An intelligent charging pile management system and method with remote diagnosis function

Abstract

This application provides an intelligent charging pile management system and method with remote diagnostic capabilities, relating to the field of power control technology. It involves performing one-dimensional variational mode decomposition and multi-dimensional variational mode decomposition on the load data and operating status signals of the charging pile, respectively, to obtain stationary components, noise components, and a set of state sub-mode signals. Based on the stationary components, external environment data, and noise components, load identification characteristics are determined. The state sub-mode signal set is then jointly analyzed to obtain fault diagnosis characteristics of the charging pile. The load identification characteristics and fault diagnosis characteristics are fused to obtain a mode fusion matrix. This matrix is ​​then subjected to joint diagnostic analysis to obtain the load prediction curve and fault diagnosis results of the charging pile. Based on the load prediction curve and fault diagnosis results, the charging pile is remotely dispatched. This application can perform collaborative analysis of the charging pile based on load prediction and fault diagnosis to improve the accuracy of remote dispatch control.

Description

Technical Field

[0001] This application relates to the field of power control technology, and more specifically, to an intelligent charging pile management system and method with remote diagnostic capabilities. Background Technology

[0002] Power control is the process of achieving efficient distribution and safe operation of electrical energy through signal acquisition, data processing, and execution control, targeting the power generation, transmission, distribution, and consumption processes of a power system. Power control encompasses multiple aspects such as monitoring the operating status of power equipment, load forecasting, dispatch optimization, and fault diagnosis, and is a core supporting technology for ensuring the stability and economy of modern energy systems.

[0003] Intelligent charging pile management, building upon traditional charging pile control and scheduling, incorporates technologies such as information collection, data analysis, pattern recognition, and intelligent decision-making to monitor the operational status of charging piles in real time, optimize load allocation, and remotely control them. This approach aims to improve the reliability, intelligence, and operational efficiency of charging piles. However, current remote diagnostic methods for charging piles treat load forecasting and fault diagnosis as two independent processes, lacking synergistic utilization. This results in limited intelligence in the diagnostic results for scheduling strategies. Furthermore, load analysis for charging piles often involves single-dimensional or simple filtering, failing to effectively separate noise from useful signals. Remote scheduling is triggered by a single indicator without considering dynamic load trends and fault diagnosis results, easily leading to false triggers or missed triggers. Therefore, how to conduct synergistic analysis of charging piles based on load forecasting and fault diagnosis to improve the accuracy of remote scheduling and control remains a challenge for the industry. Summary of the Invention

[0004] This application provides an intelligent charging pile management system and method with remote diagnostic capabilities, which can perform collaborative analysis of charging piles based on load forecasting and fault diagnosis to improve the accuracy of remote scheduling and control.

[0005] In a first aspect, this application provides an intelligent charging pile management system and method with remote diagnostic capabilities, the management method comprising the following steps:

[0006] Acquire external environmental data of the charging pile, and collect load data and operating status signals of the charging pile;

[0007] The load data is subjected to one-dimensional variational mode decomposition to obtain stationary components and noise components. The operating state signal is subjected to multi-dimensional variational mode decomposition to obtain a set of state sub-mode signals.

[0008] The noise component is correlated and identified based on the stationary component and the external environment data to obtain the load identification characteristics of the charging pile. The state sub-mode signal set is subjected to joint analysis and processing based on instantaneous frequency and spectral entropy to obtain the fault diagnosis characteristics of the charging pile.

[0009] The load identification features and the fault diagnosis features are fused based on feature correlation weight adaptive allocation to obtain a modal fusion matrix. The modal fusion matrix is ​​then subjected to joint diagnostic analysis based on a bidirectional long short-term memory network to obtain the load prediction curve and fault diagnosis results of the charging pile.

[0010] When the load prediction curve reaches the preset dynamic early warning range, the charging pile is remotely dispatched based on the fault diagnosis results.

[0011] In this embodiment, external environmental data of the charging pile is acquired through a sensor array.

[0012] In this embodiment, voltage and current sensors are used to collect load data of the charging pile, and acceleration and acoustic emission sensors are used to collect operating status signals of the charging pile.

[0013] In this embodiment, performing one-dimensional variational mode decomposition on the load data to obtain stationary components and noise components specifically includes:

[0014] The load data is preprocessed by normalization and noise suppression to obtain the processed load data;

[0015] Multiple intrinsic mode functions are determined based on the number of decomposition levels and penalty factor of one-dimensional variational mode decomposition using the processed load data;

[0016] By dividing the frequency of each intrinsic mode function, stationary components and noise components are obtained.

[0017] In this embodiment, performing multidimensional variational mode decomposition on the operating state signal to obtain a set of state sub-mode signals specifically includes:

[0018] The operating status signal is simultaneously calibrated through multiple channels to obtain the calibrated operating status signal;

[0019] Multiple multidimensional intrinsic mode functions are determined based on the number of decomposition levels and penalty factor of multidimensional variational mode decomposition using the calibrated operating status signal.

[0020] Based on the frequency distribution and energy density, all multidimensional intrinsic mode functions are grouped into states to obtain a set of state submode signals.

[0021] In this embodiment, the correlation identification of the noise component based on the stationary component and the external environment data to obtain the load identification features of the charging pile specifically includes:

[0022] The stationary components and the external environment data are timestamped to obtain a time-series correlated dataset.

[0023] Based on the correlation coefficient and mutual information entropy, correlation analysis is performed on the time-series associated dataset and the noise component, and then the load identification features of the charging pile are obtained from the analysis results.

