Direct current charging pile fault diagnosis and repair method and device and storage medium

Abstract

The invention relates to a direct current charging pile fault diagnosis and repair method and device and a storage medium. The method comprises the following steps: carrying out Hilbert transformation on voltage, current and temperature parameters of a charging pile to obtain envelope data of a multi-channel analysis signal; performing frequency domain conversion and maximum posterior probability matrix decomposition on the envelope data to obtain a frequency spectrum feature vector; inputting the spectrum feature vector into a dual-control law observer with a continuous item and a discontinuous item for state tracking to obtain a system residual matrix; performing threshold detection and accumulative error integral operation on the system residual matrix to obtain fault feature data; and performing lambda consensus iterative operation on the fault feature data to obtain a control parameter compensation scheme. According to the method, the recognition accuracy of fault features is improved, and the optimality and reliability of a repair strategy are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of direct current charging piles, and in particular to a direct current charging pile fault diagnosis and repair method, device and storage medium. Background Art

[0002] With the rapid development of the new energy vehicle industry, the operational reliability of DC charging piles as key infrastructure has received increasing attention. Charging piles are prone to various faults during long-term operation, which not only affect charging efficiency but may also lead to safety accidents. Therefore, it is necessary to establish an efficient and accurate fault diagnosis and repair mechanism. At present, the mainstream charging pile fault diagnosis method is mainly based on the spectrum analysis of operating parameters, but due to the limited sampling time, the traditional method has obvious deficiencies in diagnostic accuracy and real-time performance.

[0003] Existing charging pile fault diagnosis systems generally adopt a single controller structure, which makes it difficult to take into account both diagnostic accuracy and response speed. Traditional multi-channel data acquisition solutions face problems such as low resolution and signal interference, which affect the quality of fault feature extraction. In terms of fault repair, existing repair strategies often use fixed control parameters and lack the ability to adaptively adjust the dynamic characteristics of the system, making it difficult to meet the real-time repair needs of large-scale charging stations. Summary of the invention

[0004] The main purpose of the present invention is to provide a DC charging pile fault diagnosis and repair method, device and storage medium, which improves the recognition accuracy of fault characteristics and ensures the optimality and reliability of the repair strategy.

[0005] To achieve the above object, the present invention provides a DC charging pile fault diagnosis and repair method, comprising the following steps: Perform Hilbert transform on the voltage, current and temperature parameters of the charging pile to obtain the envelope data of the multi-channel analytical signal; Performing frequency domain conversion and maximum a posteriori probability matrix decomposition on the envelope data to obtain a frequency spectrum feature vector; Inputting the frequency spectrum feature vector into a dual control law observer having continuous terms and discontinuous terms for state tracking to obtain a system residual matrix; Performing threshold detection and cumulative error integration operations on the system residual matrix to obtain fault feature data; A lambda consensus iterative operation is performed on the fault characteristic data to obtain a control parameter compensation scheme.

[0006] The present invention also provides a DC charging pile fault diagnosis and repair device, comprising: A transformation module is used to perform Hilbert transformation on the voltage, current and temperature parameters of the charging pile to obtain envelope data of a multi-channel analytical signal; A decomposition module, used for performing frequency domain conversion and maximum a posteriori probability matrix decomposition on the envelope data to obtain a frequency spectrum feature vector; A state tracking module, used for inputting the spectrum feature vector into a dual control law observer having continuous terms and discontinuous terms for state tracking to obtain a system residual matrix; An integral operation module, used for performing threshold detection and cumulative error integral operation on the system residual matrix to obtain fault feature data; The iteration module is used to perform a lambda consensus iteration operation on the fault characteristic data to obtain a control parameter compensation scheme.

[0007] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0008] The present invention also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of any of the above-mentioned methods are implemented.

[0009] In summary, the technical solution provided by the present invention effectively improves the accuracy of fault feature extraction and reduces the interference of background noise by introducing a multi-channel data preprocessing mechanism based on Hilbert transform and combining the frequency domain analysis of 0.1Hz high-resolution frequency grid. The dual control law observer structure is adopted, and the unification of fast convergence and precise tracking is achieved through the adaptive weight allocation of discontinuous control law and continuous control law, thereby enhancing the robustness of fault diagnosis. The feature extraction method based on sparse envelope spectrum analysis improves the recognition accuracy of fault features through matrix sparsification processing with L1 norm constraints. The lambda consensus algorithm is innovatively applied to the optimization of repair strategies, and the computational efficiency of parameter optimization is significantly improved through Lagrangian dual decomposition and parallel computing architecture. A double-layer detector structure is designed, combined with real-time threshold detection and cumulative error integral analysis, to achieve comprehensive monitoring of instantaneous faults and gradual faults. By establishing a multidimensional optimization model of the repair objective function and combining the decoupling and separation of control parameters and compensator parameters, the optimality and reliability of the repair strategy are guaranteed. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 This is a schematic diagram of the steps of a DC charging pile fault diagnosis and repair method in one embodiment of the present invention; Figure 2 This is a structural block diagram of a DC charging pile fault diagnosis and repair device in one embodiment of the present invention; Figure 3 It is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0011] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0012] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0013] Reference Figure 1 , this embodiment provides a DC charging pile fault diagnosis and repair method, including the following steps: S1, perform Hilbert transform on the voltage, current and temperature parameters of the charging pile to obtain the envelope data of the multi-channel analytical signal; Among them, the voltage, current and temperature parameters are input into a high-precision analog-to-digital converter for digital conversion, and the analog signal is converted into digital parameter data. The Hilbert kernel function based on the Fourier series is used to perform convolution calculation on the digital parameter data. The Hilbert kernel function generates the orthogonal components of the signal through the symmetry of the frequency domain, so that the real and imaginary parts of the signal can maintain a 90-degree difference in phase. The Fourier series provides a mathematical framework that represents the signal as a linear combination of a series of sine and cosine functions, and the Hilbert kernel function uses this representation to extract the orthogonal component data of the signal through convolution operations. These orthogonal components contain the instantaneous characteristics of the input signal. The orthogonal component data and the original digital parameter data are used for complex synthesis to generate analytical signal data. The orthogonal component is used as the imaginary part, and the digital parameter data is used as the real part, and the two together constitute an analytical signal in complex form. In order to extract the envelope amplitude data, the data of the complex analytical signal is subjected to the square sum operation of the real and imaginary parts, and then the square root of the sum of squares is calculated. The mathematical expression of this operation is ,in represents the envelope amplitude, and Represent the real part and imaginary part of the signal respectively. The envelope amplitude data generated in this step is an intuitive reflection of the energy change of the signal. Perform time domain alignment on the envelope amplitude data. Through a method based on linear interpolation, data with different sampling frequencies or time points are mapped to a unified time axis to solve the problem of timing inconsistency. Linear interpolation uses the linear relationship between adjacent data points to interpolate unaligned signals to obtain envelope data with time synchronization, reflecting the coordination and consistency of signals in different channels. The synchronized envelope data is grouped and merged according to the voltage channel, current channel and temperature channel respectively. By arranging different types of signals, the organization of the data is clearer, which is convenient for subsequent multi-channel signal analysis. The grouped and merged data structure effectively supports the subsequent fault feature extraction and diagnosis algorithm, and finally forms multi-channel analytical signal envelope data containing all channel information.

