Charging module grading diagnosis method, system and device based on big data

Abstract

The invention discloses a charging module grading diagnosis method, system and device based on big data, which are used for improving the accuracy of a charging module. The method comprises the following steps: acquiring multi-modal big data of a plurality of charging modules in real time; dividing the multi-modal big data into a plurality of sub-data modules according to a preset operation sample grade, and obtaining a complete proportion, a noise level and a variable coefficient; calculating a data quality score based on the complete proportion and the noise level, and determining fault sensitivity based on the coefficient of variation; calculating the data quality score and the fault sensitivity through a weighted fusion algorithm to obtain a comprehensive priority; extracting a multi-scale feature vector through a preset multi-scale feature extraction model; comparing the fluctuation amplitude of the multi-scale feature vector with a fluctuation envelope of a preset fluctuation reference to obtain a deviation value; judging whether the deviation value is higher than a preset health degree threshold value or not; and if yes, calling a preset dynamic twin model for simulation, and outputting a fault evolution trend and a diagnosis report.

Description

Technical Field

[0001] This application relates to the field of charging technology, and in particular to a method, system and device for hierarchical diagnosis of charging modules based on big data. Background Technology

[0002] With the large-scale deployment of charging modules in electric vehicle charging stations, battery energy management, and industrial energy storage, the operational stability of these modules directly determines the user's charging experience, the equipment's lifespan, and energy efficiency. Because charging modules operate in a high-frequency charging and discharging environment, they are prone to hidden faults such as cell capacity decay, poor circuit contact, and malfunctioning cooling fans. If these faults are not identified promptly and accurately, they can gradually escalate into safety incidents such as charging interruptions, equipment overload, or even fires.

[0003] In existing technologies, multimodal big data is typically collected, and then signal processing and machine learning techniques are combined to achieve fault diagnosis of charging modules. Specifically, the collected multimodal big data is first preprocessed using methods such as filtering, missing value imputation, and normalization. Then, representative features of each modality in the multimodal big data are extracted based on time domain, frequency domain, and time-frequency domain methods. Finally, a diagnostic model is used to perform diagnostic analysis on these representative features to obtain diagnostic results.

[0004] However, in actual data collection, unstable working environments exist. For example, data may be lost due to communication interruptions or subjected to strong electromagnetic interference during the collection process. As a result, the integrity and noise levels of the collected big data vary. Current technologies only preprocess these big data with varying states, which dilutes the characteristics of each modality, thus seriously affecting the accuracy and reliability of the diagnosis. Summary of the Invention

[0005] To address the aforementioned technical issues, this application provides a method, system, and apparatus for hierarchical diagnosis of charging modules based on big data.

[0006] The technical solution provided in this application is described below: The first aspect of this application provides a hierarchical diagnostic method for charging modules based on big data, the hierarchical diagnostic method for charging modules including: Real-time acquisition of multimodal big data from multiple charging modules, wherein the multimodal big data represents data from each dimension when the charging module is in operation; The multimodal big data is divided into multiple sub-data modules according to the preset running sample level, and the complete proportion, noise level and coefficient of variation of each sub-data module are obtained. The data quality score of each sub-data module is calculated based on the completeness ratio and the noise level, and the fault sensitivity of each sub-data module is determined based on the coefficient of variation. The data quality score and the fault sensitivity are weighted and calculated using a preset weighted fusion algorithm to obtain the comprehensive priority of each sub-data module; According to the overall priority order, the corresponding sub-data modules are input into the preset multi-scale feature extraction model, and the multi-scale feature vectors in the sub-data modules are extracted through the preset multi-scale feature extraction model. The fluctuation amplitude of the multi-scale feature vector in time is compared with the fluctuation envelope of the preset fluctuation benchmark, and the deviation value of the current operating state of the charging module is obtained according to the proportion by which the fluctuation amplitude exceeds the upper or lower envelope boundary of the fluctuation envelope. Determine whether the deviation value is higher than a preset health threshold; If so, a preset dynamic twin model is invoked to simulate the multi-scale feature vectors, and the fault evolution trend and diagnostic report are output.

[0007] Optionally, obtaining the deviation value of the current operating state of the charging module based on the proportion by which the fluctuation amplitude exceeds the upper or lower envelope boundary of the fluctuation envelope includes: The frequency of the fluctuation amplitude exceeding the upper or lower envelope boundary of the fluctuation envelope is determined based on the number of times the fluctuation amplitude exceeds the upper or lower envelope boundary of the fluctuation envelope. The over-limit intensity of the fluctuation amplitude is determined based on the difference between each fluctuation amplitude exceeding the upper or lower envelope boundary of the fluctuation envelope. The deviation value of the current operating state of the charging module is obtained by combining the over-limit frequency and the over-limit intensity with a preset weighting coefficient.