[0024] In this embodiment, the fault diagnosis features of the charging pile are obtained by performing joint analysis processing based on instantaneous frequency and spectral entropy on the set of state sub-mode signals. Specifically, this includes:

[0025] Perform a Hilbert transform on the set of state submode signals to obtain the instantaneous frequency sequences corresponding to different submode signals;

[0026] Based on the instantaneous frequency sequence, the frequency value and spectral entropy value of the corresponding sub-mode signal are determined, and then the frequency entropy feature pairs corresponding to different sub-mode signals are obtained;

[0027] After normalizing all frequency entropy features, principal component analysis is performed to obtain the fault diagnosis features of the charging pile.

[0028] In this embodiment, the modality fusion matrix obtained by adaptively allocating feature correlation weights based on feature identification features and fault diagnosis features specifically includes:

[0029] The load identification features and the fault diagnosis features are synchronized in time before the correlation coefficient is calculated, thereby obtaining the feature correlation coefficient matrix;

[0030] The adaptive weight allocation coefficients are determined based on the aforementioned feature correlation coefficient matrix;

[0031] The load identification features and the fault diagnosis features are weighted and fused using the adaptive weight allocation coefficients to obtain a modal fusion matrix.

[0032] In this embodiment, remote scheduling of the charging pile based on the fault diagnosis results specifically includes:

[0033] The remote dispatch command is determined based on the fault diagnosis results and the load prediction curve of the charging pile.

[0034] The remote scheduling command is sent to the charging pile, and the remote scheduling command is executed by the controller of the charging pile to obtain the remote scheduling result.

[0035] Secondly, this application provides an intelligent charging pile management system with remote diagnostic capabilities, used to execute an intelligent charging pile management method with remote diagnostic capabilities, the management system comprising:

[0036] The data acquisition module is used to acquire external environmental data of the charging pile, as well as load data and operating status signals of the charging pile.

[0037] The mode decomposition module is used to perform one-dimensional variational mode decomposition on the load data to obtain stationary components and noise components, and to perform multi-dimensional variational mode decomposition on the operating state signal to obtain a set of state sub-mode signals.

[0038] The feature extraction module is used to correlate and identify the noise component based on the stationary component and the external environment data to obtain the load identification features of the charging pile, and to perform joint analysis processing based on instantaneous frequency and spectral entropy on the state sub-mode signal set to obtain the fault diagnosis features of the charging pile.

[0039] The feature fusion and diagnosis module is used to perform feature fusion on the load identification features and the fault diagnosis features based on feature correlation weight adaptive allocation to obtain a modal fusion matrix. The modal fusion matrix is ​​then used for joint diagnostic analysis based on a bidirectional long short-term memory network to obtain the load prediction curve and fault diagnosis results of the charging pile.

[0040] The remote scheduling module is used to remotely schedule the charging piles based on the fault diagnosis results when the load prediction curve reaches the preset dynamic early warning range.

[0041] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects:

[0042] The system acquires external environmental data of the charging pile and collects its load data and operating status signals. It performs one-dimensional variational mode decomposition on the load data to obtain stationary and noise components, and performs multi-dimensional variational mode decomposition on the operating status signals to obtain a set of state sub-mode signals. Based on the stationary components and the external environmental data, it identifies the noise components to obtain the load identification features of the charging pile. It then performs joint analysis processing based on instantaneous frequency and spectral entropy on the set of state sub-mode signals to obtain the fault diagnosis features of the charging pile. Finally, it performs feature fusion based on adaptive allocation of feature correlation weights on the load identification features and the fault diagnosis features to obtain a mode fusion matrix. A bidirectional long short-term memory network is used to perform joint diagnostic analysis on the mode fusion matrix to obtain the load prediction curve and fault diagnosis results of the charging pile. When the load prediction curve reaches a preset dynamic warning interval, the system remotely schedules the charging pile based on the fault diagnosis results.

[0043] Therefore, this application demonstrates that charging piles can be collaboratively analyzed based on load forecasting and fault diagnosis. First, by acquiring external environmental data, load data, and operational status signals of the charging piles, a multi-source data foundation covering external disturbances and equipment operating characteristics is established, facilitating a comprehensive understanding of the charging piles' operational status and environmental impacts by the model. Then, by performing one-dimensional variational mode decomposition on the load data and multi-dimensional variational mode decomposition on the operational status signals, stationary components, noise components, and state sub-mode signal sets are extracted, achieving accurate separation of different signal components and providing identifiable signal inputs for subsequent feature extraction. Second, based on the correlation between the stationary components and external environmental data, the noise components are identified, extracting load identification features based on load variation patterns, and combining these with the state sub-mode data. By jointly analyzing the instantaneous frequency and spectral entropy of the state signal set, fault diagnosis features characterizing equipment operation anomalies are obtained, thus taking into account both load trends and operational health. Then, through feature fusion based on feature correlation weight adaptive allocation, load identification features and fault diagnosis features are fused into a modal fusion matrix, and a bidirectional long short-term memory network is used for joint diagnostic analysis to obtain a load prediction curve and fault diagnosis results that have both prediction accuracy and fault identification capabilities. Finally, when the load prediction curve reaches the dynamic warning interval, targeted remote scheduling is carried out in combination with the fault diagnosis results to achieve early intervention and precise control of potential risks, thereby effectively improving the response speed and decision accuracy of remote scheduling, reducing equipment operation risks, and optimizing the allocation of operation and maintenance resources.