[0014] S2, frequency domain conversion and maximum a posteriori probability matrix decomposition are performed on the envelope data to obtain the spectrum feature vector; Specifically, the envelope data is subjected to discrete Fourier transform. By setting a frequency resolution grid of 0.1 Hz, the envelope data in the time domain is mapped to the frequency domain to generate initial spectrum data. By reasonably designing the sampling frequency and frequency resolution, the low-frequency fault characteristics that may be contained in the signal can be captured while avoiding the interference of high-frequency noise. The discrete Fourier transform can transform the signal from the time domain to the frequency domain and generate a complex spectrum containing amplitude and phase information. The initial spectrum data is subjected to matrix sparse processing using the L1 norm constraint method. The L1 norm constraint effectively compresses the low-intensity spectrum components by introducing sparse regularization while retaining significant frequency components with important information. The goal of sparseness is to generate a sparse spectrum matrix so that only a few frequency points with significant amplitudes are retained in the matrix, while the values ​​of the rest are close to zero. The sparse spectrum matrix is ​​input into the Bayesian prior probability model for parameter optimization and statistical modeling of spectrum characteristics. The introduction of the Bayesian prior probability model is to combine prior knowledge and data distribution characteristics to improve the descriptive ability of spectrum data through statistical inference. Estimate the parameters of the prior distribution of the sparse spectrum matrix, for example, assume that its amplitude follows a Gaussian distribution or a Laplace distribution, and use these assumptions to establish a spectrum prior distribution model. The spectrum prior distribution is iteratively calculated using the maximum a posteriori probability criterion to obtain the posterior probability density function. In this process, the maximum a posteriori probability criterion captures the statistical characteristics of the signal and eliminates noise and interference by continuously optimizing the model parameters. Based on the posterior probability density function, the maximum likelihood method is used to solve the parameters to generate the consistency data of the multi-channel spectrum. The core idea of ​​maximum likelihood estimation is to find a set of parameter values ​​that maximizes the likelihood value of the given data under the model. This result ensures that the generated multi-channel spectrum consistency data can reflect the common characteristics between different signal channels while maintaining sensitivity to the unique information of specific channels. Perform real-value transformation and dimensionality reduction operations on the consistency data. The real-valued transformation converts the spectral data into real form by removing the complex part in the spectral data, which is convenient for subsequent analysis and processing; the dimensionality reduction operation converts high-dimensional data into low-dimensional feature vectors through principal component analysis or other dimensionality reduction algorithms, extracts the most representative spectral feature information, and obtains the spectral feature vector.

[0015] S3, input the spectrum feature vector into the dual control law observer with continuous terms and discontinuous terms for state tracking, and obtain the system residual matrix; It should be noted that the spectral feature vector is input into the state space model, and the dynamic characteristics of the charging pile system are analyzed to obtain the system state equation. The state space model reflects the time-varying characteristics of the charging pile during operation by describing the dynamic relationship between the system state variables. By analyzing the spectral feature vector, the dynamic relationship between the input and output of the system is expressed in the form of a state equation. After establishing the system state equation, a sliding surface based on a sign function is designed to generate a discontinuous control law. Sliding mode control is a nonlinear control method with strong robustness. It guides the system state to the sliding surface through the design of the sliding surface, and moves along the sliding surface to achieve the control target. The design of the discontinuous control law uses the sign function to feedback and adjust the state deviation of the system, so that the tracking accuracy and stability of the system can still be guaranteed when there is strong disturbance or uncertainty. However, due to the inherent discontinuity of the sign function, there is a chattering phenomenon in the control input, which is improved by introducing a continuous control law. In order to achieve the continuity and robustness of the system control, the system state equation is processed continuously based on the hyperbolic tangent function to generate a continuous control law. The hyperbolic tangent function can effectively alleviate the chattering problem caused by the discontinuous control law due to its smooth transition characteristics, while maintaining sensitivity to system disturbances. The design of the continuous control law replaces the sign function with the hyperbolic tangent function, so that the control input is gradually adjusted with the change of the state error, thereby achieving stable control of the system within a certain range. According to the error size of the system state equation, the discontinuous control law and the continuous control law are adaptively weighted to generate a combined control law. The adaptive weight allocation dynamically adjusts the action ratio of the discontinuous control law and the continuous control law according to the amplitude and frequency changes of the system state error. When the error is large, the discontinuous control law can respond quickly and eliminate large deviations; when the error decreases, the continuous control law gradually takes the dominant position to achieve a smooth transition of the control input. The adaptive characteristics of the combined control law can take into account the rapidity and stability of the system control. The combined control law is input into the repetitive controller to compensate for periodic disturbances. The repetitive controller is a control strategy that can effectively suppress periodic interference and achieve precise control of system operation by memorizing and compensating for repetitive errors. In the compensation process, the generated compensation control quantity is used to correct the deviation of the input signal so that the system can still maintain stable operation under complex disturbance environment. In order to ensure the effectiveness of the compensation control quantity and the stability of the system, a convergence analysis based on Lyapunov stability theory is performed. Lyapunov stability theory verifies whether the system state can tend to be stable over time by constructing Lyapunov function, thereby ensuring that the designed compensation control strategy has theoretical stability guarantee. After completing the state observation, the residual calculation is performed based on the observation results and the input spectrum feature vector to generate an error sequence. The error sequence is the deviation between the actual operating state of the system and the theoretical model state, which is used to reflect the dynamic change characteristics of the system under different operating conditions.In order to enhance the feature expression of the residual, the error sequence is reconstructed by sliding. Sliding reconstruction introduces a sliding window to perform local weighted averaging or filtering on the error sequence to eliminate high-frequency noise interference while retaining significant feature information in system operation. The result of sliding reconstruction is organized into a system residual matrix.