[0008] Optionally, after obtaining the deviation value of the current operating state of the charging module based on the proportion by which the fluctuation amplitude exceeds the upper or lower envelope boundary of the fluctuation envelope, the charging module graded diagnosis method further includes: The fluctuation envelope of the preset fluctuation benchmark is dynamically updated based on the over-limit frequency and the over-limit intensity. The step of dynamically updating the fluctuation envelope of the preset fluctuation reference based on the exceeding frequency and the exceeding intensity includes: Based on the over-limit frequency and the over-limit intensity, and combined with the preset adjustment coefficient, the suggested adjustment amounts for the upper envelope boundary and the lower envelope boundary are calculated respectively. Determine whether the suggested adjustment amount is greater than the preset maximum allowable single adjustment threshold; If not, the upper envelope boundary and the lower envelope boundary are dynamically updated according to the suggested adjustment amount.

[0009] Optionally, after determining whether the suggested adjustment amount is greater than the preset maximum allowable single adjustment threshold, the charging module graded diagnosis method further includes: If so, the preset maximum allowable single adjustment threshold is used as the suggested adjustment amount to dynamically update the upper envelope boundary and the lower envelope boundary.

[0010] Optionally, dividing the multimodal big data into multiple sub-data modules according to a preset running sample level includes: Based on the preset running sample level, a cross-modal joint feature template is constructed for each level, wherein the cross-modal joint feature template includes each modal feature template and the cross-constraint relationship between different modalities; The multimodal big data is synchronously sliced ​​to generate multiple cross-modal data units within the same time window, and the single-modal features and inter-modal cross features in each cross-modal data unit are extracted. The single-modal features and the inter-modal cross features are matched with the modal feature templates and the cross constraint relationships, respectively, and the matching degree is calculated. Based on the matching degree, the cross-modal data units are divided into the corresponding preset running sample levels, and the cross-modal data units in different levels are aggregated to obtain multiple sub-data modules.

[0011] Optionally, after calculating the matching degree, the charging module hierarchical diagnosis method further includes: Sort all the cross-modal data units from high to low according to the matching degree; After sorting, check whether the difference between the matching degree corresponding to each cross-modal data unit and the matching degree corresponding to the previous cross-modal data unit is greater than or equal to a preset minimum discrimination margin. If so, then the step of dividing the cross-modal data unit into the corresponding preset running sample level according to the matching degree is performed.

[0012] Optionally, the preset multi-scale feature extraction model includes a variational mode decomposition unit, a multi-dimensional computation unit, and an attention fusion unit; The step of extracting multi-scale feature vectors from the sub-data module using the preset multi-scale feature extraction model includes: The variational mode decomposition unit decomposes the time-series signal in the sub-data module to obtain multiple intrinsic mode components with different center frequencies. The time-domain statistics, frequency-domain energy spectrum, and joint time-frequency distribution of each intrinsic mode component are calculated by the multi-dimensional computing unit. The attention fusion unit fuses the time-domain statistics, the frequency-domain energy spectrum, and the time-frequency domain joint distribution to generate the multi-scale feature vector.

[0013] A second aspect of this application provides a charging module hierarchical diagnostic system based on big data, the charging module hierarchical diagnostic system comprising: The first acquisition unit is used to acquire multimodal big data of multiple charging modules in real time, wherein the multimodal big data represents the data of each dimension when each charging module is in operation. The second acquisition unit is used to divide the multimodal big data into multiple sub-data modules according to a preset running sample level, and to acquire the complete proportion, noise level and coefficient of variation of each sub-data module. A determining unit is configured to calculate the data quality score of each sub-data module based on the completeness ratio and the noise level, and to determine the fault sensitivity of each sub-data module based on the coefficient of variation; The calculation unit is used to perform weighted calculation of the data quality score and the fault sensitivity through a preset weighted fusion algorithm to obtain the comprehensive priority of each sub-data module; An extraction unit is used to input the corresponding sub-data modules into a preset multi-scale feature extraction model according to the order of the comprehensive priority, and extract multi-scale feature vectors from the sub-data modules through the preset multi-scale feature extraction model. The comparison unit is used to compare the fluctuation amplitude of the multi-scale feature vector in time with the fluctuation envelope of the preset fluctuation benchmark, and obtain the deviation value of the current operating state of the charging module according to the proportion of the fluctuation amplitude exceeding the upper or lower envelope boundary of the fluctuation envelope. The judgment unit is used to determine whether the deviation value is higher than a preset health threshold. The simulation unit is used to call a preset dynamic twin model to simulate the multi-scale feature vector if the condition is met, and output the fault evolution trend and diagnosis report.