[0044] In summary, the technical solution adopted in this application can perform collaborative analysis of charging piles based on load forecasting and fault diagnosis to improve the accuracy of remote dispatch control. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of this application 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 for this embodiment of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 This is a flowchart of a smart charging pile management method with remote diagnostic function provided in this application;

[0047] Figure 2 This is an exemplary flowchart for determining stationary components and noise components according to the present application;

[0048] Figure 3 This is an exemplary flowchart for determining the fault diagnosis characteristics of a charging pile according to the present application;

[0049] Figure 4 This is a module structure diagram of the management system provided in this application. Detailed Implementation

[0050] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0051] This application provides an intelligent charging pile management system and method with remote diagnostic capabilities. The core of the system involves acquiring external environmental data of the charging pile and collecting its load data and operating status signals. The load data undergoes one-dimensional variational mode decomposition to obtain stationary components and noise components. The operating status signals undergo multi-dimensional variational mode decomposition to obtain a set of state sub-mode signals. The noise components are correlated and identified based on the stationary components and the external environmental data to obtain the charging pile's load identification features. The state sub-mode signal set undergoes joint analysis based on instantaneous frequency and spectral entropy to obtain the charging pile's fault diagnosis features. The load identification features and fault diagnosis features are fused using feature correlation weight adaptive allocation to obtain a mode fusion matrix. The mode fusion matrix is ​​then subjected to joint diagnostic analysis based on a bidirectional long short-term memory network to obtain the charging pile's load prediction curve and fault diagnosis results. When the load prediction curve reaches a preset dynamic warning interval, the charging pile is remotely scheduled based on the fault diagnosis results.

[0052] Example 1: To better understand the above technical solution, the following will provide a detailed description of the technical solution in conjunction with the accompanying drawings and specific implementation methods. (Refer to...) Figure 1 As shown in the figure, this is an exemplary flowchart of a smart charging pile management method with remote diagnostic function according to this embodiment of the present application. The management method includes the following steps:

[0053] In step S1, external environmental data of the charging pile is acquired, and load data and operating status signals of the charging pile are collected.

[0054] In practice, the external environmental data of the charging pile is acquired through a sensor array. The sensor array includes temperature sensors, humidity sensors, wind speed sensors, and particulate matter concentration sensors deployed around the charging pile. The deployment location and number of the sensor array can be determined according to the spatial structure of the actual installation environment. In practice, the collected analog signals are converted into digital signals through the sensor array, and then the digital signals are used as the external environmental data of the charging pile.

[0055] It should be noted that the external environmental data in this application are measurements of physical and chemical quantities that describe the operating environment of the charging pile. The external environmental data includes ambient temperature, relative humidity, wind speed, wind direction, and atmospheric pressure, etc., and is used to characterize the impact of external environmental conditions on the heat dissipation efficiency, insulation performance, and operational stability of the charging pile.

[0056] In specific implementation, voltage and current sensors are used to collect load data of the charging pile, and acceleration and acoustic emission sensors are used to collect operating status signals of the charging pile. The load data, which describes the changes in electrical load carried by the charging pile within a specific time period, includes charging voltage, charging current, and charging power, and is used to assess the power supply capacity of the charging pile under different load conditions. The operating status signals represent non-electrical signals indicating the operating status of the charging pile, and include mechanical vibration signals, acoustic emission signals, and displacement change signals on the shell surface, used to identify mechanical abnormalities occurring during operation. Preferably, the sampling frequency of the acceleration sensor can be set between 1-20kHz, which is beneficial for comprehensively capturing the low-frequency and mid-to-high-frequency vibration characteristics of the charging pile's mechanical structure; the sampling frequency of the acoustic emission sensor can be set between 100kHz and 1MHz, which is beneficial for accurately recording wave signals generated by high-frequency transient events such as partial discharge.

[0057] It should be noted that the accelerometer in this embodiment is a sensing element that measures the magnitude and direction of the vibration acceleration of an object, and is used to collect mechanical vibration information of the charging pile shell or internal components during operation; the acoustic emission sensor is a transducer that can convert the high-frequency stress wave signal generated inside the material or structure into an electrical signal, and is used to collect the high-frequency acoustic emission signal generated by the internal electrical components, contactors or conductors of the charging pile when they are broken down or discharged, so as to facilitate the timely detection of early faults such as potential insulation breakdown and partial discharge.

[0058] In step S2, the load data is subjected to one-dimensional variational mode decomposition to obtain stationary components and noise components, and the operating state signal is subjected to multi-dimensional variational mode decomposition to obtain a set of state sub-mode signals.

[0059] Preferably, in this embodiment, reference Figure 2 As shown, this figure is an exemplary flowchart for determining stationary components and noise components according to the present application. In this embodiment, the one-dimensional variational mode decomposition of the load data to obtain stationary components and noise components can be achieved by the following steps:

[0060] In step S21, the load data is normalized and noise suppressed preprocessing to obtain processed load data;

[0061] In step S22, multiple intrinsic mode functions are determined based on the number of decomposition layers and penalty factor of one-dimensional variational mode decomposition using the processed load data.

[0062] In step S23, the frequency of each intrinsic mode function is divided to obtain stationary components and noise components.

