A method and device for monitoring the operational status and providing early warning of faults in urban charging piles.

Abstract

This invention provides a method and device for monitoring the operational status and providing early warning of faults in urban charging piles. The invention preprocesses the acquired raw monitoring data, extracts edge feature vectors and alarm flags of the charging piles from the preprocessed data, statistically analyzes the charging pile status identification information, and maps it to a digital twin model of the charging pile corresponding to the physical charging pile to generate a virtual simulation feature vector. The virtual simulation feature vector, edge feature vector, and historical fault sample data are fused into a fault prediction input matrix. A fault prediction model constructed using a CNN-BiLSTM hybrid network is used to predict the fault prediction input matrix, obtaining the latent fault prediction result. This invention employs a CNN-BiLSTM hybrid network combined with an attention mechanism to construct the fault prediction model, effectively improving the accuracy of fault type identification and the precision of remaining service life prediction; thus enhancing the detection and early warning of latent faults in charging piles.

Description

Technical Field

[0001] This invention relates to the field of new energy charging pile monitoring technology, and in particular to a method and device for monitoring the operation status and providing early warning of faults in urban charging piles. Background Technology

[0002] With the rapid iteration and popularization of the new energy vehicle industry, urban public AC charging piles and DC fast charging piles have become core components of new urban municipal infrastructure, smart energy networks, and new energy vehicle charging systems, and are widely deployed in various scenarios such as urban business districts, residential communities, road stations, and industrial parks. Charging piles operate in an outdoor environment with frequent load fluctuations, alternating temperature and humidity, and complex electromagnetic interference. They also experience high daily start-stop charging frequency and large dynamic changes in load conditions, making them highly susceptible to safety hazards such as electrical aging, overheating of lines, insulation degradation, leakage, water leakage, and fire caused by heat accumulation. If faults cannot be detected and intervened in a timely manner, they can easily lead to equipment burnout, electric shock injuries, fires, and large-scale charging paralysis, seriously affecting urban public safety and the normal charging order for new energy vehicles.

[0003] Current charging pile monitoring methods rely on basic online and offline voltage and current values ​​of the equipment for simple monitoring. This can only identify complete offline shutdown faults and cannot accurately identify potential fault hazards of the charging pile. Moreover, it does not consider the parameter drift characteristics caused by different charging rates, different load conditions, and seasonal changes in ambient temperature and humidity. This easily leads to false alarms under low load and missed alarms under high load. It has extremely poor ability to identify gradual and coupled faults, and the early warning triggering is severely delayed. Often, the fault is not discovered until it becomes apparent or even after safety signs appear, thus losing the best opportunity for early warning and pre-emptive action.

[0004] Therefore, there is an urgent need to provide a method and device for monitoring the operation status and providing early warning of faults of urban charging piles. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and device for monitoring the operational status and providing early warning of faults in urban charging piles.

[0006] In a first aspect, the present invention provides a method for monitoring the operational status and providing early warning of faults in urban charging piles, comprising:

[0007] Obtain raw monitoring data of electrical operating parameters, temperature parameters, environmental safety parameters, and port operating parameters of charging stations in the city;

[0008] The raw monitoring data is processed by time-series alignment, noise reduction filtering, and outlier removal to obtain preprocessed monitoring data.

[0009] According to the fixed slices, the slice mean, slice peak, slice valley, slice temperature difference, and slice change slope of each parameter are extracted from the preprocessed monitoring data and spliced ​​into an edge feature vector;

[0010] The preprocessed monitoring data is judged using the security threshold of the rule base, the alarm flag bit of each charging pile is obtained, and the status identification information of the charging pile is statistically analyzed.

[0011] The pre-processed monitoring data, edge feature vectors, charging pile status identification information, alarm flags, timestamps, unique charging pile identifiers, operator numbers, and site numbers are encapsulated into a standard data package for urban charging piles and transmitted to the cloud.

[0012] After parsing the standard data of urban charging piles received from the cloud, the data is mapped to the digital twin model of the charging pile corresponding to the physical charging pile. Virtual simulation feature vectors are then generated through the digital twin model of the charging pile.

[0013] The virtual simulation feature vector, edge feature vector, and historical fault sample data are fused to obtain the fault prediction input matrix;

[0014] A fault prediction model constructed using a CNN-BiLSTM hybrid network is used to predict the fault prediction input matrix, thereby obtaining the latent fault prediction results.

[0015] Based on the prediction results of latent faults, the early warning level is classified in combination with the real-time operating parameters of the charging piles.

[0016] Preferably, the original monitoring data undergoes time-series alignment, noise reduction filtering, and outlier removal to obtain preprocessed monitoring data, including:

[0017] Using the local master clock of the edge terminal as a unified benchmark, a standard timestamp is added to the raw monitoring data sampled by each sensor, and associated with the unique identifier of the charging pile, the operator to which it belongs, and the site number to which it belongs;

[0018] Linear interpolation is used to correct time-displaced data, bringing the sampling data from all sensing units to a fixed sampling period. This forms a time-aligned sequence of original monitoring data.

[0019] A joint denoising algorithm combining sliding median filtering and Kalman filtering is used to perform median filtering and denoising on the time-aligned original monitoring data sequence, resulting in filtered and denoised monitoring data.

[0020] The monitoring data is filtered and denoised to remove outliers, resulting in preprocessed monitoring data.

[0021] Preferably, the standard data of urban charging piles received from the cloud is parsed and mapped to a digital twin model of the charging pile corresponding to the physical charging pile, and virtual simulation features are generated through the digital twin model of the charging pile; including:

[0022] The standard data packets of urban charging piles received from the cloud are parsed to extract preprocessed monitoring data, edge feature vectors, timestamps and unique identifiers of charging piles, and to complete data verification and classification caching.

[0023] The digital twin model of the charging pile is created using SolidWorks software at a 1:1 scale based on the structure of the physical charging pile.

[0024] Establish a one-to-one mapping relationship between the sensor points of physical charging piles and the parameter nodes of the digital twin model of charging piles, and push standard data synchronously to the corresponding parameter nodes according to timestamps;

[0025] Using preprocessed monitoring data as boundary conditions, the digital twin model of the charging pile is driven to perform multiphysics numerical simulation calculations to obtain virtual simulation data;

[0026] The virtual simulation data is processed by extracting the virtual slice mean, virtual slice peak value, virtual slice valley value, virtual slice temperature difference, and virtual change slope from fixed slices, and then splicing them together to obtain the virtual simulation feature vector.

[0027] Preferably, the virtual simulation feature vector, edge feature vector, and historical fault sample data are fused to obtain the fault prediction input matrix, including:

[0028] Mini-maximum normalization was performed on the virtual simulation feature vector, edge feature vector, and historical fault sample data, respectively.

[0029] The normalized virtual simulation feature vector, edge feature vector, and historical fault sample data are concatenated according to feature dimensions to obtain the fused feature vector;

[0030] The fused feature vectors are arranged sequentially according to time slices to form the fault prediction input matrix.

[0031] As a preferred embodiment, the fault prediction model constructed by the CNN-BiLSTM hybrid network includes a local feature extraction module, a temporal feature extraction module, a feature fusion module, and a classifier;

[0032] The local feature extraction module includes convolutional layers and max pooling layers, which extract local features from the fault prediction input matrix and perform dimensionality reduction.