[0016] S4, performing threshold detection and cumulative error integration operations on the system residual matrix to obtain fault feature data; Specifically, the real-time data segments of the system residual matrix are input into the threshold logic detector. The threshold logic detector performs instant analysis on the real-time residual data through boundary comparison operations to determine whether it exceeds the preset safe operating range, and generates instantaneous fault detection results accordingly. The main goal of instantaneous fault detection is to quickly identify abnormal conditions that occur and ensure that the system can capture potential fault signals in the early stages. According to the instantaneous fault detection results, a probability density function based on normal distribution is established to quantify the probability characteristics of fault occurrence. Through statistical analysis of residual data, its distribution parameters, such as mean and variance, are estimated to construct a normal distribution model. The probability density function reflects the possibility of fault occurrence in the current operating state of the system. On this basis, the threshold range corresponding to different fault levels is clarified by performing interval segmentation processing on the fault probability distribution data. In order to enhance the perception of the fault evolution trend, the system residual matrix is ​​accumulated to generate a sequence containing historical data. By accumulating historical data, the operating status of the system can be observed from a wider time scale and potential long-term problems can be identified more accurately. The historical data sequence is subjected to exponentially weighted cumulative integral operations to calculate the fault accumulation eigenvalue. Exponential weighted integration gives different time weights to historical data, which highlights the impact of recent data and avoids excessive interference of long-term data on the current state. After obtaining the cumulative characteristic value of the fault, a multi-dimensional feature fusion operation is performed according to the relationship between the fault level threshold and the cumulative characteristic value of the fault to generate a fault type mark. Multi-dimensional feature fusion achieves a comprehensive evaluation and classification of system faults by comprehensively considering fault information in different dimensions, such as the size of the cumulative error, the probability of failure, and the running time. The fault type mark is subjected to feature extraction operation through sparse envelope spectrum analysis to obtain the frequency component of the fault. Sparse envelope spectrum analysis is a signal analysis method based on sparse representation, which extracts key features from complex spectrum data and avoids unnecessary spectrum redundancy. The analysis process accurately identifies the frequency components related to the fault, such as the specific frequency of mechanical vibration or the characteristic frequency of current fluctuation. The matching degree is calculated based on the extracted fault frequency component and the preset fault feature pattern to determine the fault feature data. The preset fault feature pattern is a template library based on historical data and experience summary, which is used to describe the typical characteristics of various known faults. The matching degree calculation determines whether the current fault matches a known fault mode by comparing the similarity between the current frequency component and the template library, and generates the final fault feature data based on this, including the type and severity of the fault, and provides preliminary analysis results of the cause of the fault.

[0017] S5, perform λ consensus iteration operation on the fault feature data to obtain a control parameter compensation scheme.

[0018] Among them, the fault feature data is input into the repair target optimization model to construct a cost function for describing the fault repair target of the charging pile. The repair target optimization model needs to comprehensively consider multiple factors, such as operating efficiency, power loss, maintenance cost and system stability, to construct a multidimensional repair objective function. The control parameters and compensator parameters of the multidimensional repair objective function are decoupled and separated to clarify the variable space involved in the optimization process. The purpose of decoupling and separation is to separate the control parameters (such as the control weights of voltage and current) from the compensator parameters (such as disturbance correction values ​​and feedback gains) to avoid mutual interference between variables during the optimization process. The Lagrange duality principle is used to decompose and optimize the optimization variable space to generate a set of sub-problems. The Lagrange duality principle transforms the constraints of the original problem into a dual problem by introducing Lagrange multipliers, thereby achieving relaxation and decomposition of the problem. This step can effectively decompose the global optimization problem into multiple smaller sub-problem sets, each of which only focuses on some variables and constraint relationships. For the generated sub-problem set, λ consensus parallel computing is performed to obtain a distributed solution set. By introducing the lambda consensus mechanism, consistency constraints are established between sub-problems, so that the solution results of each sub-problem tend to the global optimal solution. The lambda consensus calculation performs multiple rounds of iterations between each sub-problem by sharing key parameters and update information, achieving coordinated consistency of solutions and obtaining a distributed solution set. The distributed solution set is iteratively processed by gradient descent based on adaptive step size to obtain an iterative convergence result. The adaptive step size mechanism dynamically adjusts the update amplitude according to the current gradient size to avoid the oscillation and excessive jump problems in the traditional gradient descent method. In each iteration, by reducing the error term, the optimal solution is gradually approached and global convergence is finally achieved. After multiple adjustments to the iterative convergence results, stability analysis is performed to ensure that the results have steady-state characteristics in practical applications. After obtaining the steady-state solution, the control system of the charging pile is optimized according to the solution. According to the repair solution provided by the steady-state solution, the key parameters in the charging pile control system, such as controller gain, current response time and voltage set value, are readjusted to improve system performance. After optimization, the control system can more efficiently cope with the impact of faults, while improving the stability and energy efficiency of operation, and obtaining the final control parameter compensation solution.