[0063] In practical implementation, firstly, the minimum-maximum normalization method can be used to linearly map the load data to its minimum and maximum values ​​to eliminate the influence of dimensions. Then, the moving average of the load data is calculated over a time window to smooth random noise and abrupt changes. These steps yield the preprocessed load data. Next, the number of decomposition levels and the penalty factor of the one-dimensional variational mode decomposition algorithm are used as control parameters. Through iterative optimization, multiple intrinsic mode functions are decomposed from the processed load data. The number of decomposition levels can be determined based on the mode of the load data and can also be adjusted according to the actual data characteristics. When adjusted, the penalty factor is used to regulate the bandwidth. Preferably, in this embodiment, the penalty factor of the one-dimensional variational mode decomposition algorithm can be set between 2000 and 5000, which is beneficial to balance decomposition accuracy and computational efficiency. The frequency division threshold can be dynamically set according to the power spectral density analysis results. Finally, a fast Fourier transform is performed on each intrinsic mode function, that is, the intrinsic mode function is divided into low-frequency stationary components and high-frequency noise components according to the center frequency range. The low-frequency components can describe the stable change trend of the charging pile load, and the high-frequency components are used to identify instantaneous fluctuations caused by external disturbances or measurement noise.

[0064] It should be noted that the one-dimensional variational mode decomposition in this embodiment is an adaptive signal decomposition process based on the variational principle. By iteratively solving the center frequency and bandwidth of each mode in the frequency domain, multi-scale decomposition of non-stationary signals is achieved. The stationary component refers to the signal component that changes slowly and has a stable frequency over time, and is used to describe the load change trend of the charging pile during long-term operation. The noise component refers to the signal component that changes rapidly in frequency, has a small amplitude, and is highly random, and is mainly caused by electromagnetic interference, sensor measurement errors, or sudden changes in the external environment.

[0065] In this embodiment, the multidimensional variational mode decomposition of the operating state signal to obtain the state sub-mode signal set can be performed in the following manner:

[0066] The operating status signal is simultaneously calibrated through multiple channels to obtain the calibrated operating status signal;

[0067] Multiple multidimensional intrinsic mode functions are determined based on the number of decomposition levels and penalty factor of multidimensional variational mode decomposition using the calibrated operating status signal.

[0068] Based on the frequency distribution and energy density, all multidimensional intrinsic mode functions are grouped into states to obtain a set of state submode signals.

[0069] In specific implementation, firstly, the operating status signal is synchronously calibrated across multiple channels using a timestamp-based alignment method. This involves using the system's unified clock as a reference and interpolating and resampling the signal sequences acquired from each channel according to their sampling time, thus obtaining the calibrated operating status signal. Then, the synchronously calibrated multi-channel operating status signal is input into a multidimensional variational mode decomposition (MMD) algorithm. The number of decomposition levels and a penalty factor are set as control parameters. Through iterative optimization, multiple multidimensional intrinsic mode functions (IMFs) are simultaneously decomposed from the multi-channel signal. The number of decomposition levels in the multidimensional variational mode decomposition can be determined based on the length of the signal's spectral smoothing interval, and the penalty factor is used to approximate... In this embodiment, the penalty factor of the multidimensional variational mode decomposition layer can be set between 1500 and 4000, which is beneficial to achieving a balance between decomposition accuracy and computation speed. Finally, the center frequency (i.e., frequency distribution) of each multidimensional intrinsic mode function is extracted using fast Fourier transform, and the energy density (e.g., root mean square value or power spectrum integral) of the multidimensional intrinsic mode function is calculated. Modes with similar frequencies and energy levels are grouped into the same state group, and this state group is then used as a state submode signal set. Each submode signal in the state submode signal set is a feature characterizing the mechanical vibration mode or fault.

[0070] It should be noted that the multidimensional variational mode decomposition in this embodiment is an adaptive decomposition algorithm for multi-channel signals. By synchronously optimizing the center frequency and bandwidth of multi-channel signals in the frequency domain, it achieves time alignment and feature coordination of different channel information in the mode decomposition process. The multidimensional intrinsic mode function refers to the component signal with consistent frequency characteristics in multiple sensor signals obtained by multidimensional variational mode decomposition, which is used to represent the common dynamic characteristics of the device in a certain frequency band. The state sub-mode signal set is a set of modes obtained by dividing according to frequency and energy characteristics. The signals in each set have similar vibration characteristics or acoustic emission characteristics, which facilitates subsequent feature extraction and pattern recognition, and improves the accuracy of identifying mechanical faults, structural anomalies and early defects.

[0071] In addition, it should be noted that the one-dimensional variational mode decomposition and multi-dimensional variational mode decomposition in this application can be applied to different types of data respectively. The one-dimensional variational mode decomposition is used to process single-channel load data, while the multi-dimensional variational mode decomposition is used to process multi-channel operating status signals, which is beneficial for realizing the decomposition and extraction of signal features in different data dimensions.

[0072] In step S3, the noise component is correlated and identified based on the stationary component and the external environment data to obtain the load identification characteristics of the charging pile. The state sub-mode signal set is subjected to joint analysis processing based on instantaneous frequency and spectral entropy to obtain the fault diagnosis characteristics of the charging pile.

[0073] In this embodiment, the noise component is correlated and identified based on the stationary component and the external environmental data to obtain the load identification characteristics of the charging pile. Specifically, this can be achieved in the following manner:

[0074] The stationary components and the external environment data are timestamped to obtain a time-series correlated dataset.

[0075] Based on the correlation coefficient and mutual information entropy, correlation analysis is performed on the time-series associated dataset and the noise component, and then the load identification features of the charging pile are obtained from the analysis results.