[0033] The temporal feature extraction module uses a bidirectional long short-term memory network to extract bidirectional hidden features from the dimensionality-reduced local features;

[0034] The feature fusion module uses an attention mechanism to weight the bidirectional hidden features to obtain the global fused features;

[0035] The classifier uses a fully connected layer and a Softmax function to predict the global fusion features, outputs the probability of occurrence of various types of faults, and outputs the remaining useful life (RUL) through a regression layer.

[0036] Preferably, a fault prediction model constructed using a CNN-BiLSTM hybrid network is used to predict the fault prediction input matrix. The prediction process yields latent fault prediction results, including fault type, probability of occurrence of the corresponding fault type, and remaining service life (RUL) of the charging pile under its current state; including:

[0037] Input the fault prediction matrix The convolutional layer of the input local feature extraction module extracts local features, and the local features are downsampled by the max pooling layer to obtain the dimensionality-reduced local features.

[0038] The local features after dimensionality reduction by the max pooling layer are input into the temporal feature extraction module, and bidirectional long short-term memory network is used to extract bidirectional hidden features from the local features.

[0039] The bidirectional hidden features are input into the feature fusion module, which uses an attention mechanism to weight the bidirectional hidden features to obtain the global fused features.

[0040] The global fusion features are input into the classifier, and the global fusion features are nonlinearly transformed and dimension-mapped through a fully connected layer, mapping the global fusion features to the output dimension corresponding to the number of fault types.

[0041] The output of the fully connected layer is normalized by using the Softmax activation function, and the output value is mapped to the [0,1] interval to obtain the occurrence probability of each fault type.

[0042] The classifier's regression layer performs a linear and nonlinear combination mapping on the global fusion features, outputting the remaining lifespan of the charging pile in its current state.

[0043] The output layer outputs the fault category, the probability of occurrence corresponding to the fault type, and the remaining service life of the charging pile under the current state, to obtain the latent fault prediction result.

[0044] As a preferred approach, the early warning level is classified based on the latent fault prediction results and the real-time operating parameters of the charging pile, including:

[0045] Three levels of probability thresholds and three levels of preset lifespan are set, wherein the first probability threshold is less than the second probability threshold, the second probability threshold is less than the third probability threshold, the first preset lifespan is greater than the second preset lifespan, and the second preset lifespan is greater than the third preset lifespan.

[0046] The warning level is set to Level 4 based on the three-level probability threshold and the three-level preset lifespan. The judgment is made by combining the predicted probability of the fault type, the remaining lifespan of the charging pile under its current state, and real-time alarm flags. Among these factors:

[0047] Level 1 warning: The predicted probability of the fault type is less than the first probability threshold, and the remaining service life (RUL) of the charging pile in its current state is greater than the first preset service life, and there is no alarm signal.

[0048] Level 2 warning: The predicted probability of the fault category is greater than or equal to the first probability threshold and less than the second probability threshold, and the remaining service life RUL of the charging pile in the current state is greater than the second preset service life and less than or equal to the first probability threshold, and there is no alarm signal.

[0049] Level 3 warning: The predicted probability of the fault type is greater than or equal to the second probability threshold and less than the third probability threshold, and the remaining service life (RUL) of the charging pile in the current state is greater than the third preset service life and less than or equal to the second preset service life, and the alarm signs are high temperature alarm sign, water immersion alarm sign, and leakage alarm sign.

[0050] Level 4 warning: The predicted probability of the fault type is greater than or equal to the third probability threshold, and the remaining service life (RUL) of the charging pile in its current state is less than or equal to the third preset service life, and the alarm sign is a smoke alarm sign.

[0051] Secondly, the present invention provides a device for monitoring the operation status and providing early warning of faults in urban charging piles, comprising:

[0052] The data acquisition module is used to acquire raw monitoring data of electrical operating parameters, temperature parameters, environmental safety parameters, and port operating parameters of charging stations in the city.

[0053] The preprocessing module is used to perform time-series alignment, noise reduction filtering, and outlier removal on the raw monitoring data to obtain preprocessed monitoring data.

[0054] The edge feature extraction module is used to extract the mean, peak, valley, temperature difference, and slope of each parameter from the preprocessed monitoring data according to fixed slices and concatenate them into an edge feature vector.

[0055] The alarm flag acquisition module is used to judge the pre-processed monitoring data using the security threshold of the rule base, acquire the alarm flag bit of each charging pile, and count the status identification information of the charging pile.

[0056] The data transmission module is used to encapsulate the pre-processed monitoring data, edge feature vectors, charging pile status identification information, alarm flags, timestamps, unique charging pile identifiers, operator numbers, and site numbers into a standard data packet for urban charging piles and transmit it to the cloud.

[0057] The simulation module is used to parse the standard data of urban charging piles received from the cloud and map it to the digital twin model of the charging pile corresponding to the physical charging pile, and generate virtual simulation feature vectors through the digital twin model of the charging pile.

[0058] The feature fusion module is used to fuse virtual simulation feature vectors, edge feature vectors, and historical fault sample data to obtain a fault prediction input matrix.

[0059] The fault prediction module is used to predict the fault prediction input matrix using a fault prediction model built with a CNN-BiLSTM hybrid network, and obtain the latent fault prediction result.

[0060] The early warning level classification module is used to classify early warning levels based on the prediction results of latent faults and the real-time operating parameters of charging piles.

[0061] Thirdly, the present invention provides an electronic device, comprising:

[0062] At least one processor;

[0063] and memory that is communicatively connected to at least one processor;

[0064] The memory stores a computer program that can be executed by at least one processor. When the computer program is executed by at least one processor, it implements a method for monitoring the operation status and providing early warning of faults in urban charging piles.

[0065] Fourthly, the present invention provides a computer storage medium storing a computer program, which, when executed by a processor, implements a method for monitoring the operating status and providing early warning of faults in urban charging piles.

[0066] The beneficial technical effects of this invention are as follows:

[0067] 1. This invention uses a CNN-BiLSTM hybrid network combined with an attention mechanism to construct a fault prediction model. It extracts local features of electrical, temperature, and environmental parameters through convolutional layers, and mines temporal evolution patterns through a bidirectional long short-term memory network. The attention mechanism is used to strengthen the feature weights of key fault periods, which effectively improves the accuracy of fault type identification and the accuracy of remaining service life prediction.

[0068] 2. This invention integrates measured multi-operational data of edge terminal charging piles with simulation data of digital twin models, and introduces historical fault sample data for multi-feature fusion modeling, thereby significantly improving the completeness and accuracy of latent fault feature mining of charging piles.

[0069] 3. This invention completes data timing alignment, noise reduction filtering, and outlier removal at the edge terminal side, and obtains high temperature, water immersion, leakage current, and smoke alarm flags based on local thresholds, realizing edge-end preprocessing and local initial judgment.