[0019] The optimization variable space is decomposed into a convex optimization of the repaired objective function to generate normalized constraints. By analyzing and separating the objective function and the constraints, the non-convex problem is converted into multiple solvable convex problems. Through this step, the convex characteristics of each part in the variable space are clarified, providing mathematical feasibility and decoupling for subsequent optimization. At the same time, the generated normalized constraints can reasonably limit the scope of the solution to ensure the stability and convergence of the optimization process. The normalized constraints are input into the Lagrange multiplier dual transformer to construct the dual problem and generate the dual constraint equation. The Lagrange dual theory decomposes the original optimization problem into multiple more tractable sub-problems by introducing Lagrange multipliers, and converts the constraints into the form of dual variables. The core of the transformation is to use the properties of the Lagrange function so that the solution of the problem is obtained by jointly solving the original problem and the dual problem. In the dual constraint equation, the relationship between the variables is systematically expressed as a set of equations. According to the constructed dual constraint equation, the optimization problem is decomposed based on the KKT condition and complementary slackness to generate a series of sub-problem sequences. KKT conditions are necessary conditions for optimization problems. Combined with complementary slackness analysis, complex global problems are decomposed into multiple local sub-problems. These sub-problems are relatively independent and can be solved in parallel. The generated sub-problem sequence significantly reduces the complexity of the global optimization problem by clarifying the variable range and constraints of each sub-problem. The sub-problem sequence is deployed in a parallel computing architecture to form a parallel optimization subsystem. The introduction of a parallel computing architecture can significantly improve computing efficiency, while supporting distributed processing of large-scale problems, making the optimization process more efficient and scalable. In the parallel optimization subsystem, in order to achieve collaborative solutions between sub-problems, local variables are iteratively calculated based on the lambda consensus of neighbor nodes. The lambda consensus algorithm achieves coordinated updates of local variables by introducing a communication mechanism between neighbor nodes. Each node calculates the consistency update value based on its own local variables and the information of neighbor nodes, so that the system as a whole gradually converges to a consistent solution. This process coordinates the solutions of each node to the vicinity of the global optimal solution through multiple rounds of iterative calculations, while maintaining a balance between local constraints and global consistency. The parallel characteristics of the lambda consensus algorithm enable it to make full use of distributed computing resources to ensure the efficiency and robustness of the solution. In the lambda consensus iterative calculation, in addition to updating local variables, it is also necessary to synchronize global variables and make convergence judgments on the consistency solution. The goal of global variable synchronization is to aggregate the local solutions of each node into an overall result, thereby ensuring the overall consistency of the optimization process. At the same time, by defining convergence judgment conditions, such as the amplitude of the solution change or the convergence rate of the objective function, the progress of the solution process is dynamically monitored, and the iteration is terminated when the convergence conditions are met. After completing global synchronization and convergence judgment, a distributed solution set containing the global optimal solution is generated.

[0020] In one example, the voltage, current, and temperature parameters of the charging pile are subjected to Hilbert transform to obtain envelope data of a multi-channel analytical signal, including: Input the voltage, current and temperature parameters into a high-precision analog-to-digital converter for digital conversion to obtain digital parameter data; The orthogonal component data is obtained by performing convolution calculation on the digital parameter data through the Hilbert kernel function based on the Fourier series; Performing complex synthesis on the orthogonal component data and the digital parameter data to obtain analytical signal data, and performing square and square root operations on the real and imaginary parts of the analytical signal data to obtain envelope amplitude data; The envelope amplitude data is aligned in the time domain based on linear interpolation to obtain synchronous envelope data, and the synchronous envelope data is grouped and merged according to the voltage, current and temperature channels to obtain the envelope data of the multi-channel analytical signal.

[0021] In this example, a high-precision analog-to-digital converter is used to convert the key analog signals (voltage , Current ,temperature ) is converted into corresponding digital signals. These parameters are the core indicators reflecting the operating status of the charging pile. Assuming the sampling frequency is , then at any point in time Next, the analog signal Discretized into digital signals ,in is the sampling value at the corresponding time point. Taking the voltage signal as an example, if Represents a sine wave signal. After being converted by the analog-to-digital converter, its discretized form is ,in is the signal frequency, is the sampling frequency. The orthogonal component data is obtained by convolution calculation of the digital parameter data based on the Hilbert kernel function of the Fourier series. The Hilbert transform generates the orthogonal component (imaginary part) of the signal through the phase shift operation in the frequency domain. For the discretized signal , whose Hilbert kernel function is expressed as (when )and By and Perform convolution calculation to obtain orthogonal components ,Right now: ; in, is the quadrature component (imaginary part) of the signal, is the sample value of the original signal, is the corresponding value of the kernel function. For the voltage signal , whose orthogonal components It's a phase shift. The cosine signal of By obtaining the orthogonal component data and the original signal , complex synthesis is performed to obtain analytical signal data The formula for complex number synthesis is: ; in, is the analytical signal, is an imaginary unit, is the real part of the signal, is the imaginary part of the signal. For example, for a voltage signal and its orthogonal components , the analytical signal is written as: ; For analytical signals Perform square root operations on the real and imaginary parts to obtain the envelope amplitude data The envelope amplitude is calculated as: ; For the voltage signal , its envelope amplitude is: ; For envelope amplitude data Perform time domain alignment based on linear interpolation. Unify the data of different channels (such as voltage, current, temperature) to the same time base. Assume that the sampling frequency of the voltage signal is , the current signal is , the temperature signal is , through linear interpolation, the envelope amplitude data of these signals are mapped to a unified time axis. For example, the linear interpolation formula is: ; in is the interpolated value, and are the two nearest sampling points, and is the corresponding sampling value. By interpolation, the envelope amplitudes of voltage, current, and temperature are aligned to the same time axis to form synchronous envelope data. The synchronous envelope data are grouped and merged according to the voltage, current, and temperature channels to generate envelope data of multi-channel analytical signals. For example, for the envelope data of the voltage channel , envelope data of current channel , envelope data of temperature channel , multi-channel envelope data is expressed in matrix form: .

[0022] In one example, frequency domain conversion and maximum a posteriori probability matrix decomposition are performed on the envelope data to obtain a spectrum feature vector, including: Perform discrete Fourier transform on the envelope data through a frequency grid with a resolution of 0.1 Hz to obtain initial spectrum data, and perform matrix sparsification on the initial spectrum data with L1 norm constraints to obtain a sparse spectrum matrix; The sparse spectrum matrix is ​​input into the Bayesian prior probability model for parameter optimization to obtain the spectrum prior distribution, and the spectrum prior distribution is iteratively calculated using the maximum a posteriori probability criterion to obtain the posterior probability density function; The maximum likelihood parameters are solved according to the posterior probability density function to obtain multi-channel spectrum consistency data, and the multi-channel spectrum consistency data is transformed into real values ​​and reduced in dimension to obtain the spectrum feature vector.

