature extraction, frequency-domain feature extraction, and waveform feature extraction on each phase current signal in the three-phase current signals to obtain the corresponding phase fault features.

7. The method for accurate fault diagnosis of hybrid grid-connected offshore wind power as described in claim 6, characterized in that, The three-phase feature fusion includes splicing the fault features of each phase according to the phase sequence to obtain the comprehensive fault features.

8. The method for accurate fault diagnosis of hybrid grid-connected offshore wind power according to claim 4, characterized in that, The fault classification model is a random forest classification model. The random forest classification model is used to classify the bridge arm open circuit fault, the single-phase ground short circuit fault and the three-phase ground short circuit fault according to the comprehensive fault characteristics, and to determine the location of the faulty bridge arm in the classification results corresponding to the bridge arm open circuit fault.

9. A precise fault diagnosis system for hybrid grid-connected wind power in offshore and medium-sized cities, characterized in that, The system includes: The fault sample construction module acquires the three-phase electrical quantity signals of the COSCO offshore wind power hybrid grid-connected system under different fault states, and constructs a fault sample set based on the three-phase electrical quantity signals and the corresponding fault categories. The multi-domain feature fusion module performs multi-domain feature extraction and three-phase feature fusion on the three-phase electrical quantity signals in the fault sample set to obtain comprehensive fault features; The classification model training module trains a fault classification model based on comprehensive fault features and fault categories. The fault diagnosis output module acquires the three-phase electrical quantity signals to be diagnosed, obtains the comprehensive fault characteristics to be diagnosed by multi-domain feature extraction and three-phase feature fusion, and outputs the fault diagnosis results using the fault classification model.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which includes program instructions that, when executed by a processor, can implement the method for accurate fault diagnosis of hybrid grid-connected offshore wind power as described in any one of claims 1-8.

Images