[0070] 4. This invention sets up four warning levels to achieve gradient identification from minor hidden dangers to emergency faults;

[0071] 5. The digital twin model of the present invention uses measured data as boundary conditions to carry out multi-physics numerical simulation and generate virtual simulation features, which makes up for the data gaps in which physical sensors cannot be deployed. Attached Figure Description

[0072] Figure 1 This is a flowchart illustrating Embodiment 1 of the method of the present invention;

[0073] Figure 2 This is a structural framework diagram of the CNN-BiLSTM hybrid network in Embodiment 1 of the present invention;

[0074] Figure 3 This is a structural framework diagram of the device in Embodiment 2 of the present invention;

[0075] In the diagram, 100 is the data acquisition module; 200 is the preprocessing module; 300 is the edge feature extraction module; 400 is the alarm flag acquisition module; 500 is the data transmission module; 600 is the simulation module; 700 is the feature fusion module; 800 is the fault prediction module; and 900 is the early warning level classification module. Detailed Implementation

[0076] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings:

[0077] like Figure 1 As shown in the figure, this embodiment provides a method for monitoring the operation status and providing fault early warning of urban charging piles, including:

[0078] S1. Obtain raw monitoring data of electrical operating parameters, temperature parameters, environmental safety parameters, and port operating parameters of charging stations in the city.

[0079] S2. Perform time-series alignment, noise reduction filtering, and outlier removal on the raw monitoring data to obtain preprocessed monitoring data.

[0080] S3. Extract the mean, peak, valley, temperature difference, and slope of each parameter from the preprocessed monitoring data according to fixed slices, and concatenate them into an edge feature vector. ;

[0081] S4. Use the security threshold of the rule base to judge the preprocessed monitoring data, obtain the alarm flag bit of each charging pile, and count the status identification information of the charging pile.

[0082] S5. Process the preprocessed monitoring data and edge feature vectors. The charging pile status identification information, alarm flag, timestamp, unique identifier of the charging pile, operator number, and site number are encapsulated into a standard data packet for urban charging piles, and the standard data packet for urban charging piles is transmitted to the cloud through a dual primary and backup communication link.

[0083] S6. After parsing the standard data of urban charging piles received from the cloud, map it to the digital twin model of the charging pile corresponding to the physical charging pile, and generate virtual simulation feature vectors through the digital twin model of the charging pile. ;

[0084] S7. Transfer the virtual simulation feature vector Edge feature vectors Historical fault sample data are fused to obtain the fault prediction input matrix Y;

[0085] S8. A fault prediction model constructed using a CNN-BiLSTM hybrid network is used to predict the fault prediction input matrix. Prediction is performed to obtain latent fault prediction results, including fault type, probability of occurrence of the fault type, and remaining service life (RUL) of the charging pile under the current state.

[0086] S9. Based on the prediction results of latent faults, the early warning level is classified in combination with the real-time operating parameters of the charging pile.

[0087] In this embodiment, step S1 involves acquiring raw monitoring data of the electrical operating parameters, temperature parameters, environmental safety parameters, and port operating condition parameters of each charging station in the city, including:

[0088] Distributed sensing units are deployed inside and outside the charging piles, charging guns, power modules, busbars, and cabinets at various charging stations in the city.

[0089] The electrical operating parameters, temperature parameters, environmental safety parameters, and port operating status parameters of the charging pile are collected in parallel at a set high-frequency sampling period using distributed sensing units.

[0090] In this embodiment, the electrical operating parameters include input / output voltage, current, power, bus ripple, insulation resistance, and residual leakage current.

[0091] In this embodiment, the temperature parameters include the power module temperature, the nozzle contact point temperature, the busbar temperature, and the ambient temperature inside the cabinet.

[0092] In this embodiment, environmental safety parameters include humidity inside and outside the cabinet, water accumulation / water immersion sensor signals, and smoke sensor signals;

[0093] In this embodiment, the port operating parameters include the working status of each charging port, the gun position locking status, the relay operation status, and the charging order status.

[0094] In this embodiment, step S2 involves performing time-series alignment, noise reduction filtering, and outlier removal on the original monitoring data to obtain preprocessed monitoring data, including:

[0095] S21. Using the local master clock of the edge terminal as a unified reference, add a standard timestamp to the raw monitoring data sampled by each sensor. Associate the unique identifier of the charging pile, its operator, and its site number to complete the data structuring;

[0096] Linear interpolation is used to correct time-displaced data, bringing the sampling data from all sensing units to a fixed sampling period. This forms a time-aligned sequence of original monitoring data. ;

[0097] ;

[0098] In the formula, This is the corrected monitoring data; , These are monitoring data from two adjacent sampling points; These are the timestamps of two adjacent sampling points;

[0099] ;

[0100] In the formula, for The raw monitoring data at any given moment.

[0101] S22. A joint noise reduction algorithm combining sliding median filtering and Kalman filtering is used to process the time-aligned original monitoring data sequence. Median filtering and noise reduction are performed to obtain the filtered and denoised monitoring data. The details are as follows:

[0102] Time-aligned raw monitoring data sequences Iterate through the sliding window point by point, for the... Original monitoring data The original monitoring data within the sliding window is sorted, and the median is taken as the first value. Original monitoring data Median filtered data; traversing all original monitoring data sequences Median filter monitoring data was then obtained. ;

[0103] Based on median filter monitoring data The Kalman filter algorithm was used for noise reduction to obtain the filtered and denoised monitoring data. ;Right now:

[0104] ;

[0105] ;

[0106] ;

[0107] ;

[0108] ;

[0109] ;

[0110] ;

[0111] In the formula, Monitoring data after Kalman denoising; Kalman gain; for Median filtered monitoring data at any given time; The prior error covariance matrix; The process noise variance is corrected by residuals; To measure the noise variance; For residuals; , The residual coefficient is used in this embodiment. , Set them to 0.01 and 0.05 respectively; The observation matrix; Here is the state transition matrix; in this embodiment, A=1, H=1; This represents the predicted value of the prior state.

[0112] S23, Monitoring data based on filtering and noise reduction Outlier removal was performed to obtain preprocessed monitoring data. ;

[0113] Calculate the mean of the data within a sliding window. and standard deviation and based on the mean and standard deviation Set the rejection threshold to ;

[0114] If the current noise-reduced monitoring data Not at the rejection threshold If it falls within the range, it is considered a suspected outlier;

[0115] Calculate the current noise-reduced monitoring data for suspected outliers The slope of the previous three normal monitoring data after noise reduction If the current noise-reduced monitoring data is suspected to contain outliers The absolute value of the slope Greater than the maximum allowable slope change Then determine the current noise-reduced monitoring data. If the value is a true outlier, it will be removed; otherwise, it will be considered a normal fluctuation.

[0116] In this embodiment, in step S3, the mean value, peak value, valley value, temperature difference, and slope of each parameter are extracted from the preprocessed monitoring data according to fixed slices and then concatenated into an edge feature vector. The details are as follows:

[0117] The slice mean is expressed as:

[0118] ;

[0119] In the formula, The average value of the slice; This represents the number of sampling points within the slice; For the first Monitoring data after removing outliers at any given time;

[0120] The peak and valley values ​​of a slice are represented as follows:

[0121] ;

[0122] ;

[0123] In the formula, , These represent the peak value and valley value of the slice, respectively.