[0076] In practical implementation, firstly, the sampling times of different data sources are interpolated and resampled to obtain the stationary component values ​​arranged in chronological order and the external environmental data as a time-series correlated dataset. Preferably, the time synchronization accuracy can be set at the second level, which is beneficial for the deviation of the correspondence between different signals. Then, the correlation coefficient and mutual information entropy between the time-series correlated dataset and the noise component are calculated. Mutual information entropy measures the degree of correlation between two variables in a nonlinear relationship and can be calculated using the mutual information entropy calculation formula. The correlation coefficient measures the degree of correlation between two variables in a linear relationship and can be calculated using the Pearson correlation coefficient calculation formula. In actual implementation, a mutual information entropy threshold and a correlation coefficient threshold are preset, and time-series correlated datasets and noise components that simultaneously meet the mutual information entropy threshold and the correlation coefficient threshold are selected. Then, the corresponding stationary component in the selected time-series correlated dataset is used as the load identification feature. Preferably, the mutual information entropy threshold can be set to 0.1-0.5 and the correlation coefficient threshold can be set to 0.6-0.9 to simultaneously meet the nonlinear and linear correlation requirements of the stationary component and the external environmental data.

[0077] It should be noted that the mutual information entropy in this application is an indicator based on probability distribution, used to represent the nonlinear dependency between variables; the correlation coefficient is an indicator based on numerical change trend, used to represent the linear correlation between variables; the load identification feature is a set of features representing the relationship between changes in the external environment and the load fluctuation of the charging pile, which can provide highly correlated input for load forecasting and condition diagnosis.

[0078] Preferably, in this embodiment, reference Figure 3As shown, this figure is an exemplary flowchart for determining the fault diagnosis characteristics of a charging pile according to the present application. In this embodiment, the fault diagnosis characteristics of the charging pile are obtained by performing joint analysis processing based on instantaneous frequency and spectral entropy on the state sub-mode signal set, which can be implemented by the following steps:

[0079] In step S31, the Hilbert transform is performed on the set of state sub-mode signals to obtain the instantaneous frequency sequences corresponding to different sub-mode signals;

[0080] In step S32, the frequency value and spectral entropy value of the corresponding sub-mode signal are determined based on the instantaneous frequency sequence, thereby obtaining the frequency entropy feature pairs corresponding to different sub-mode signals;

[0081] In step S33, all frequency entropy feature pairs are normalized before principal component analysis is performed to obtain the fault diagnosis features of the charging pile.

[0082] In specific implementation, firstly, the state sub-mode signals can be transformed using the Hilbert algorithm to obtain complex analytic signals. Then, this complex analytic signal is used as an instantaneous frequency sequence. Preferably, the Hilbert algorithm can be combined with a sliding window for segmented calculation to maintain the continuity of frequency changes. Next, for each sub-mode signal's corresponding instantaneous frequency sequence, the mean of the instantaneous frequency sequence can be used as the frequency value of the corresponding sub-mode signal. This frequency value describes the dominant vibration frequency of the sub-mode signal under its main operating state. After normalizing the spectral amplitude of the instantaneous frequency sequence corresponding to the sub-mode signal, statistical methods are used to determine the frequency band energy proportion. The uniformity (i.e., information entropy) of the frequency and spectral entropy of the sub-mode signal is used as the spectral entropy value, which is an indicator of the uncertainty of the signal energy distribution. The vector composed of the frequency value and spectral entropy value of the sub-mode signal is then used as the corresponding frequency entropy feature pair. In this way, the frequency entropy feature pairs corresponding to all sub-mode signals can be obtained. Finally, all normalized frequency entropy feature pairs are input into the principal component analysis algorithm. That is, by calculating the covariance matrix of the frequency entropy feature pairs, the principal component direction is extracted, which can compress the original multidimensional features into a low-dimensional space. The matrix composed of the frequency entropy feature pairs in the low-dimensional space is used as the fault diagnosis feature.

[0083] It should be noted that the fault diagnosis features in this application are feature matrices obtained by jointly analyzing instantaneous frequency and spectral entropy and then performing dimensionality reduction. These matrices can represent the vibration modes and energy distribution characteristics of the charging pile under different operating conditions, and are used to identify early faults and potential abnormal trends of the equipment. The instantaneous frequency in the instantaneous frequency sequence refers to the instantaneous frequency change of the signal at each time point, representing the dynamic frequency characteristics of the equipment under non-stationary conditions. The Hilbert algorithm can convert real-number signals into complex analytic signals and extract instantaneous frequencies. Principal component analysis is a linear dimensionality reduction technique that can reduce the feature dimension while retaining as much of the main feature information related to the fault as possible.

[0084] In step S4, feature fusion based on feature correlation weight adaptive allocation is performed on the load identification features and the fault diagnosis features to obtain a modal fusion matrix. The modal fusion matrix is ​​then subjected to joint diagnostic analysis based on a bidirectional long short-term memory network to obtain the load prediction curve and fault diagnosis results of the charging pile.

[0085] In this embodiment, the modality fusion matrix obtained by adaptively allocating feature correlation weights based on feature identification features and fault diagnosis features can be achieved by the following method:

[0086] The load identification features and the fault diagnosis features are synchronized in time before the correlation coefficient is calculated, thereby obtaining the feature correlation coefficient matrix;

[0087] The adaptive weight allocation coefficients are determined based on the aforementioned feature correlation coefficient matrix;

[0088] The load identification features and the fault diagnosis features are weighted and fused using the adaptive weight allocation coefficients to obtain a modal fusion matrix.