[0014] Optionally, the comparison unit is specifically used for: The frequency of the fluctuation amplitude exceeding the upper or lower envelope boundary of the fluctuation envelope is determined based on the number of times the fluctuation amplitude exceeds the upper or lower envelope boundary of the fluctuation envelope. The over-limit intensity of the fluctuation amplitude is determined based on the difference between each fluctuation amplitude exceeding the upper or lower envelope boundary of the fluctuation envelope. The deviation value of the current operating state of the charging module is obtained by combining the over-limit frequency and the over-limit intensity with a preset weighting coefficient.

[0015] Optionally, it also includes a first update unit, specifically used for: The fluctuation envelope of the preset fluctuation benchmark is dynamically updated based on the over-limit frequency and the over-limit intensity. The step of dynamically updating the fluctuation envelope of the preset fluctuation reference based on the exceeding frequency and the exceeding intensity includes: Based on the over-limit frequency and the over-limit intensity, and combined with the preset adjustment coefficient, the suggested adjustment amounts for the upper envelope boundary and the lower envelope boundary are calculated respectively. Determine whether the suggested adjustment amount is greater than the preset maximum allowable single adjustment threshold; If not, the upper envelope boundary and the lower envelope boundary are dynamically updated according to the suggested adjustment amount.

[0016] Optionally, a second update unit may also be included, specifically for: If so, the preset maximum allowable single adjustment threshold is used as the suggested adjustment amount to dynamically update the upper envelope boundary and the lower envelope boundary.

[0017] Optionally, the second acquisition unit is specifically used for: Based on the preset running sample level, a cross-modal joint feature template is constructed for each level, wherein the cross-modal joint feature template includes each modal feature template and the cross-constraint relationship between different modalities; The multimodal big data is synchronously sliced ​​to generate multiple cross-modal data units within the same time window, and the single-modal features and inter-modal cross features in each cross-modal data unit are extracted. The single-modal features and the inter-modal cross features are matched with the modal feature templates and the cross constraint relationships, respectively, and the matching degree is calculated. Based on the matching degree, the cross-modal data units are divided into the corresponding preset running sample levels, and the cross-modal data units in different levels are aggregated to obtain multiple sub-data modules.

[0018] Optionally, a detection unit may also be included, specifically for: Sort all the cross-modal data units from high to low according to the matching degree; After sorting, check whether the difference between the matching degree corresponding to each cross-modal data unit and the matching degree corresponding to the previous cross-modal data unit is greater than or equal to a preset minimum discrimination margin. If so, then the step of dividing the cross-modal data unit into the corresponding preset running sample level according to the matching degree is performed.

[0019] Optionally, the preset multi-scale feature extraction model includes a variational mode decomposition unit, a multi-dimensional computation unit, and an attention fusion unit; The step of extracting multi-scale feature vectors from the sub-data module using the preset multi-scale feature extraction model includes: The variational mode decomposition unit decomposes the time-series signal in the sub-data module to obtain multiple intrinsic mode components with different center frequencies. The time-domain statistics, frequency-domain energy spectrum, and joint time-frequency distribution of each intrinsic mode component are calculated by the multi-dimensional computing unit. The attention fusion unit fuses the time-domain statistics, the frequency-domain energy spectrum, and the time-frequency domain joint distribution to generate the multi-scale feature vector.

[0020] A third aspect of this application provides a charging module hierarchical diagnostic device based on big data, the charging module hierarchical diagnostic device comprising: Processor, memory, input / output units, and bus; The processor is connected to the memory, the input / output unit, and the bus; The memory stores a program, which the processor calls to execute the first aspect and any optional charging module graded diagnostic method in the first aspect.

[0021] The fourth aspect of this application provides a computer-readable storage medium storing a program that, when executed on a computer, performs the first aspect and any optional charging module graded diagnosis method of the first aspect.