[0124] The temperature difference of the slices is expressed as:

[0125] ;

[0126] In the formula, Temperature difference between slices; , These are the highest and lowest temperatures within the slice;

[0127] Slice change slope Represented as:

[0128] ;

[0129] The slice mean, slice peak, slice trough, slice temperature difference, and slice slope are concatenated to form an edge feature vector. .

[0130] In this embodiment, in step S4, the preprocessed monitoring data is judged using the security threshold of the rule base, the alarm flag bit of each charging pile is obtained, and the status identification information of the charging pile is statistically analyzed, as follows:

[0131] When any of the following values—power module temperature, nozzle contact point temperature, busbar temperature, or ambient temperature inside the cabinet—exceeds the preset high temperature threshold, the high temperature alarm flag is set to 1; otherwise, it is set to 0.

[0132] When the water level or switch signal collected by the water immersion sensor at the bottom of the cabinet exceeds the preset water immersion threshold, the water immersion alarm flag is set to 1; otherwise, it is set to 0.

[0133] When either the residual leakage current or the insulation resistance exceeds the leakage current threshold or falls below the lower insulation limit threshold, the leakage current alarm flag is set to 1; otherwise, it is set to 0.

[0134] The smoke alarm flag is set to 1 when the smoke concentration collected by the smoke sensor exceeds the preset smoke threshold; otherwise, it is set to 0.

[0135] In this embodiment, when the charging pile is online, the status identifier of the charging pile is 1; when the charging pile is offline, the status identifier of the charging pile is 0.

[0136] In this embodiment, in step S5, the primary and backup dual communication links include a primary communication link and a backup communication link. When the primary communication link fails, the backup communication link is switched.

[0137] In this embodiment, step S6 involves parsing the standard data of urban charging piles received from the cloud and mapping it to a digital twin model of the charging pile corresponding to the physical charging pile, and generating virtual simulation features through the digital twin model of the charging pile; including:

[0138] S61. Parse the standard data packets of urban charging piles received from the cloud, and extract preprocessed monitoring data and edge feature vectors. The data is verified and categorized using timestamps and unique identifiers for charging piles.

[0139] S62. The digital twin model of the charging pile is created using SolidWorks software at a 1:1 3D scale based on the structure of the physical charging pile, constructing a digital twin that includes a 3D geometric model, an electrical topology model, a thermal coupling model, and multiphysics coupling equations; among which...

[0140] The 3D geometric model is a 1:1 solid structure model built using SolidWorks, including components such as cabinets, power modules, busbars, charging guns, cables, relays, and heat dissipation ducts.

[0141] The electrical topology model includes an equivalent circuit model that includes AC input, rectification, DC / DC conversion, output circuit, insulation detection, and leakage protection.

[0142] The thermal field coupling model is a temperature field model constructed based on the equations of heat conduction, heat convection, and heat radiation.

[0143] Multiphysics coupling equations are used for joint simulation of electrical, thermal, mechanical, and environmental systems.

[0144] S63. Establish a one-to-one mapping relationship between the sensor points of the physical charging pile and the parameter nodes of the digital twin model of the charging pile, clarify the parameter mapping path and update frequency, and push the standard data to the corresponding parameter nodes synchronously according to the timestamp.

[0145] S64. Using preprocessed monitoring data as boundary conditions, drive the digital twin model of the charging pile to perform multi-physics numerical simulation calculations to obtain virtual simulation data.

[0146] Virtual simulation data includes virtual electrical simulation data, virtual thermal field simulation data, virtual environmental safety simulation data, and virtual port operating condition simulation data;

[0147] Virtual electrical simulation data includes virtual input / output voltage, current, virtual bus voltage, virtual insulation resistance, virtual residual leakage current, and virtual power;

[0148] The virtual thermal simulation data includes virtual power module temperature, virtual nozzle contact temperature, virtual busbar temperature, virtual cabinet temperature, and virtual temperature rise gradient.

[0149] Virtual environment safety simulation data includes virtual temperature and humidity, virtual water immersion status, and virtual smoke concentration;

[0150] The virtual port operating condition simulation data includes the virtual port status, virtual latching status, and virtual relay status.

[0151] The thermal field simulation follows the heat conduction equation:

[0152] ;

[0153] In the formula, The thermal conductivity of the material; For node temperature; Volumetric heating power; The density of the material; Specific heat capacity; For time; , , They represent respectively to Find the partial derivative along the axis; , , Temperature along Gradient of direction;

[0154] Virtual insulation resistance and virtual leakage current are calculated using the following formulas:

[0155] , ;

[0156] In the formula, The virtual insulation resistance is the insulation resistance of the charging pile busbar to ground, simulated by the digital twin model. The virtual leakage current is the total system leakage current output from the digital twin model simulation. The DC bus voltage is the simulated output of the digital twin model, representing the virtual bus voltage. The virtual A-phase leakage current is the leakage current to ground of the A-phase winding or line output by the digital twin simulation. For virtual B-phase leakage current, the digital twin simulation outputs the B-phase winding or line leakage current to ground. This refers to the virtual C-phase leakage current, which is the C-phase winding or line leakage current to ground output by the digital twin simulation.

[0157] Virtual power The calculation formula is: ;

[0158] In the formula, Virtual input / output voltage and current respectively;

[0159] The virtual temperature gradient is represented as:

[0160] ;

[0161] In the formula, Indicates the virtual simulation temperature; , These represent the virtual temperatures corresponding to the two time slices; These represent the times of two adjacent time slices.

[0162] S65. Extract the virtual slice mean from the virtual simulation data according to fixed slices. Virtual slice peak Virtual slice valley value Virtual slice temperature difference Virtual slope of change The virtual simulation feature vector is obtained by splicing the vectors together. ;

[0163] .

[0164] The virtual slice mean is represented as:

[0165] ;

[0166] In the formula, This represents the number of sampling points within the slice; For the first Virtual data of time;

[0167] The virtual slice peak and virtual slice valley are represented as follows:

[0168] ;

[0169] ;

[0170] In the formula, , These are the virtual slice peak value and the virtual slice valley value, respectively.

[0171] The virtual slice temperature difference is represented as:

[0172] ;

[0173] In the formula, For virtual slice temperature difference; , These are the highest and lowest virtual temperatures within the slice;

[0174] Virtual slice change slope Represented as:

[0175] .

[0176] In this embodiment, in step S7, the virtual simulation feature vector is... Edge feature vectors Historical fault sample data are fused to obtain the fault prediction input matrix. ;include:

[0177] S71, Regarding virtual simulation feature vectors Edge feature vectors Historical fault sample data Perform min-max normalization separately to unify the data units to the [0,1] interval;

[0178] S72, Normalizing the virtual simulation feature vector Edge feature vectors Historical fault sample data Concatenate the features along their respective dimensions to obtain the fused feature vector. ,Right now:

[0179] ;

[0180] S73, Fusion Feature Vectors Sliced ​​by time Arranged sequentially, they form the fault prediction input matrix. ,Right now:

[0181] ;

[0182] In the formula, N is the total number of time slices; For the first The fused feature vectors corresponding to each time slice.