[0023] In this example, the envelope data Perform a discrete Fourier transform and select a frequency grid with a resolution of 0.1 Hz. The interval within the frequency range is 0.1 Hz. Assume that the sampling frequency is , the number of sampling points is , frequency resolution Defined as In order to achieve a resolution of 0.1 Hz, it is necessary to ensure Large enough so that Hz. The formula for discrete Fourier transform is: ; in, is the frequency The corresponding spectrum value is The envelope data is The value of the sampling point, is an imaginary unit, is the frequency index. By calculating , get the initial spectrum data, including the amplitude and phase information of the signal. Perform L1 norm constraint matrix sparseness on the initial spectrum data to get a sparse spectrum matrix. The L1 norm constraint adds a sparse regularization term so that the optimization objective function gives priority to a small number of important frequency components when solving, while compressing the remaining irrelevant components to zero. Assume is the initial spectrum matrix, and the sparsification problem is expressed as the following optimization problem: ; in, is the reconstruction error term, representing the sparse matrix With the original spectrum matrix differences; is a sparse regularization term, which represents the L1 norm of the matrix (i.e., the sum of the absolute values ​​of all elements); is the sparsification weight parameter, which controls the balance between sparsity and reconstruction error. By solving this optimization problem, the sparse spectrum matrix is ​​obtained. , which only retains the frequency components that are significant to the signal. The sparse spectrum matrix is ​​input into the Bayesian prior probability model for parameter optimization to obtain the spectrum prior distribution. The Bayesian model defines the prior distribution by assuming that the amplitude of the sparse spectrum follows a certain probability distribution (such as Gaussian distribution or Laplace distribution) and combining historical data or expert knowledge. . Assume that the likelihood function of the measured data is , and the posterior distribution is obtained through the Bayesian formula: ; in, is the marginal distribution, ensuring the normalization of the posterior distribution. The spectrum prior distribution is iteratively calculated through the maximum a posteriori probability criterion to find a set of parameters maximize The optimization problem of maximum a posteriori probability is: ; In each iteration, the parameters are updated to approximate the optimal posterior probability density function. Based on the posterior probability density function, the maximum likelihood estimation method is used to solve the multi-channel spectrum consistency data. The maximum likelihood method selects a set of parameters , so that the probability of data occurrence is maximized. Define the likelihood function of consistent data , and its optimization goal is: ; in, represents the spectrum consistency parameter, Indicates the number of channels, It is The sparse spectrum data of the channel. By solving this problem, the consistency data of the multi-channel spectrum is obtained. The multi-channel spectrum consistency data is transformed into a real value and reduced in dimension to generate a spectrum feature vector. The real value transformation converts the complex spectrum into a symmetrical real form by taking half the frequency band of the sparse spectrum. The formula is: ; in, and Represent the real and imaginary parts of the spectrum respectively. Through dimensionality reduction algorithms (such as principal component analysis), the most representative low-dimensional feature vectors are extracted from high-dimensional data. For example, for the spectrum data matrix , principal component analysis finds the main direction of the data through eigenvalue decomposition and generates a spectrum feature vector ,Right now: ; in, is a dimension reduction matrix that defines the weights of the main components. Through the above steps, the conversion, sparsification, Bayesian modeling and feature extraction of spectral data are realized, and the spectral feature vector for fault diagnosis is generated.

[0024] In one example, the spectrum feature vector is input into a dual control law observer with continuous terms and discontinuous terms for state tracking, and a system residual matrix is ​​obtained, including: The frequency spectrum feature vector is input into the state space model to analyze the dynamic characteristics and obtain the system state equation; According to the system state equation, a sliding surface design based on a sign function is established to obtain a discontinuous control law, and the system state equation is processed continuously based on a hyperbolic tangent function to obtain a continuous control law; According to the error size of the system state equation, the discontinuous control law and the continuous control law are adaptively weighted to obtain the combined control law; The combined control law is input into the repetitive controller to compensate for periodic disturbances, and the compensation control quantity is obtained. The convergence analysis of the compensation control quantity based on Lyapunov stability is performed to obtain the state observation result. The residual is calculated based on the state observation results and the spectrum feature vector to obtain the error sequence, and the error sequence is reconstructed by sliding to obtain the system residual matrix.

[0025] In this example, the spectral feature vector Input state space model to analyze the dynamic characteristics of the system. Assume that the dynamic behavior of the system is described by a discrete time state space model in the form of: ; ; in, Indicates that the system The state vector at the moment; is the control input; is the output of the system; It is the system matrix, input matrix and output matrix, which describe the dynamic relationship of the system; and Represent process noise and measurement noise respectively. By analyzing the distribution and evolution of the spectrum feature vector, we estimate The parameters of the system are used to obtain the state equation that describes the dynamic characteristics of the system. After obtaining the system state equation, the sliding surface is designed based on the sign function to construct the discontinuous control law. Sliding mode control is to force the system state to the sliding surface and slide along the sliding surface to achieve the control target. The sliding surface is defined as: ; in, is the sliding surface function, whose zeros represent the ideal system trajectory. The discontinuous control law is defined by the symbolic function as: ; in, is the sliding mode gain; is a symbolic function defined as: ; The role of the sign function is to quickly respond to deviations, but its inherent discontinuity has the potential to introduce chattering problems. In order to alleviate the chattering phenomenon caused by the discontinuous control law, the system state equation is processed continuously based on the hyperbolic tangent function to design a continuous control law. The hyperbolic tangent function is a smooth nonlinear function defined as: ; Based on this, the expression of the continuous control law is: ; in, is the continuous control gain. Compared with the sign function, the hyperbolic tangent function can transition smoothly, reduce the jitter of the control input, and is suitable for handling small-scale errors. According to the error size of the system state equation, the discontinuous control law and the continuous control law are adaptively weighted to generate a combined control law. The strategy of adaptive weight allocation is to dynamically adjust the ratio of the two control laws based on the error amplitude. Assume that the error amplitude is , then the weight distribution is defined as: ; in, and is the weight of the discontinuous control law and the continuous control law; is a small positive number used to prevent the denominator from being zero. The expression of the combined control law is: ; Through weight distribution, the combined control law has both the rapidity of discontinuous control and the smoothness of continuous control. The combined control law is input into the repetitive controller to compensate for periodic disturbances. The repetitive controller is a control method optimized for periodic signals. It gradually reduces the steady-state deviation of the system by memorizing and compensating for repetitive errors. To compensate the control amount, the update rule is: ; in, is the expected output, is the compensation gain. After multiple rounds of iterations, the output of the repetitive controller can effectively eliminate periodic disturbances. The convergence analysis of the compensation control quantity based on the Lyapunov stability theory is carried out to verify the stability of the system. The Lyapunov function is defined as: ; in, is a symmetric positive definite matrix. If monotonically decreases over time, the system is Lyapunov stable. According to the state observation results With the spectral feature vector Perform residual calculations and generate error sequences: ; The error sequence is reconstructed by sliding, by introducing a sliding window Take a weighted average: ; The error sequence after sliding reconstruction is used to generate the system residual matrix , which is of the form: ; The system residual matrix is ​​used to characterize the deviations in the dynamic characteristics of the system.