[0022] As can be seen from the above technical solutions, the embodiments of this application have the following advantages: 1. Acquire multimodal big data, divide it into sub-data modules according to the preset running sample level, and extract the complete proportion, noise level and coefficient of variation to provide a basis for data quality assessment; 2. Calculate the quality score of each sub-data module based on the extracted complete proportion and noise level, determine the fault sensitivity based on the coefficient of variation, and obtain the comprehensive priority by weighting to avoid homogenized input of low-quality data; 3. After obtaining the overall priority, the sub-data modules are input into the preset multi-scale feature extraction model in order of overall priority to accurately extract multi-scale feature vectors, further preventing low-quality data from diluting key features; 4. Based on the preset fluctuation benchmark, the fluctuation amplitude of the multi-scale feature vector is compared to obtain the deviation value. Then, combined with dynamic twin simulation and cross-validation, the fault evolution trend and diagnosis report are obtained. This not only filters high-quality and high-value data, but also enhances the accuracy of features and the reliability of diagnosis, and improves the accuracy of charging module hierarchical diagnosis and fault prediction capability. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 A schematic flowchart of an embodiment of the charging module graded diagnosis method provided in this application; Figure 2 A flowchart illustrating one implementation of step S102 in the charging module graded diagnosis method provided in this application; Figure 3 A flowchart illustrating one implementation of step S105 in the charging module graded diagnosis method provided in this application; Figure 4 A flowchart illustrating one implementation of step S106 in the graded diagnostic method for charging modules provided in this application; Figure 5 A schematic diagram of an embodiment of the charging module hierarchical diagnostic system provided in this application; Figure 6 This is a schematic diagram of an embodiment of the charging module hierarchical diagnostic device provided in this application. Detailed Implementation

[0025] In this embodiment, the entity executing the charging module hierarchical diagnostic method is not limited to a specific type of charging control device, computing terminal, or diagnostic platform. The charging module hierarchical diagnostic method can be executed by any hardware, software, or a combination of hardware and software system capable of real-time acquisition, analysis, processing, and diagnostic decision-making of multimodal big data from charging modules, such as a dedicated controller for charging systems, a general-purpose computer, a cloud server cluster, an edge computing node, a virtualization processing platform, or an integrated management environment for charging networks.

[0026] The charging module hierarchical diagnostic method of this embodiment can also be implemented by controlling the remote execution unit of the charging module through the local embedded control program of the charging module, the charging network management software module, and the network communication interface of the remote diagnostic platform. Regardless of whether the specific execution entity is a single diagnostic device directly connected to the charging module, multiple parallel collaborative data analysis and processing units, or a scalable distributed charging diagnostic system, this method can be operated according to the step sequence and logic in the embodiments below.

[0027] Based on this, the big data-based hierarchical diagnosis method for charging modules proposed in this application is particularly suitable for scenarios such as public charging stations with multiple charging modules operating in parallel, new energy vehicle battery swapping stations, and dedicated charging systems in the industrial field.

[0028] To enable those skilled in the art to more clearly understand the technical terms involved in the charging module graded diagnosis method of this application, the key terms are defined as follows: Multimodal Big Data: refers to the collection of multi-dimensional and multi-type operating data collected in real time when the charging module is in operation. Specifically, it includes, but is not limited to, the output current, output voltage, cavity temperature, cooling fan speed, charging time, energy consumption data, cell equalization voltage, IGBT device temperature, etc., which can comprehensively reflect the different dimensions of the module's operating status.

[0029] Coefficient of Variation (CVA): This refers to the ratio of the standard deviation to the mean of a certain dimension of data in a sub-data module. It is used to quantify the dispersion of the data in that dimension. The larger the CVA, the more drastic the data fluctuation, the more prone the charging module is to failure, and the higher the fault sensitivity.

[0030] Multi-scale Feature Vector: Refers to vector data that can quantify the operating characteristics of a charging module, covering feature information of different frequency scales and dimensions.

[0031] Preset Fluctuation Benchmark: This refers to a pre-defined standard for the normal fluctuation range of multi-scale feature vectors based on historical data under normal operating conditions of the charging module. It serves as a benchmark for judging whether the current operating state of the charging module is abnormal.

[0032] Digital Twin Model: A digital virtual model built on the physical structure and operating principle of a charging module, which simulates the operation and evolution of the charging module under the current abnormal state and outputs the fault evolution trend and diagnostic report.

[0033] Cross-modal constraint relationship: refers to the association rules between different modal data at a specific operational sample level, used to verify the consistency and rationality of multimodal data.

[0034] Please see Figure 1 This application first provides an embodiment of a big data-based hierarchical diagnostic method for charging modules, which includes: S101. Real-time acquisition of multimodal big data from multiple charging modules, where multimodal big data represents data from each dimension when each charging module is in operation; In this embodiment, comprehensive collection of multimodal big data provides complete raw data support for subsequent data quality assessment, feature extraction, and fault diagnosis, avoiding one-sided diagnosis due to missing data in a single dimension. Multimodal big data refers to data in various dimensions that can reflect the operating status of the charging module, covering types such as electrical parameters, environmental parameters, and equipment status parameters.

[0035] Specifically, data is collected in real time through the sensors built into the charging module. For example, the output voltage and charging current of the charging module are collected every 100 milliseconds, and the internal temperature and cooling fan speed of the charging module are collected every 5 seconds. At the same time, information such as the running time and start / stop status of the charging module is recorded.