[0089] In practical implementation, firstly, linear interpolation can be used to interpolate and resample the load identification features and fault diagnosis features according to a unified clock. Then, the correlation coefficients of the interpolated and resampled load identification features and fault diagnosis features are calculated. The matrix composed of all correlation coefficient values ​​is then used as the feature correlation coefficient matrix, where each element represents the degree of linear correlation between the load identification features and fault diagnosis features. The correlation coefficient can be calculated using the Pearson correlation coefficient formula. Next, the sum of the absolute values ​​of the correlation coefficients in the feature correlation coefficient matrix is ​​calculated row by row, and all sums of absolute values ​​are normalized. The values ​​are normalized to between 0 and 1, and the sum of the normalized absolute values ​​is used as the initial weights. The initial weights are then smoothed using a moving average method, and the smoothed initial weights are used as adaptive weight allocation coefficients. Finally, the adaptive weight allocation coefficients are multiplied by the load identification features and fault diagnosis features respectively to obtain the weighted load identification features and fault diagnosis features. The weighted load identification features and fault diagnosis features are then concatenated into a two-dimensional matrix in the order of sampling time, and this two-dimensional matrix is ​​used as the modality fusion matrix. In this matrix, the rows correspond to the sampling time, and the columns correspond to the weighted feature dimensions.

[0090] It should be noted that the modality fusion matrix in this embodiment refers to a two-dimensional data structure that fuses different types of features; the adaptive weight allocation coefficient is a coefficient that is automatically adjusted based on the correlation between features, with features with high correlation having larger weights and features with low correlation having smaller weights; the feature correlation coefficient matrix is ​​a symmetric matrix, with the diagonal elements of the feature correlation coefficient matrix having a value of 1, indicating that the feature is completely correlated with itself, and the elements outside the diagonal are used to represent the degree of correlation between different features.

[0091] In this embodiment, the joint diagnostic analysis of the modal fusion matrix based on a bidirectional long short-term memory network to obtain the load prediction curve and fault diagnosis results of the charging pile can be performed in the following manner:

[0092] The modality fusion matrix is ​​used to divide the training set and the test set for load prediction, and the training set and the test set are respectively input into the forward layer and the backward layer of the bidirectional long short-term memory network to obtain bidirectional dependency features.

[0093] The load prediction vector and the fault probability vector are obtained from the output layer of the bidirectional long short-term memory network.

[0094] The load prediction vector is monitored and plotted to obtain the load prediction curve of the charging pile;

[0095] The fault probability vector and the preset fault threshold are used to determine the fault diagnosis result of the charging pile.

[0096] In practical implementation, firstly, the modality fusion matrix can be converted into a sample set using a sliding window slicing method. The sliding window length can be set to 24 time steps, and the window sliding step size can be set to 1 time step. The sample set is then proportionally divided into training and testing sets using the sliding window slicing method. 70% of the sample set can be allocated to the training set, and 30% to the testing set. The training and testing sets are then input into the forward and backward layers of a Bidirectional Long Short-Term Memory (BiLSTM) network, respectively. The forward layer of the BiLSTM can sequentially process data from the first time step to the last time step. The input data is processed sequentially at each time step. The reverse layer of BiLSTM can process the input data sequentially from the last time step to the first time step in reverse chronological order. Inside the forward and reverse layers, the Long Short-Term Memory (LSTM) units can control the transmission and forgetting of information through input gates, forget gates, and output gates, thereby extracting bidirectional time-dependent features. The bidirectional dependency features are the result of splicing two sets of hidden states extracted by the forward and reverse processing paths at the same time step, which can more comprehensively describe the dependency relationship between load identification features and fault diagnosis features in the modality fusion matrix.

[0097] In addition, in specific implementation, the hidden states of the forward and backward layers in the BiLSTM are concatenated along the feature dimension to obtain a bidirectional dependency feature vector. This bidirectional dependency feature vector is then passed through the fully connected output layer of the BiLSTM to output a load prediction vector and a fault probability vector. Then, the load prediction vector is visualized using a line graph, i.e., the load prediction curve of the charging pile is plotted with time on the horizontal axis and the predicted load value on the vertical axis. Finally, the fault probability vector can be read from the BiLSTM to output the magnitude of the time step, and the magnitude of the fault probability vector is compared with a preset fault threshold. When the fault probability corresponding to a certain time step is greater than or equal to the threshold, it is determined that there is a fault risk at that time step, and the corresponding time point is marked as a fault diagnosis result and output. The preset fault threshold can be set according to the actual service life and quality of the charging pile. The higher the fault threshold value, the more likely the diagnosis is to reduce false alarms, and the lower the fault threshold value, the more likely the diagnosis is to issue an early warning.

[0098] It should be noted that the bidirectional long short-term memory network is a recurrent neural network structure that combines forward and backward information. It can simultaneously utilize historical and future time information to improve the ability to capture patterns in time series data. The long short-term memory unit alleviates the gradient vanishing and gradient exploding problems in traditional recurrent neural networks during long sequence training by introducing gating mechanisms (including input gate, forget gate, and output gate). The load prediction vector refers to the numerical prediction result of the charging pile load value in the future time period, and the fault probability vector refers to the probability value of the charging pile failing at the corresponding time node. The fault diagnosis result refers to the structured conclusion of the health status of the charging pile equipment, including the fault level, which includes four levels: "minor", "moderate", "severe" and "dangerous".

[0099] In step S5, when the load prediction curve reaches the preset dynamic early warning range, the charging pile is remotely scheduled based on the fault diagnosis result.

[0100] It should be noted that in this application, when the load forecast curve reaches the preset dynamic warning interval, a warning mechanism will be triggered. The warning mechanism includes two steps: threshold determination and graded response. The dynamic warning interval refers to the dynamic upper and lower limit interval obtained by regression analysis based on historical operating load data. The upper and lower limits will be adaptively adjusted with changes in time, season, and charging pile usage frequency. The graded response can output the load risk level. The load risk level is the degree of overload risk divided based on the relative position of the load forecast curve in the dynamic warning interval, including four levels: "slight", "moderate", "severe", and "dangerous".