[0183] In this embodiment, in step S8, the fault prediction model constructed using the CNN-BiLSTM hybrid network is used to predict the fault prediction input matrix. The prediction process yields latent fault prediction results, including fault type, probability of occurrence of the corresponding fault type, and remaining service life (RUL) of the charging pile under its current state; including:

[0184] S81. Construct a fault prediction model using a CNN-BiLSTM hybrid network, such as Figure 2 As shown, the fault prediction model constructed by the CNN-BiLSTM hybrid network includes a local feature extraction module, a temporal feature extraction module, a feature fusion module, and a classifier.

[0185] The local feature extraction module includes convolutional layers and max-pooling layers, which extract features from the fault prediction input matrix. Extract local features from;

[0186] The temporal feature extraction module uses a bidirectional long short-term memory network to extract bidirectional hidden features from local features;

[0187] The feature fusion module uses an attention mechanism to weight the bidirectional hidden features to obtain the global fused features;

[0188] The classifier uses a fully connected layer and a Softmax function to predict the global fusion features, outputs the probability of occurrence of various types of faults, and outputs the remaining useful life (RUL) through a regression layer.

[0189] S82, Input fault prediction into the matrix The convolutional layer of the local feature extraction module extracts local features, and then a max pooling layer downsamples these local features to obtain the dimensionality-reduced local features, i.e.:

[0190] ;

[0191] ;

[0192] In the formula, These are local features extracted by the convolutional layer. These are local features after dimensionality reduction of the max pooling layer; For activation functions; This is a convolution operation; , These are the weight matrix and bias term of the convolutional layer, respectively; This is a max pooling operation;

[0193] S83, Local features after dimensionality reduction of the max pooling layer The input temporal feature extraction module uses a bidirectional long short-term memory network to extract bidirectional hidden features from local features, namely:

[0194] ;

[0195] ;

[0196] ;

[0197] In the formula, for Positive hidden features at any given moment; for Inverse hidden features at each moment; This indicates vector concatenation; It is a two-way hidden feature; for Local features after dimensionality reduction of the max pooling layer at time 1;

[0198] S84, Bidirectional Hidden Features The input feature fusion module uses an attention mechanism to weight the bidirectional hidden features to obtain the global fused features;

[0199] Based on bidirectional hidden features Calculate the attention score for each time-step feature ;Right now:

[0200] ;

[0201] In the formula, It is a value matrix; Indicates the transpose operation; It is the hyperbolic tangent function; , These are the weight matrix and bias term of the attention mechanism, respectively;

[0202] Attention score at all times Softmax normalization is performed to obtain the attention weights. ;Right now:

[0203] ;

[0204] Based on attention weights For bidirectional hidden features We perform a weighted summation to obtain the global fusion feature. ;

[0205] ;

[0206] In the formula, This represents the total number of time-series slices.

[0207] S85, Global Fusion Features Input to the classifier, then globally fuse features through a fully connected layer. Perform nonlinear transformations and dimension mappings to fuse global features. The output is mapped to the output dimension corresponding to the number of fault types; then, the output of the fully connected layer is normalized using the Softmax activation function, mapping the output value to the [0,1] interval to obtain the occurrence probability of each fault type; that is:

[0208] ;

[0209] In the formula, The predicted probability corresponding to fault type k; Number of fault categories; , These are the weight matrix and bias term of the fully connected layer, respectively.

[0210] S86. Global fusion features through the regression layer of the classifier. Perform a combined linear and nonlinear mapping to output the remaining service life (RUL) of the charging pile in its current state, i.e.:

[0211] ;

[0212] In the formula, , These are the weight matrix and bias term of the regression layer, respectively.

[0213] S87. Output the fault category, the probability of occurrence corresponding to the fault type, and the remaining service life (RUL) of the charging pile under the current state through the output layer to obtain the latent fault prediction result. ,Right now:

[0214] ;

[0215] In the formula, Fault category; The probability of the fault category; The remaining service life (RUL) of the charging station under its current condition.

[0216] In this embodiment, step S9 involves classifying the early warning level based on the latent fault prediction results and the real-time operating parameters of the charging pile, including:

[0217] Three levels of probability thresholds and three levels of preset lifespan are set, wherein the first probability threshold is less than the second probability threshold, the second probability threshold is less than the third probability threshold, the first preset lifespan is greater than the second preset lifespan, and the second preset lifespan is greater than the third preset lifespan.

[0218] In this embodiment, the first probability threshold, the second probability threshold, and the third probability threshold are 0.3, 0.6, and 0.9, respectively.

[0219] In this embodiment, the first preset lifespan, the second preset lifespan, and the third preset lifespan are 72h, 24h, and 1h, respectively.

[0220] In this embodiment, the warning level is set based on a three-level probability threshold and a three-level preset lifespan. The warning level is four, determined by combining the predicted probability of the fault type, the remaining service life (RUL) of the charging pile under its current condition, and real-time alarm flags; that is: ;

[0221] In the formula, the alarm sign This indicates that there is no alarm.

[0222] The Level 4 early warning information will be displayed in real time on the monitoring terminal.

[0223] like Figure 3 As shown, an embodiment of this application provides a device for monitoring the operation status and providing early warning of faults in urban charging piles, comprising:

[0224] The data acquisition module 100 is used to acquire raw monitoring data of electrical operating parameters, temperature parameters, environmental safety parameters, and port operating parameters of charging stations in the city.

[0225] The preprocessing module 200 is used to perform time-series alignment, noise reduction filtering, and outlier removal on the raw monitoring data to obtain preprocessed monitoring data.

[0226] The edge feature extraction module 300 is used to extract the mean value, peak value, valley value, temperature difference, and slope of each parameter from the preprocessed monitoring data according to fixed slices and splice them into an edge feature vector.

[0227] The alarm flag acquisition module 400 is used to judge the pre-processed monitoring data using the security threshold of the rule base, acquire the alarm flag bit of each charging pile, and count the status identification information of the charging pile.

[0228] The data transmission module 500 is used to encapsulate the pre-processed monitoring data, edge feature vectors, charging pile status identification information, alarm flags, timestamps, unique charging pile identifiers, operator numbers, and site numbers into a standard data packet for urban charging piles and transmit it to the cloud.

[0229] The simulation module 600 is used to parse the standard data of urban charging piles received from the cloud and map it to the digital twin model of the charging pile corresponding to the physical charging pile, and generate virtual simulation feature vectors through the digital twin model of the charging pile.

[0230] The feature fusion module 700 is used to fuse virtual simulation feature vectors, edge feature vectors, and historical fault sample data to obtain a fault prediction input matrix.

[0231] The fault prediction module 800 is used to predict the fault prediction input matrix using a fault prediction model constructed with a CNN-BiLSTM hybrid network, and obtain the latent fault prediction result.

[0232] The early warning level classification module 900 is used to classify early warning levels based on the prediction results of latent faults and the real-time operating parameters of charging piles.