[0036] It is important to note that this big data needs to be acquired continuously and without interruption to ensure coverage of the entire operation phase of the charging module, from standby, charging, full charge to shutdown.

[0037] S102. Divide the multimodal big data into multiple sub-data modules according to the preset running sample level, and obtain the complete proportion, noise level and coefficient of variation of each sub-data module; To provide a basis for subsequent quality assessment, this embodiment mainly achieves data classification management through grade division and extracts the basic characteristics of quantitative data.

[0038] To achieve data classification management through grading, it is necessary to first determine the preset operating sample grades. The preset operating sample grades are divided according to the operating load and working scenario of the charging module, for example, into three grades: "light load operation (charging power < 30% of rated power)", "medium load operation (30% ≤ charging power ≤ 70% of rated power)" and "heavy load operation (charging power > 70% of rated power)".

[0039] After determining the preset operating sample level based on actual needs, the acquired multimodal big data is classified according to the operating sample level to form corresponding sub-data modules. For example, the voltage, current and temperature data collected when the charging power is 25% are classified as the sub-data module for light load operation.

[0040] Meanwhile, the system acquires the completeness ratio, noise level, and coefficient of variation for each sub-data module. The completeness ratio is the proportion of valid data in the sub-data module to the total amount of data to be collected. For example, if a light-load module should collect 1000 data points and actually collects 950 valid data points, then the completeness ratio is 95%. The noise level is determined by analyzing the irregular components of data fluctuations in the sub-data module, such as the irregular fluctuation amplitude caused by electromagnetic interference in current data. The coefficient of variation reflects the dispersion of data in the sub-data module, such as the degree of difference in current data at different times in the same heavy-load module.

[0041] See Figure 2 The following provides a specific implementation method for step S102, which includes: S1021. Construct cross-modal joint feature templates for each level based on preset running sample levels, wherein the cross-modal joint feature templates include each modal feature template and the cross-constraint relationships between different modalities; To construct cross-modal joint feature templates for each level, it is necessary to first determine the feature templates for each mode based on each level in the preset running sample levels. For example, for electrical parameter modes, the light load template is set to current 5-10A and voltage 220-230V, the medium load template is 10-20A and 220-235V, and the heavy load template is 20-30A and 230-240V; for visual modes, the light load template is temperature 30-40℃ and speed 1500-2000r / min, the medium load template is 38-48℃ and 2000-2500r / min, and the heavy load template is 45-55℃ and 2500-3000r / min; for acoustic modes, the light load template is fan sound pressure level 50-55dB, the medium load template is 55-60dB, and the heavy load template is 60-65dB.

[0042] Simultaneously, it is necessary to determine the cross-constraint relationships between different modes. For example, under light load, "for every 1A increase in current, the temperature rises by 0.5℃ and the fan speed increases by 100 r / min"; under medium load, "for every 1A increase in current, the temperature rises by 0.6℃ and the fan speed increases by 120 r / min"; and under heavy load, "for every 1A increase in current, the temperature rises by 0.7℃ and the fan speed increases by 150 r / min". Finally, the feature templates of each mode and the cross-constraint relationships between different modes are combined to construct a complete cross-modal joint feature template.

[0043] S1022. Perform synchronous time slicing on multimodal big data to generate multiple cross-modal data units within the same time window, and extract the single-modal features and inter-modal cross features in each cross-modal data unit; To synchronize time slices for multimodal big data, it is necessary to first determine the granularity of the time slice to ensure that the timestamps of all modal data within the slice are consistent, and to avoid feature misalignment caused by time asynchrony. For example, collect current, temperature and fan noise data of the charging module from 14:00 to 14:01 to form a cross-modal data unit. The cross-modal data unit includes electrical parameters, visual and acoustic data within 1 minute.

[0044] Subsequently, single-mode features are extracted from the cross-modal data units. For example, single-mode features such as "average current 8.5A, voltage fluctuation ±0.5V" are extracted from electrical parameter data; "average temperature 35℃, rotational speed stability 95%" are extracted from visual data; and "average sound pressure level 52dB, noise-free percentage 100%" are extracted from acoustic data. Simultaneously, inter-modal cross-features are extracted, such as the correlation features between electrical parameters and visual data, electrical parameters and acoustic data, and visual data and acoustic data, to ensure that each cross-modal data unit contains both single-mode and cross-feature features.

[0045] S1023. Match the single-modal features and inter-modal cross features with each modal feature template and cross constraint relationship respectively, and calculate the matching degree; After extracting the single-modal features and inter-modal cross features, the single-modal features and inter-modal cross features extracted from each cross-modal data unit are compared and matched one by one with the modal feature templates and cross-constraint relationships constructed at each preset running sample level.