[0101] In this embodiment, the remote scheduling of charging piles based on the fault diagnosis results can be carried out in the following manner:

[0102] The remote dispatch command is determined based on the fault diagnosis results and the load prediction curve of the charging pile.

[0103] The remote scheduling command is sent to the charging pile, and the remote scheduling command is executed by the controller of the charging pile to obtain the remote scheduling result.

[0104] In specific implementation, a threshold determination algorithm is used to jointly analyze the fault diagnosis results and load prediction curves, and the joint analysis results are converted into remote dispatch commands based on a communication protocol. These remote dispatch commands may include: load reduction operation commands, phased current limiting commands, charging pause commands, and switching to backup charging pile commands. Different types of commands can be preset based on a combination of fault level and load risk level. The threshold determination algorithm is a multi-level threshold algorithm. Specifically, the fault level in the fault diagnosis results is quantified into a numerical score to obtain a fault score, and the load risk level is also quantified into a numerical score to obtain a load risk score. For example, both the fault level and load risk level can be quantified numerically as follows: "minor" is quantified as 1, "moderate" as 2, "severe" as 3, and "dangerous" as 4. This quantification method yields the fault score and load risk score. The fault score refers to the numerical result after quantifying the fault level, used to characterize the severity of the equipment fault; the load risk score refers to the numerical result after quantifying the load risk level. The result is used to characterize the degree of overload risk. Then, the fault score and load risk score are integrated through a weighted scoring model to calculate the comprehensive risk index. Preferably, the initial weight of the weighted scoring model can be set to 0.5, which can be set according to the actual diagnostic needs. That is, the initial mathematical expression of the weighted scoring model is: comprehensive risk index = (0.5 * fault score + 0.5 load risk score) / 4, where the denominator 4 can be set according to the maximum value of the quantitative score, so that the comprehensive risk index takes the value between 0 and 1. The comprehensive risk index is a unified dimension charging pile diagnostic decision basis, used to match the remote dispatch command corresponding to the risk level. Then, the comprehensive risk index is compared with multiple preset threshold intervals. The corresponding remote dispatch command is triggered according to the interval in which the comprehensive risk index is located. In addition, the preset threshold intervals include: load reduction operation interval, phased current limiting interval, charging suspension interval and switching to backup charging pile interval. These can be set according to the actual service life, quality and actual diagnostic needs of the charging pile. The value of the threshold interval can be set between [0,1], without specific limitation.

[0105] In addition, the remote scheduling command is sent to the control terminal of the charging pile through the communication module of the charging pile. Preferably, the communication module can use a 4G / 5G cellular network to ensure the real-time and reliability of the command. The controller of the charging pile receives and parses the remote scheduling command, and then controls the charging pile to perform the operation corresponding to the remote scheduling command (e.g., load reduction, current limiting, pausing charging and switching to standby). The status of the charging pile is also transmitted back through the communication module to obtain the remote scheduling result. The controller of the charging pile can be an embedded microprocessor control unit (MCU) or a programmable logic controller (PLC), both of which can be used to parse the remote scheduling command and drive the power electronic devices to perform corresponding control actions. The specific type of controller is not limited.

[0106] It should be noted that in this embodiment, the remote scheduling command is a set of commands generated based on the load trend and fault risk of the charging pile, which is used to optimize load allocation and reduce equipment risk; the remote scheduling result refers to the data on changes in the operating status generated after the charging pile executes the scheduling command, such as reduced power output, interruption of charging session, or smoothing of load curve.

[0107] In summary, the technical solution adopted in this application can perform collaborative analysis of charging piles based on load forecasting and fault diagnosis to improve the accuracy of remote dispatch control.

[0108] Example 2: This application provides an intelligent charging pile management system with remote diagnostic capabilities, referencing... Figure 4 As shown, this diagram is a module structure diagram of the management system provided in this application. The management system includes:

[0109] The data acquisition module 100 is used to acquire external environmental data of the charging pile, and to collect load data and operating status signals of the charging pile.

[0110] The mode decomposition module 200 is used to perform one-dimensional variational mode decomposition on the load data to obtain stationary components and noise components, and to perform multi-dimensional variational mode decomposition on the operating state signal to obtain a set of state sub-mode signals.

[0111] The feature extraction module 300 is used to correlate and identify the noise component based on the stationary component and the external environment data to obtain the load identification features of the charging pile, and to perform joint analysis processing based on instantaneous frequency and spectral entropy on the state sub-mode signal set to obtain the fault diagnosis features of the charging pile.

[0112] The feature fusion and diagnosis module 400 is used to perform feature fusion on the load identification features and the fault diagnosis features based on feature correlation weight adaptive allocation to obtain a modal fusion matrix. The modal fusion matrix is ​​then used for joint diagnostic analysis based on a bidirectional long short-term memory network to obtain the load prediction curve and fault diagnosis results of the charging pile.

[0113] The remote scheduling module 500 is used to remotely schedule the charging piles based on the fault diagnosis results when the load prediction curve reaches the preset dynamic early warning range.