[0233] In this embodiment, the preprocessing module 200 performs time-series alignment, noise reduction filtering, and outlier removal on the original monitoring data to obtain preprocessed monitoring data, including:

[0234] Using the local master clock of the edge terminal as a unified reference, a standard timestamp is added to the raw monitoring data sampled by each sensor. Associate the unique identifier of the charging pile, its operator, and its site number to complete the data structuring;

[0235] Linear interpolation is used to correct time-displaced data, bringing the sampling data from all sensing units to a fixed sampling period. This forms a time-aligned sequence of original monitoring data. ;

[0236] ;

[0237] In the formula, This is the corrected monitoring data; , These are monitoring data from two adjacent sampling points; These are the timestamps of two adjacent sampling points;

[0238] ;

[0239] In the formula, for The raw monitoring data at any given moment.

[0240] A joint noise reduction algorithm combining moving median filtering and Kalman filtering is used to denoise the time-aligned original monitoring data sequences. Median filtering and noise reduction are performed to obtain the filtered and denoised monitoring data. The details are as follows:

[0241] Time-aligned raw monitoring data sequences Iterate through the sliding window point by point, for the... Original monitoring data The original monitoring data within the sliding window is sorted, and the median is taken as the first value. Original monitoring data Median filtered data; traversing all original monitoring data sequences Median filter monitoring data was then obtained. ;

[0242] Based on median filter monitoring data The Kalman filter algorithm was used for noise reduction to obtain the filtered and denoised monitoring data. ;Right now:

[0243] ;

[0244] ;

[0245] ;

[0246] ;

[0247] ;

[0248] ;

[0249] ;

[0250] In the formula, Monitoring data after Kalman denoising; Kalman gain; for Median filtered monitoring data at any given time; The prior error covariance matrix; The process noise variance is corrected by residuals; To measure the noise variance; For residuals; , The residual coefficient is used in this embodiment. , Set them to 0.01 and 0.05 respectively; The observation matrix; Here is the state transition matrix; in this embodiment, A=1, H=1; This represents the predicted value of the prior state.

[0251] Monitoring data based on filtering and noise reduction Outlier removal was performed to obtain preprocessed monitoring data. ;

[0252] Calculate the mean of the data within a sliding window. and standard deviation and based on the mean and standard deviation Set the rejection threshold to ;

[0253] If the current noise-reduced monitoring data Not at the rejection threshold If it falls within the range, it is considered a suspected outlier;

[0254] Calculate the current noise-reduced monitoring data for suspected outliers The slope of the previous three normal monitoring data after noise reduction If the current noise-reduced monitoring data is suspected to contain outliers The absolute value of the slope Greater than the maximum allowable slope change Then determine the current noise-reduced monitoring data. If the value is a true outlier, it will be removed; otherwise, it will be considered a normal fluctuation.

[0255] In this embodiment, the average value of the slices extracted by the edge feature extraction module 300 is expressed as:

[0256] ;

[0257] In the formula, The average value of the slice; This represents the number of sampling points within the slice; For the first Monitoring data after removing outliers at any given time.

[0258] In this embodiment, the slice peak value and slice valley value extracted by the edge feature extraction module 300 are respectively represented as:

[0259] ;

[0260] ;

[0261] In the formula, , These represent the peak value and valley value of the slice, respectively.

[0262] In this embodiment, the slice temperature difference extracted by the edge feature extraction module 300 is expressed as:

[0263] ;

[0264] In the formula, Temperature difference between slices; , These represent the highest and lowest temperatures within the slice.

[0265] In this embodiment, the slope of the slice change extracted by the edge feature extraction module 300 is expressed as:

[0266] .

[0267] In this embodiment, the edge feature extraction module 300 concatenates the slice mean, slice peak value, slice valley value, slice temperature difference, and slice slope into an edge feature vector. .

[0268] In this embodiment, the alarm flag acquisition module 400 uses the security threshold of the rule base to determine the preprocessed monitoring data, acquires the alarm flag bit of each charging pile, and counts the status identification information of the charging pile, as follows:

[0269] When any of the following values—power module temperature, nozzle contact point temperature, busbar temperature, or ambient temperature inside the cabinet—exceeds the preset high temperature threshold, the high temperature alarm flag is set to 1; otherwise, it is set to 0.

[0270] When the water level or switch signal collected by the water immersion sensor at the bottom of the cabinet exceeds the preset water immersion threshold, the water immersion alarm flag is set to 1; otherwise, it is set to 0.

[0271] When either the residual leakage current or the insulation resistance exceeds the leakage current threshold or falls below the lower insulation limit threshold, the leakage current alarm flag is set to 1; otherwise, it is set to 0.

[0272] The smoke alarm flag is set to 1 when the smoke concentration collected by the smoke sensor exceeds the preset smoke threshold; otherwise, it is set to 0.

[0273] In this embodiment, when the charging pile is online, the status identifier of the charging pile is 1; when the charging pile is offline, the status identifier of the charging pile is 0.

[0274] In this embodiment, the simulation module 600 parses the standard data of urban charging piles received from the cloud and maps it to a digital twin model of the charging pile corresponding to the physical charging pile, and generates virtual simulation features through the digital twin model of the charging pile; including:

[0275] The standard data packets of urban charging piles received from the cloud are parsed to extract preprocessed monitoring data and edge feature vectors. The system uses timestamps and unique identifiers for charging piles to complete data verification and categorized caching.

[0276] The digital twin model of the charging pile is created using SolidWorks software at a 1:1 3D scale, based on the structure of the physical charging pile. This constructs a digital twin containing a 3D geometric model, an electrical topology model, a thermal coupling model, and multiphysics coupling equations.

[0277] The 3D geometric model is a 1:1 solid structure model built using SolidWorks, including components such as cabinets, power modules, busbars, charging guns, cables, relays, and heat dissipation ducts.

[0278] The electrical topology model includes an equivalent circuit model that incorporates AC input, rectification, DC / DC conversion, output circuit, insulation detection, and leakage protection.

[0279] The thermal field coupling model is a temperature field model constructed based on the equations of heat conduction, heat convection, and heat radiation.

[0280] Multiphysics coupling equations are used for joint simulation of electrical, thermal, mechanical, and environmental systems.

[0281] Establish a one-to-one mapping relationship between the sensor locations of physical charging piles and the parameter nodes of the digital twin model of charging piles, clarify the parameter mapping path and update frequency, and push standard data synchronously to the corresponding parameter nodes according to timestamps.

[0282] Using preprocessed monitoring data as boundary conditions, the digital twin model of the charging pile is driven to perform multiphysics numerical simulation calculations to obtain virtual simulation data;

[0283] Virtual simulation data includes virtual electrical simulation data, virtual thermal field simulation data, virtual environmental safety simulation data, and virtual port operating condition simulation data;

[0284] Virtual electrical simulation data includes virtual input / output voltage, current, virtual bus voltage, virtual insulation resistance, virtual residual leakage current, and virtual power;

[0285] The virtual thermal simulation data includes virtual power module temperature, virtual nozzle contact temperature, virtual busbar temperature, virtual cabinet temperature, and virtual temperature rise gradient.

[0286] Virtual environment safety simulation data includes virtual temperature and humidity, virtual water immersion status, and virtual smoke concentration;

[0287] The virtual port operating condition simulation data includes the virtual port status, virtual latching status, and virtual relay status.