[0046] Specifically, single-modal feature matching matches single-modal features with single-modal feature templates to quantify the fit of a single modality; inter-modal cross-feature matching matches inter-modal cross-features with cross-constraint relationships to quantify the fit of modal associations. While completing the matching, a matching degree is also calculated, primarily to transform the fit of single-modal features and the fit of modal associations into a single, comparable quantitative indicator, avoiding bias caused by single-path judgments.

[0047] S1024. Sort all cross-modal data units from high to low according to their matching degree; After calculating the matching degree, all cross-modal data units are sorted from high to low according to the matching degree. By sorting them in an ordered manner, the differences in the degree of fit between the cross-modal data units and the templates at each level are presented intuitively. For example, if there are 3 cross-modal data units with matching degrees of 94.6, 91.8, and 93.2 respectively, sorting them from high to low will result in the following order: Cross-modal data unit 1: 94.6 → Cross-modal data unit 3: 93.2 → Cross-modal data unit 2: 91.8.

[0048] It is important to note that during the sorting process, the matching degree value of each cross-modal data unit must be accurate to avoid sorting deviations caused by numerical errors.

[0049] S1025. Check whether the difference between the matching degree corresponding to each cross-modal data unit after sorting and the matching degree corresponding to the previous cross-modal data unit is greater than or equal to the preset minimum discrimination margin. In this embodiment, starting from the second cross-modal data unit after sorting, the matching degree difference between the current cross-modal data unit and the previous cross-modal data unit is calculated sequentially. For example, the preset minimum discrimination margin is 10 points. The difference between cross-modal data unit 2 (93.2 points) and cross-modal data unit 1 (94.6 points) is 1.4 points < 10 points; the difference between cross-modal data unit 3 (91.6 points) and cross-modal data unit 2 (93.2 points) is 1.6 points, 1.6 points < 10 points, and so on until the difference calculation of all cross-modal data units is completed.

[0050] When it is detected that the difference between the matching degree corresponding to each cross-modal data unit after sorting and the matching degree corresponding to the previous cross-modal data unit is greater than or equal to the preset minimum discrimination margin, step S1026 is executed; when it is detected that the difference between the matching degree corresponding to one or more cross-modal data units after sorting and the matching degree corresponding to the previous cross-modal data unit is less than the preset minimum discrimination margin, all cross-modal data units with differences less than the preset minimum discrimination margin are uniformly classified into the same preset running sample level, and the aggregation step is executed.

[0051] It is important to note that during the detection process, the differences between adjacent cross-modal data units need to be checked one by one to ensure that no boundary points are missed.

[0052] S1026. Based on the matching degree, divide the cross-modal data units into the corresponding preset running sample levels, and aggregate the cross-modal data units in different levels to obtain multiple sub-data modules.

[0053] When it is detected that the difference between the matching degree of each cross-modal data unit after sorting and the matching degree of the previous cross-modal data unit is greater than or equal to the preset minimum discrimination margin, the cross-modal data units are classified into levels according to the matching degree and aggregated to obtain sub-data modules.

[0054] Specifically, first determine the boundaries of each level in the preset running sample levels, and then divide the data according to these boundaries. For example, the boundary of level A is greater than or equal to 85 points, the boundary of level B is less than 85 points but greater than or equal to 75 points, and the boundary of level C is less than 75 points. When the matching degree is 98 points, the cross-modal data unit corresponding to the matching degree is classified as level A; when the matching degree is 88 points, the cross-modal data unit corresponding to the matching degree is classified as level B; and when the matching degree is 60 points, the cross-modal data unit corresponding to the matching degree is classified as level C.

[0055] After being assigned to the corresponding preset operational sample levels, all cross-modal data units within the same level are integrated into a sub-data module. This avoids subsequent processing biases caused by data mixing from different levels. The aggregation process preserves the complete characteristics of each cross-modal data unit, ensuring that the sub-data module can comprehensively reflect the multimodal operational status of that level.

[0056] S103. Calculate the data quality score of each sub-data module based on the integrity ratio and noise level, and determine the fault sensitivity of each sub-data module based on the coefficient of variation. After obtaining the complete proportion, noise level, and coefficient of variation of each sub-data module, a data quality score is calculated based on the obtained complete proportion and noise level, and the fault sensitivity is determined based on the coefficient of variation.

[0057] Specifically, when calculating the data quality score, the weights for the completeness ratio and noise level should be set according to their importance. For example, the completeness ratio accounts for 60% of the weight and the noise level accounts for 40%. If the completeness ratio of a certain sub-data module is 95%, the corresponding score is 95 points, and the noise level is 90%, the corresponding score is 90 points. Then the data quality score = 95 × 60% + 90 × 40% = 93 points. The higher the score, the better the data quality.