[0114] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0115] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compactdisc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0116] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

Claims

1. A method for managing intelligent charging piles with remote diagnostic capabilities, characterized in that, The management method includes the following steps: Acquire external environmental data of the charging pile, and collect load data and operating status signals of the charging pile; The load data is subjected to one-dimensional variational mode decomposition to obtain stationary components and noise components. The operating state signal is subjected to multi-dimensional variational mode decomposition to obtain a set of state sub-mode signals. The noise component is correlated and identified based on the stationary component and the external environment data to obtain the load identification characteristics of the charging pile. The state sub-mode signal set is subjected to joint analysis and processing based on instantaneous frequency and spectral entropy to obtain the fault diagnosis characteristics of the charging pile. The load identification features and the fault diagnosis features are fused based on feature correlation weight adaptive allocation to obtain a modal fusion matrix. The modal fusion matrix is ​​then subjected to joint diagnostic analysis based on a bidirectional long short-term memory network to obtain the load prediction curve and fault diagnosis results of the charging pile. When the load prediction curve reaches the preset dynamic early warning range, the charging pile is remotely dispatched based on the fault diagnosis results.

2. The intelligent charging pile management method with remote diagnostic function as described in claim 1, characterized in that, The external environmental data of the charging pile is acquired through a sensor array.

3. The intelligent charging pile management method with remote diagnostic function as described in claim 1, characterized in that, Voltage and current sensors are used to collect load data of the charging pile, while acceleration and acoustic emission sensors are used to collect operating status signals of the charging pile.

4. The intelligent charging pile management method with remote diagnostic function as described in claim 1, characterized in that, Performing one-dimensional variational mode decomposition on the load data to obtain stationary components and noise components specifically includes: The load data is preprocessed by normalization and noise suppression to obtain the processed load data; Multiple intrinsic mode functions are determined based on the number of decomposition levels and penalty factor of one-dimensional variational mode decomposition using the processed load data; By dividing the frequency of each intrinsic mode function, stationary components and noise components are obtained.

5. The intelligent charging pile management method with remote diagnostic function as described in claim 1, characterized in that, The multidimensional variational mode decomposition of the operating state signal yields a set of state sub-mode signals, specifically including: The operating status signal is simultaneously calibrated through multiple channels to obtain the calibrated operating status signal; Multiple multidimensional intrinsic mode functions are determined based on the number of decomposition levels and penalty factor of multidimensional variational mode decomposition using the calibrated operating status signal. Based on the frequency distribution and energy density, all multidimensional intrinsic mode functions are grouped into states to obtain a set of state submode signals.

6. The intelligent charging pile management method with remote diagnostic function as described in claim 1, characterized in that, The load identification features of the charging pile are obtained by associating and identifying the noise component based on the stationary component and the external environment data, specifically including: The stationary components and the external environment data are timestamped to obtain a time-series correlated dataset. Based on the correlation coefficient and mutual information entropy, correlation analysis is performed on the time-series associated dataset and the noise component, and then the load identification features of the charging pile are obtained from the analysis results.

7. The intelligent charging pile management method with remote diagnostic function as described in claim 1, characterized in that, The fault diagnosis features of the charging pile are obtained by performing joint analysis based on instantaneous frequency and spectral entropy on the set of state sub-mode signals. Specifically, these features include: Perform a Hilbert transform on the set of state submode signals to obtain the instantaneous frequency sequences corresponding to different submode signals; Based on the instantaneous frequency sequence, the frequency value and spectral entropy value of the corresponding sub-mode signal are determined, and then the frequency entropy feature pairs corresponding to different sub-mode signals are obtained; After normalizing all frequency entropy features, principal component analysis is performed to obtain the fault diagnosis features of the charging pile.

8. The intelligent charging pile management method with remote diagnostic function as described in claim 1, characterized in that, The modality fusion matrix is ​​obtained by performing feature fusion on the load identification features and the fault diagnosis features based on adaptive allocation of feature correlation weights. Specifically, this includes: The load identification features and the fault diagnosis features are synchronized in time before the correlation coefficient is calculated, thereby obtaining the feature correlation coefficient matrix; The adaptive weight allocation coefficients are determined based on the aforementioned feature correlation coefficient matrix; The load identification features and the fault diagnosis features are weighted and fused using the adaptive weight allocation coefficients to obtain a modal fusion matrix.

9. A smart charging pile management method with remote diagnostic function as described in claim 1, characterized in that, Remote scheduling of charging piles based on the fault diagnosis results specifically includes: The remote dispatch command is determined based on the fault diagnosis results and the load prediction curve of the charging pile. The remote scheduling command is sent to the charging pile, and the remote scheduling command is executed by the controller of the charging pile to obtain the remote scheduling result.

10. A smart charging pile management system with remote diagnostic function, used to execute a smart charging pile management method with remote diagnostic function as described in any one of claims 1 to 9, characterized in that, The management system includes: The data acquisition module is used to acquire external environmental data of the charging pile, as well as load data and operating status signals of the charging pile. The mode decomposition module is used to perform one-dimensional variational mode decomposition on the load data to obtain stationary components and noise components, and to perform multi-dimensional variational mode decomposition on the operating state signal to obtain a set of state sub-mode signals. The feature extraction module is used to correlate and identify the noise component based on the stationary component and the external environment data to obtain the load identification features of the charging pile, and to perform joint analysis processing based on instantaneous frequency and spectral entropy on the state sub-mode signal set to obtain the fault diagnosis features of the charging pile. The feature fusion and diagnosis module is used to perform feature fusion on the load identification features and the fault diagnosis features based on feature correlation weight adaptive allocation to obtain a modal fusion matrix. The modal fusion matrix is ​​then used for joint diagnostic analysis based on a bidirectional long short-term memory network to obtain the load prediction curve and fault diagnosis results of the charging pile. The remote scheduling module is used to remotely schedule the charging piles based on the fault diagnosis results when the load prediction curve reaches the preset dynamic early warning range.

Images