[0288] The thermal field simulation follows the heat conduction equation:

[0289] ;

[0290] In the formula, The thermal conductivity of the material; For node temperature; Volumetric heating power; The density of the material; Specific heat capacity; For time; , , They represent respectively to Find the partial derivative along the axis; , , Temperature along Gradient of direction;

[0291] Virtual insulation resistance and virtual leakage current are calculated using the following formulas:

[0292] , ;

[0293] In the formula, The virtual insulation resistance is the insulation resistance of the charging pile busbar to ground, simulated by the digital twin model. The virtual leakage current is the total system leakage current output from the digital twin model simulation. The DC bus voltage is the simulated output of the digital twin model, representing the virtual bus voltage. The virtual A-phase leakage current is the leakage current to ground of the A-phase winding or line output by the digital twin simulation. For virtual B-phase leakage current, the digital twin simulation outputs the B-phase winding or line leakage current to ground. This refers to the virtual C-phase leakage current, which is the C-phase winding or line leakage current to ground output by the digital twin simulation.

[0294] Virtual power The calculation formula is: ;

[0295] In the formula, Virtual input / output voltage and current respectively;

[0296] The virtual temperature gradient is represented as:

[0297] ;

[0298] In the formula, Indicates the virtual simulation temperature; , These represent the virtual temperatures corresponding to the two time slices; These represent the times of two adjacent time slices.

[0299] Extract the virtual slice mean from the virtual simulation data by dividing it into fixed slices. Virtual slice peak Virtual slice valley value Virtual slice temperature difference Virtual slope of change The virtual simulation feature vector is obtained by splicing the vectors together. ;

[0300] .

[0301] The virtual slice mean is represented as:

[0302] ;

[0303] In the formula, This represents the number of sampling points within the slice; For the first Virtual data of time;

[0304] The virtual slice peak and virtual slice valley are represented as follows:

[0305] ;

[0306] ;

[0307] In the formula, , These are the virtual slice peak value and the virtual slice valley value, respectively.

[0308] The virtual slice temperature difference is represented as:

[0309] ;

[0310] In the formula, For virtual slice temperature difference; , These are the highest and lowest virtual temperatures within the slice;

[0311] Virtual slice change slope Represented as:

[0312] .

[0313] In this embodiment, the fault prediction module 800 uses a fault prediction model constructed with a CNN-BiLSTM hybrid network to predict the fault prediction input matrix. The prediction process yields latent fault prediction results, including fault type, probability of occurrence of the corresponding fault type, and remaining service life (RUL) of the charging pile under its current state; including:

[0314] A fault prediction model constructed using a CNN-BiLSTM hybrid network is presented. The fault prediction model constructed using the CNN-BiLSTM hybrid network includes a local feature extraction module, a temporal feature extraction module, a feature fusion module, and a classifier.

[0315] The local feature extraction module includes convolutional layers and max-pooling layers, which extract features from the fault prediction input matrix. Extract local features from;

[0316] The temporal feature extraction module uses a bidirectional long short-term memory network to extract bidirectional hidden features from local features;

[0317] The feature fusion module uses an attention mechanism to weight the bidirectional hidden features to obtain the global fused features;

[0318] The classifier uses a fully connected layer and a Softmax function to predict the global fusion features, outputs the probability of occurrence of various types of faults, and outputs the remaining useful life (RUL) through a regression layer.

[0319] Embodiments of this application also provide an electronic device, including:

[0320] At least one processor;

[0321] and memory that is communicatively connected to at least one processor;

[0322] The memory stores a computer program that can be executed by at least one processor. When the computer program is executed by at least one processor, it implements a method for monitoring the operation status and providing early warning of faults in urban charging piles.

[0323] In this embodiment, the memory can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. A processor, coupled to the memory, is used to execute computer programs stored in the memory.

[0324] This application provides a computer storage medium storing a computer program, which, when executed by a processor, implements a method for monitoring the operating status and providing fault early warning of urban charging piles.

[0325] Computer programs include computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. Computer-readable media can include at least: any entity or device capable of carrying computer program code to a photographic / electronic device, recording media, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media. Examples include USB flash drives, portable hard drives, magnetic disks, or optical discs.

[0326] The embodiments and descriptions above are merely illustrative of the principles and preferred embodiments of the present invention. Various changes and modifications may be made to the present invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed.

Claims

1. A method for monitoring the operational status and providing early warning of faults in urban charging piles, characterized in that, include: Obtain raw monitoring data of electrical operating parameters, temperature parameters, environmental safety parameters, and port operating parameters of charging stations in the city; The original monitoring data is subjected to time-series alignment, noise reduction filtering, and outlier removal to obtain preprocessed monitoring data; According to the fixed slices, the slice mean, slice peak, slice valley, slice temperature difference, and slice change slope of each parameter are extracted from the preprocessed monitoring data and spliced ​​into an edge feature vector; The preprocessed monitoring data is judged using the security threshold of the rule base, the alarm flag bit of each charging pile is obtained, and the status identification information of the charging pile is statistically analyzed. The preprocessed monitoring data, the edge feature vector, the charging pile status identification information, the alarm flag, the timestamp, the unique identifier of the charging pile, the operator number, and the site number are encapsulated into a standard data packet for urban charging piles and transmitted to the cloud. After parsing the standard data of the city charging piles received from the cloud, the data is mapped to a digital twin model of the charging pile corresponding to the physical charging pile, and a virtual simulation feature vector is generated through the digital twin model of the charging pile. The virtual simulation feature vector, edge feature vector, and historical fault sample data are fused to obtain the fault prediction input matrix; A fault prediction model constructed using a CNN-BiLSTM hybrid network is used to predict the fault prediction input matrix, thereby obtaining the latent fault prediction results. Based on the predicted latent faults, early warning levels are classified in conjunction with the real-time operating parameters of the charging piles.

2. The method for monitoring the operation status and providing early warning of faults of urban charging piles according to claim 1, characterized in that, The original monitoring data is subjected to time-series alignment, noise reduction filtering, and outlier removal to obtain preprocessed monitoring data, including: Using the local master clock of the edge terminal as a unified benchmark, a standard timestamp is added to the raw monitoring data sampled by each sensor, and associated with the unique identifier of the charging pile, the operator to which it belongs, and the site number to which it belongs; Linear interpolation is used to correct time-displaced data, and the sampling data of all sensing units are converged to a fixed sampling period to form a time-aligned original monitoring data sequence. A joint denoising algorithm combining sliding median filtering and Kalman filtering is used to perform median filtering and denoising on the time-aligned original monitoring data sequence, resulting in filtered and denoised monitoring data. The monitoring data is filtered and denoised to remove outliers, resulting in preprocessed monitoring data.

3. The method for monitoring the operation status and providing early warning of faults of urban charging piles according to claim 1, characterized in that, The standard data of the city charging piles received from the cloud is parsed and mapped to a digital twin model of the charging pile corresponding to the physical charging pile. Virtual simulation features are then generated using the digital twin model of the charging pile; including: The standard data packets of urban charging piles received from the cloud are parsed to extract the preprocessed monitoring data, the edge feature vector, the timestamp, and the unique identifier of the charging pile, and to complete data verification and classification caching. The digital twin model of the charging pile is created using SolidWorks software at a 1:1 scale based on the structure of the physical charging pile. Establish a one-to-one mapping relationship between the sensor points of the physical charging pile and the parameter nodes of the digital twin model of the charging pile, and push the standard data synchronously to the corresponding parameter nodes according to the timestamp; Using the preprocessed monitoring data as boundary conditions, the digital twin model of the charging pile is driven to perform multiphysics numerical simulation calculations to obtain virtual simulation data; The virtual simulation data is processed by extracting the virtual slice mean, virtual slice peak value, virtual slice valley value, virtual slice temperature difference, and virtual change slope according to fixed slices, and then splicing them to obtain the virtual simulation feature vector.