[0058] While calculating the data quality score, fault sensitivity is determined based on the coefficient of variation. The coefficient of variation reflects the data dispersion of each sub-data module. If the coefficient of variation of a sub-data module is far beyond the normal range, it indicates that the data fluctuation of that sub-data module is abnormal and is more likely to reflect potential faults in the charging module, i.e., the fault sensitivity is high. Conversely, sub-data modules with a coefficient of variation close to the normal range have low fault sensitivity.

[0059] S104. The data quality score and fault sensitivity are weighted and calculated using a preset weighted fusion algorithm to obtain the comprehensive priority of each sub-data module. In this embodiment, the preset weighted fusion algorithm is used to set reasonable weights for the calculated data quality score and fault sensitivity and then fuse them. The weight setting should be determined in combination with the actual diagnostic needs, and there is no limitation here. For example, if the actual diagnostic needs focus more on data reliability, the weight of the data quality score can be set to 60% and the weight of the fault sensitivity to 40%; if the focus is more on fault detection efficiency, the weight of the data quality score can be set to 40% and the weight of the fault sensitivity to 60%.

[0060] Then, the overall priority of each sub-data module is calculated according to the weight set determined by the preset weighted fusion algorithm. For example, if the data quality score of a certain sub-data module is 90 points and the fault sensitivity is 85 points, after weighted calculation by the preset weighted fusion algorithm, the overall priority of the sub-data module is 90×50%+85×50%=87.5 points.

[0061] S105. Input the corresponding sub-data modules into the preset multi-scale feature extraction model according to the order of comprehensive priority, and extract the multi-scale feature vectors in the sub-data modules through the preset multi-scale feature extraction model. By prioritizing data from high-quality, high-value sub-data modules, the pre-defined multi-scale feature extraction model is prioritized for processing. This allows the model to capture key information from the data more comprehensively.

[0062] Specifically, the sub-data modules are sorted from largest to smallest according to the calculated comprehensive priority value, and then the corresponding sub-data modules are sequentially input into the preset multi-scale feature extraction model in descending order. The preset multi-scale feature extraction model has the ability to extract features at different scales, and can extract features from the micro-details, meso-trends and macro-laws of the sub-data modules.

[0063] Furthermore, the preset multi-scale feature extraction model first preprocesses the input sub-data modules, and then transforms the data of different dimensions in the sub-data modules into a multi-scale feature vector of a unified dimension. For example, the voltage micro-fluctuation, current meso-trend, and temperature macro-law of module A are integrated into a vector containing multiple feature values.

[0064] See Figure 3 The following provides a specific implementation method for step S105, which includes: S1051. The preset multi-scale feature extraction model includes a variational mode decomposition unit, a multi-dimensional computation unit, and an attention fusion unit. The variational mode decomposition unit decomposes the time-series signals in the sub-data modules to obtain multiple intrinsic mode components with different center frequencies. The variational mode decomposition unit primarily processes the timing signals in the input sub-data module. These timing signals include voltage and current timing signals generated during the operation of the charging module. For example, the sub-data module records the voltage timing data of the charging module over one hour. The variational mode decomposition unit uses its built-in signal processing logic to decompose these timing signals into multiple eigenmode components with different center frequencies. Different center frequencies correspond to signal characteristics at different scales. For instance, the voltage timing signal can be decomposed into three eigenmode components with center frequencies of 5Hz, 1Hz, and 0.2Hz.

[0065] The intrinsic mode components with different center frequencies obtained by decomposition correspond to the signal characteristics at the micro, meso, and macro levels in the sub-data module, respectively. This provides a clear and specific processing object for subsequent multi-dimensional computing units, ensuring that the extracted features can fully cover the information at all scales of the sub-data module.

[0066] S1052. The time-domain statistics, frequency-domain energy spectrum, and joint time-frequency distribution of each intrinsic mode component are calculated by multi-dimensional computing units. After obtaining multiple intrinsic mode components with different center frequencies through the variational mode decomposition unit, the multidimensional computing unit performs multidimensional feature calculations on each intrinsic mode component to obtain the time-domain statistics, frequency-domain energy spectrum, and time-frequency joint distribution of each intrinsic mode component.

[0067] Time-domain statistical features can intuitively reflect the overall change pattern and fluctuation of intrinsic mode components in the time dimension; frequency-domain energy spectrum can reveal the energy distribution of intrinsic mode components in the frequency dimension, helping to identify specific frequency anomalies in the operation of the charging module; time-frequency joint distribution can combine time and frequency information to capture the dynamic characteristic changes of intrinsic mode components.