[0026] In one example, threshold detection and cumulative error integration operations are performed on the system residual matrix to obtain fault feature data, including: The real-time data segment of the system residual matrix is ​​input into the threshold logic detector for boundary comparison operation to obtain the instantaneous fault detection result; According to the instantaneous fault detection results, a probability density function based on normal distribution is established to obtain the fault probability distribution data; Perform interval segmentation processing on the fault probability distribution data to obtain the fault level threshold, and perform data accumulation operation on the system residual matrix to obtain the historical data sequence; Perform an exponentially weighted cumulative integral operation on the historical data sequence to obtain the fault cumulative characteristic value, and perform a multi-dimensional feature fusion operation based on the fault level threshold and the fault cumulative characteristic value to obtain the fault type mark; The fault type mark is subjected to feature extraction operation through sparse envelope spectrum analysis to obtain the fault frequency component, and the matching degree between the fault frequency component and the preset fault feature pattern is calculated to obtain the fault feature data.

[0027] In this example, the real-time data segment of the system residual matrix is Input threshold logic detector, which performs boundary comparison operation to identify transient faults. Assume that the normal operation state of the system corresponds to an allowable residual range , then for each time point , the judgment condition is defined as: ; in, It is the instantaneous fault detection result. A value of 1 indicates an abnormality and a value of 0 indicates normality. , quickly determine whether the system has abnormal behavior. According to the instantaneous fault detection results, a probability density function based on normal distribution is established , to quantify the probability characteristics of the failure. Assuming that the residual of the system obeys the normal distribution under normal conditions, the probability density function is expressed as: ; in, is the mean of the residuals, reflecting the central value under normal conditions; is the standard deviation, which describes the dispersion of the data; is the residual value. By statistically analyzing the instantaneous fault detection results, it is estimated and The fault probability distribution is constructed by dividing the fault probability distribution data into interval segments and defining different fault level thresholds. For example, according to the distribution characteristics of the probability density, it is divided into three levels: minor fault, medium fault and severe fault, corresponding to the threshold intervals , and The segmented processing can transform the continuous fault probability information into discrete fault levels. In order to capture the long-term behavior characteristics of the system, the system residual matrix Perform data accumulation operations to generate historical data sequences Assumptions is the cumulative time window length, then the historical data series is defined as: ; in, Indicates from time arrive The cumulative residual of . By introducing a time window, short-term fluctuations in data are smoothed and long-term trends are highlighted. Perform exponentially weighted cumulative integral operation to calculate the fault cumulative characteristic value The exponentially weighted integral is defined as: ; in, is a weighting factor used to control the importance decay of historical data; is the length of the time window. Through exponential weighting, the contribution of recent data to fault characteristics is highlighted, while the impact of long-term data is reduced. , perform multi-dimensional feature fusion operations to generate fault type signs. Assume Exceeding different thresholds When corresponding to different fault types, the fault type flag is defined as: ; Through feature fusion operation, the fault type is dynamically identified according to the change trend of the cumulative feature value. Based on sparse envelope spectrum analysis, feature extraction operation is performed on the fault type mark to extract the fault frequency component. Assume that the fault signal The sparse spectrum of , and its calculation formula is: ; Through sparse processing, the key frequency components are highlighted in the spectrum and noise interference is filtered out. , the extracted frequency components Indicates the characteristic frequency of system failure. The matching degree is calculated with the preset fault feature pattern to obtain the fault feature data. The matching degree is defined using a similarity measurement formula, such as cosine similarity: ; in, is the extracted frequency component vector, is the preset fault pattern vector. By calculating the similarity, the matching degree between the current fault and the known pattern is judged, and then the fault feature data is output.

[0028] In one example, a lambda consensus iteration operation is performed on the fault feature data to obtain a control parameter compensation scheme, including: The fault feature data is input into the repair target optimization model to construct the cost function and obtain a multi-dimensional repair target function; Decouple and separate the control parameters and compensator parameters of the multi-dimensional repair objective function to obtain the optimized variable space; The optimization variable space is decomposed and optimized through the Lagrange duality principle to obtain a set of sub-problems, and the sub-problems are parallelly calculated using lambda consensus to obtain a distributed solution set. The distributed solution set is iterated by gradient descent based on adaptive step size to obtain iterative convergence results, and the stability of the iterative convergence results is analyzed to obtain steady-state solutions; The control system of the charging pile is optimized according to the steady-state solution to obtain a control parameter compensation scheme.

[0029] In this example, the fault signature data Input the repair target optimization model to construct a multi-dimensional repair objective function. Comprehensively consider the multi-dimensional performance indicators of the charging pile operation, including power stability , Energy loss , Temperature deviation Etc. Assume that these indicators are defined as: ; ; ; in, is the actual power, is the target power, is the power loss, and are the actual and target temperatures, respectively. The constructed multidimensional repair objective function is a weighted combination of the above performance indicators: ; in, is the weight coefficient, reflecting the importance of each indicator. Control parameters of the multidimensional restoration objective function and compensator parameters Decoupling separation, get the optimization variable space. Control parameters Includes voltage, current and temperature control variables; compensator parameters It is used to dynamically compensate for disturbances and faults. The optimization variable space is defined as: ; in, are inequality constraints, which represent physical boundary conditions, such as upper and lower limits on voltage and current; is an equality constraint, which indicates the equilibrium condition of the system. Through decoupling and separation, the optimization problem is divided into two independent subspaces, which are optimized for control parameters and compensator parameters respectively. In order to solve the optimization problem, the Lagrangian duality principle is used to decompose and optimize the optimization variable space to generate a set of subproblems. The Lagrangian function of the optimization problem is defined as: ; in, and is the Lagrange multiplier, which is used to introduce constraints. Through the duality principle, the original problem is decomposed into multiple sub-problems: ; ; Through this step, the global optimization problem is decomposed into a set of independent sub-problems. For the generated sub-problem set, use Consensus parallel computing method to achieve decoupled collaboration among multiple nodes. Consensus calculations achieve consistency through the following update rules: ; ; in, and is the step size parameter, and are the gradients of the control parameters and compensator parameters, respectively. In each iteration, each node shares update information through neighboring nodes to gradually achieve the coordination and consistency of the global solution. The distributed solution set is iterated by gradient descent based on adaptive step size to improve the solution accuracy. The update rule of the adaptive step size strategy is: ; ; in, It is The step size of the iteration, is the step size adjustment factor. By dynamically adjusting the step size, a balance is achieved between fast convergence and stability. After the iteration converges, the results are analyzed for stability to verify the reliability of the system solution. The stability analysis is implemented through the Lyapunov function, which is defined as: ; in, is a state variable, is a positive definite matrix. If , then the system is stable. According to the steady-state solution, the parameters of the charging pile control system are optimized to obtain the control parameter compensation scheme.