4. The method for monitoring the operation status and providing early warning of faults of urban charging piles according to claim 1, characterized in that, The virtual simulation feature vector, edge feature vector, and historical fault sample data are fused to obtain the fault prediction input matrix, including: The virtual simulation feature vector, edge feature vector, and historical fault sample data are respectively subjected to min-max normalization processing; The normalized virtual simulation feature vector, edge feature vector, and historical fault sample data are concatenated according to feature dimensions to obtain the fused feature vector; The fused feature vectors are arranged sequentially according to time slices to form a fault prediction input matrix.

5. The method for monitoring the operation status and providing early warning of faults of urban charging piles according to claim 1, characterized in that, The fault prediction model constructed by the CNN-BiLSTM hybrid network includes a local feature extraction module, a temporal feature extraction module, a feature fusion module, and a classifier. The local feature extraction module includes a convolutional layer and a max pooling layer. The convolutional layer and the max pooling layer extract local features from the fault prediction input matrix and perform dimensionality reduction processing to obtain dimensionality-reduced local features. The temporal feature extraction module uses a bidirectional long short-term memory network to extract bidirectional hidden features from the dimensionality-reduced local features; The feature fusion module uses an attention mechanism to weight the bidirectional hidden features to obtain global fused features; The classifier uses a fully connected layer and a Softmax function to predict the global fusion features, outputs the fault category and the probability of occurrence of each fault category, and outputs the remaining useful life (RUL) through a regression layer.

6. The method for monitoring the operation status and providing early warning of faults of urban charging piles according to claim 5, characterized in that, A fault prediction model constructed using a CNN-BiLSTM hybrid network is used to predict the fault prediction input matrix. The prediction process yields latent fault prediction results, including fault type, probability of occurrence of the corresponding fault type, and remaining service life (RUL) of the charging pile under its current state; including: The fault prediction input matrix is ​​input into the convolutional layer of the local feature extraction module to extract local features, and the local features are downsampled by the max pooling layer to obtain the dimensionality-reduced local features. The local features after dimensionality reduction by the max pooling layer are input into the temporal feature extraction module, and bidirectional hidden features are extracted from the local features using a bidirectional long short-term memory network. The bidirectional hidden features are input into the feature fusion module, which uses an attention mechanism to weight the bidirectional hidden features to obtain global fused features. The global fusion features are input into the classifier, and the global fusion features are nonlinearly transformed and dimension-mapped through a fully connected layer to map the global fusion features to the output dimension corresponding to the number of fault types. The output of the fully connected layer is normalized by using the Softmax activation function, and the output value is mapped to the [0,1] interval to obtain the occurrence probability of each fault type. The classifier performs a linear and nonlinear combination mapping on the global fusion features through the regression layer of the classifier, and outputs the remaining service life of the charging pile in its current state. The fault category, the probability of occurrence corresponding to the fault type, and the remaining service life of the charging pile under the current state are output through the output layer to obtain the latent fault prediction result.

7. The method for monitoring the operation status and providing early warning of faults of urban charging piles according to claim 6, characterized in that, Based on the prediction results of latent faults, and combined with the real-time operating parameters of charging piles, the early warning levels are classified, including: Three levels of probability thresholds and three levels of preset lifespan are set, wherein the first probability threshold is less than the second probability threshold, the second probability threshold is less than the third probability threshold, the first preset lifespan is greater than the second preset lifespan, and the second preset lifespan is greater than the third preset lifespan. The warning level is set to Level 4 based on the three-level probability threshold and the three-level preset lifespan. The judgment is made by combining the predicted probability of the fault type, the remaining lifespan of the charging pile under its current state, and real-time alarm flags. Among these factors: Level 1 warning: The predicted probability of the fault type is less than the first probability threshold, and the remaining service life (RUL) of the charging pile in its current state is greater than the first preset service life, and there is no alarm signal. Level 2 warning: The predicted probability of the fault category is greater than or equal to the first probability threshold and less than the second probability threshold, and the remaining service life RUL of the charging pile in the current state is greater than the second preset service life and less than or equal to the first probability threshold, and there is no alarm signal. Level 3 warning: The predicted probability of the fault type is greater than or equal to the second probability threshold and less than the third probability threshold, and the remaining service life (RUL) of the charging pile in the current state is greater than the third preset service life and less than or equal to the second preset service life, and the alarm signs are high temperature alarm sign, water immersion alarm sign, and leakage alarm sign. Level 4 warning: The predicted probability of the fault type is greater than or equal to the third probability threshold, and the remaining service life (RUL) of the charging pile in its current state is less than or equal to the third preset service life, and the alarm sign is a smoke alarm sign.

8. A device for monitoring the operational status and providing early warning of faults in urban charging piles, characterized in that, include: The data acquisition module is used to acquire raw monitoring data of electrical operating parameters, temperature parameters, environmental safety parameters, and port operating parameters of charging stations in the city. The preprocessing module is used to perform time-series alignment, noise reduction filtering, and outlier removal on the raw monitoring data to obtain preprocessed monitoring data. The edge feature extraction module is used to extract the mean, peak, valley, temperature difference, and slope of each parameter from the preprocessed monitoring data according to fixed slices and concatenate them into an edge feature vector. The alarm flag acquisition module is used to judge the pre-processed monitoring data using the security threshold of the rule base, acquire the alarm flag bit of each charging pile, and count the status identification information of the charging pile. The data transmission module is used to encapsulate the pre-processed monitoring data, edge feature vectors, charging pile status identification information, alarm flags, timestamps, unique charging pile identifiers, operator numbers, and site numbers into a standard data packet for urban charging piles and transmit it to the cloud. The simulation module is used to parse the standard data of urban charging piles received from the cloud and map it to the digital twin model of the charging pile corresponding to the physical charging pile, and generate virtual simulation feature vectors through the digital twin model of the charging pile. The feature fusion module is used to fuse virtual simulation feature vectors, edge feature vectors, and historical fault sample data to obtain a fault prediction input matrix. The fault prediction module is used to predict the fault prediction input matrix using a fault prediction model built with a CNN-BiLSTM hybrid network, and obtain the latent fault prediction result. The early warning level classification module is used to classify early warning levels based on the prediction results of latent faults and the real-time operating parameters of charging piles.

9. An electronic device, comprising: At least one processor; and memory that is communicatively connected to at least one processor; The memory stores a computer program that can be executed by at least one processor, characterized in that when the computer program is executed by at least one processor, it implements the urban charging pile operation status monitoring and fault early warning method as described in any one of claims 1-7.

10. A computer storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the method for monitoring the operation status and providing early warning of faults of urban charging piles as described in any one of claims 1-7.

Images