[0030] In one example, the optimization variable space is decomposed and optimized by the Lagrange duality principle to obtain a set of sub-problems, and the sub-problems are parallelly calculated using lambda consensus to obtain a distributed solution set, including: Perform convex optimization decomposition of the repair objective function on the optimization variable space to obtain normalized constraints; Input the normalized constraint conditions into the Lagrange multiplier dual transformer to construct the dual problem and obtain the dual constraint equation; According to the dual constraint equation, KKT conditions and complementary slackness decomposition are performed to obtain a sub-problem sequence, and the sub-problem sequence is distributedly deployed based on a parallel computing architecture to obtain a parallel optimization subsystem; The local variables in the parallel optimization subsystem are iteratively calculated based on the lambda consensus of neighbor nodes to obtain a consistent solution, and the consistent solution is synchronized with global variables and convergence is determined to obtain a distributed solution set.

[0031] In this example, the convex optimization decomposition of the repair objective function is performed on the optimization variable space to obtain the normalized constraints. Assume that the repair objective function is ,in represents the state variable, represents the control variable, and the optimization goal is to minimize The value of , while satisfying the constraints. In order to facilitate the solution, the non-convex problem is decomposed into convex sub-problems. The optimization problem is expressed as: ; in, and Represent the inequality constraints and equality constraints respectively. Through convex optimization decomposition, the objective function Divided into multiple convex subfunctions , and normalize the constraints to be convex: ; in, and Respectively The local variables of each subproblem, and is the local constraint function of the subproblem. The decomposed problem is easy to solve in parallel. The normalized constraint conditions are input into the Lagrange multiplier dual transformer to construct the dual problem and generate the dual constraint equations. By introducing the Lagrange multiplier and , the Lagrangian function of the optimization problem is defined as: ; in, is the Lagrange multiplier corresponding to the inequality constraint, is the Lagrange multiplier corresponding to the equality constraint. The goal of the dual problem is to maximize the dual function: ; Through The dual decomposition of the global problem is transformed into multiple local dual constraint equations, each of which contains only local variables and constraints. According to the dual constraint equations, the dual problem is decomposed into a sequence of sub-problems using the KKT condition (Karush-Kuhn-Tucker condition) and complementary slackness decomposition. The KKT condition includes the following four parts: the original problem constraint and ; Dual feasibility ; Gradient conditions ; complementary slackness Through these conditions, the global problem is decomposed into several local sub-problems: ; in, is a local Lagrangian function, which only contains the The subproblems are deployed to a parallel computing architecture to improve computational efficiency, forming parallel optimization subsystems. Each subsystem solves its corresponding local optimization problem independently, while sharing key information with neighboring nodes to achieve collaborative solutions. In the parallel optimization subsystem, local variables are optimized based on neighboring nodes to ensure the consistency of the global solution. Consensus iteration calculation. The goal of consensus iteration is to make all local solutions gradually converge to the same through the interaction of neighboring nodes. and There is a communication link between them, and the consensus update rules are: ; ; in, Representation Node The neighbor set of and is the step size parameter. Through multiple rounds of iterations, the variables of each node gradually converge to the same. Global variable synchronization and convergence judgment are performed on the consistent solution to generate a distributed solution set. The synchronization process is to synchronize the local solutions. and Summarize into a global solution: ; in is the total number of subsystems. Convergence is determined by evaluating the change in the variable And the objective function change Is it less than the preset threshold? Finish.

[0032] Reference Figure 2 This embodiment provides a DC charging pile fault diagnosis and repair device, including: Transformation module 1, used to perform Hilbert transformation on the voltage, current and temperature parameters of the charging pile to obtain envelope data of the multi-channel analytical signal; Decomposition module 2, used for performing frequency domain conversion and maximum a posteriori probability matrix decomposition on the envelope data to obtain a frequency spectrum feature vector; The state tracking module 3 is used to input the spectrum feature vector into the dual control law observer with continuous terms and discontinuous terms to perform state tracking and obtain the system residual matrix; The integral operation module 4 is used to perform threshold detection and cumulative error integral operation on the system residual matrix to obtain fault feature data; Iteration module 5 is used to perform lambda consensus iteration operation on the fault feature data to obtain a control parameter compensation scheme.

[0033] In this embodiment, for the specific implementation of each unit in the above device embodiment, please refer to the above method embodiment, which will not be repeated here.

[0034] Reference Figure 3 In an embodiment of the present invention, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 3 As shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0035] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0036] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0037] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided by the present invention and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

[0038] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0039] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A DC charging pile fault diagnosis and repair method, characterized in that: The following steps are involved: Perform Hilbert transform on the voltage, current and temperature parameters of the charging pile to obtain the envelope data of the multi-channel analytical signal; Performing frequency domain conversion and maximum a posteriori probability matrix decomposition on the envelope data to obtain a frequency spectrum feature vector; Inputting the frequency spectrum feature vector into a dual control law observer having continuous terms and discontinuous terms for state tracking to obtain a system residual matrix; Performing threshold detection and cumulative error integration operations on the system residual matrix to obtain fault feature data; A lambda consensus iterative operation is performed on the fault characteristic data to obtain a control parameter compensation scheme.

2. The DC charging pile fault diagnosis and repair method according to claim 1, characterized in that: The Hilbert transform is performed on the voltage, current and temperature parameters of the charging pile to obtain envelope data of the multi-channel analytical signal, including: Input the voltage, current and temperature parameters into a high-precision analog-to-digital converter for digital conversion to obtain digital parameter data; Performing convolution calculation on the digital quantity parameter data by a Hilbert kernel function based on Fourier series to obtain orthogonal component data; Performing complex synthesis on the orthogonal component data and the digital parameter data to obtain analytical signal data, and performing square and square root operations on the real and imaginary parts of the analytical signal data to obtain envelope amplitude data; The envelope amplitude data is aligned in the time domain based on linear interpolation to obtain synchronous envelope data, and the synchronous envelope data is grouped and merged according to voltage, current and temperature channels to obtain envelope data of a multi-channel analytical signal.

3. The DC charging pile fault diagnosis and repair method according to claim 1, characterized in that: The step of performing frequency domain conversion and maximum a posteriori probability matrix decomposition on the envelope data to obtain a frequency spectrum feature vector includes: Performing discrete Fourier transform on the envelope data through a frequency grid with a resolution of 0.1 Hz to obtain initial spectrum data, and performing matrix sparsification with L1 norm constraint on the initial spectrum data to obtain a sparse spectrum matrix; Inputting the sparse spectrum matrix into a Bayesian prior probability model for parameter optimization to obtain a spectrum prior distribution, and iteratively calculating the spectrum prior distribution using a maximum a posteriori probability criterion to obtain a posterior probability density function; The maximum likelihood parameters are solved according to the posterior probability density function to obtain multi-channel spectrum consistency data, and the multi-channel spectrum consistency data is transformed into a real value and reduced in dimension to obtain a spectrum feature vector.

4. The DC charging pile fault diagnosis and repair method according to claim 1, characterized in that: The frequency spectrum feature vector is input into a dual control law observer having continuous terms and discontinuous terms for state tracking to obtain a system residual matrix, including: Inputting the frequency spectrum feature vector into a state space model to perform dynamic characteristic analysis to obtain a system state equation; A sliding surface design based on a sign function is established according to the system state equation to obtain a discontinuous control law, and the system state equation is continuous processed based on a hyperbolic tangent function to obtain a continuous control law; Adaptively weighting the discontinuous control law and the continuous control law according to the error size of the system state equation to obtain a combined control law; Inputting the combined control law into a repetitive controller to perform periodic disturbance compensation to obtain a compensation control quantity, and performing a convergence analysis on the compensation control quantity based on Lyapunov stability to obtain a state observation result; Residual calculation is performed according to the state observation result and the frequency spectrum feature vector to obtain an error sequence, and sliding reconstruction is performed on the error sequence to obtain a system residual matrix.

5. The DC charging pile fault diagnosis and repair method according to claim 1, characterized in that: The performing threshold detection and cumulative error integration operation on the system residual matrix to obtain fault feature data includes: Inputting the real-time data segment of the system residual matrix into a threshold logic detector for boundary comparison operation to obtain an instantaneous fault detection result; Establishing a probability density function based on normal distribution according to the instantaneous fault detection result to obtain fault probability distribution data; Performing interval segmentation processing on the fault probability distribution data to obtain a fault level threshold, and performing data accumulation operation on the system residual matrix to obtain a historical data sequence; Performing an exponentially weighted cumulative integral operation on the historical data sequence to obtain a fault cumulative characteristic value, and performing a multi-dimensional feature fusion operation on the fault level threshold and the fault cumulative characteristic value to obtain a fault type flag; A feature extraction operation is performed on the fault type mark through sparse envelope spectrum analysis to obtain a fault frequency component, and a matching degree calculation is performed based on the fault frequency component and a preset fault feature pattern to obtain fault feature data.

6. The DC charging pile fault diagnosis and repair method according to claim 1, characterized in that: The performing a lambda consensus iterative operation on the fault characteristic data to obtain a control parameter compensation scheme includes: Inputting the fault characteristic data into a repair target optimization model to construct a cost function to obtain a multi-dimensional repair target function; Decoupling and separating the control parameters and the compensator parameters of the multi-dimensional repair objective function to obtain an optimized variable space; Decomposing and optimizing the optimization variable space by Lagrangian duality principle to obtain a set of subproblems, and performing λ consensus parallel computing on the set of subproblems to obtain a distributed solution set; Performing gradient descent iterative processing based on an adaptive step size on the distributed solution set to obtain an iterative convergence result, and performing stability analysis on the iterative convergence result to obtain a steady-state solution; The control system of the charging pile is optimized according to the steady-state solution to obtain a control parameter compensation scheme.

7. The DC charging pile fault diagnosis and repair method according to claim 6, characterized in that: The optimization variable space is decomposed and optimized by the Lagrangian duality principle to obtain a set of sub-problems, and the sub-problems are parallelly calculated by lambda consensus to obtain a distributed solution set, including: Performing convex optimization decomposition of the repair objective function on the optimization variable space to obtain normalized constraint conditions; Inputting the normalized constraint condition into a Lagrange multiplier dual transformer to construct a dual problem, thereby obtaining a dual constraint equation; Perform KKT condition and complementary slackness decomposition according to the dual constraint equation to obtain a sub-problem sequence, and perform distributed deployment of the sub-problem sequence based on a parallel computing architecture to obtain a parallel optimization subsystem; The local variables in the parallel optimization subsystem are iteratively calculated based on the lambda consensus of neighbor nodes to obtain a consistent solution, and the consistent solution is synchronized with global variables and convergence is determined to obtain a distributed solution set.

8. A DC charging pile fault diagnosis and repair device, characterized in that: For implementing the steps of the method according to any one of claims 1 to 7, the device comprises: A transformation module is used to perform Hilbert transformation on the voltage, current and temperature parameters of the charging pile to obtain envelope data of a multi-channel analytical signal; A decomposition module, used for performing frequency domain conversion and maximum a posteriori probability matrix decomposition on the envelope data to obtain a frequency spectrum feature vector; A state tracking module, used for inputting the spectrum feature vector into a dual control law observer having continuous terms and discontinuous terms for state tracking to obtain a system residual matrix; An integral operation module, used for performing threshold detection and cumulative error integral operation on the system residual matrix to obtain fault feature data; The iteration module is used to perform a lambda consensus iteration operation on the fault characteristic data to obtain a control parameter compensation scheme.

9. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images