[0068] Furthermore, time-domain statistical features include, but are not limited to, the mean, variance, maximum, and minimum values ​​of intrinsic mode components over time. For example, for intrinsic mode components reflecting the micro-fluctuations of the charging module current, the mean and variance of the current over 10 minutes are calculated to reflect the average level and fluctuation of the current during that time period. Frequency-domain energy spectrum is mainly calculated by analyzing the energy distribution of intrinsic mode components in different frequency bands. For example, the frequency-domain energy spectrum of the current micro-fluctuation component can reflect the energy proportion in the high-frequency band, and this energy proportion determines the severity of current changes. The time-frequency joint distribution simultaneously reflects the characteristic changes of intrinsic mode components in both time and frequency dimensions. For example, the frequency distribution of the current micro-fluctuation component at a certain moment can be observed through the time-frequency joint distribution to capture the abnormal frequency characteristics of the current at a specific time point.

[0069] S1053. The attention fusion unit fuses the time-domain statistics, frequency-domain energy spectrum, and time-frequency joint distribution to generate a multi-scale feature vector.

[0070] After obtaining the time-domain statistics, frequency-domain energy spectrum, and time-frequency joint distribution features of each intrinsic mode component, the attention fusion unit fuses the time-domain statistics, frequency-domain energy spectrum, and time-frequency joint distribution features to generate a multi-scale feature vector.

[0071] Specifically, the attention fusion unit assesses the importance of the input time-domain statistics, frequency-domain energy spectrum, and joint time-frequency distribution features. This assessment process considers the charging module's operational characteristics and key monitoring requirements, assigning appropriate attention weights to different types of features. Based on these weights, the three types of features are weighted and fused, integrating features of different dimensions and importance into a single vector of unified dimension—a multi-scale feature vector. This fusion method ensures both the comprehensiveness of the features and highlights key information, enabling the generated multi-scale feature vector to accurately and efficiently reflect the core features of the sub-data module.

[0072] S106. Compare the fluctuation amplitude of the multi-scale feature vector in time with the fluctuation envelope of the preset fluctuation benchmark, and obtain the deviation value of the current operating state of the charging module according to the proportion of the fluctuation amplitude exceeding the upper or lower envelope boundary of the fluctuation envelope. After extracting the multi-scale feature vectors, the temporal fluctuation amplitude of these feature vectors is obtained. Temporal fluctuation amplitude refers to the degree of difference in the multi-scale feature vectors over time, such as the range of numerical changes of a voltage-related feature vector within one hour. Subsequently, the fluctuation amplitude of this multi-scale feature vector is compared with the fluctuation envelope of a preset fluctuation benchmark. This preset fluctuation benchmark is a normal fluctuation range statistically obtained based on the multi-scale feature vectors of a large number of healthy charging modules, presented in the form of a fluctuation envelope, specifically including an upper envelope boundary and a lower envelope boundary. The upper envelope boundary represents the maximum value of normal fluctuation, and the lower envelope boundary represents the minimum value of normal fluctuation.

[0073] During the comparison process, the proportion of times the fluctuation amplitude exceeds the upper envelope boundary or falls below the lower envelope boundary will be counted out of the total number of counts. This proportion is the deviation value of the current operating state of the charging module. For example, in 100 counts, if the fluctuation amplitude exceeds the upper envelope boundary 15 times and falls below the lower envelope boundary 5 times, the total deviation ratio is 20%, that is, the deviation value is 20%.

[0074] See Figure 4 The following provides a specific implementation method for step S106, which includes: S1061. Determine the frequency of fluctuation exceeding the upper or lower envelope boundary of the fluctuation amplitude based on the number of times the fluctuation amplitude exceeds the upper or lower envelope boundary of the fluctuation envelope. In this embodiment, the frequency of exceeding the limit is determined mainly based on the comparison results between the fluctuation amplitude of the multi-scale feature vector and the preset fluctuation envelope, thereby quantifying the frequency of occurrence of the exceeding event and avoiding the one-sidedness of focusing only on a single exceeding while ignoring the overall frequency.

[0075] Specifically, first determine the total number of statistical counts within the statistical period. For example, using a 1-hour statistical period, the fluctuation amplitude of a multi-scale feature vector is counted once every 5 minutes, resulting in a total of 12 counts. After determining this, each count is checked to see if the fluctuation amplitude exceeds the upper envelope boundary or falls below the lower envelope boundary. If a count meets the criteria of exceeding the upper envelope boundary or falling below the lower envelope boundary, it is recorded as one instance of exceeding the limit. This process is repeated to count all instances of exceeding the limit within the entire statistical period. Then, the number of instances of exceeding the limit is divided by the total number of counts to obtain the frequency of fluctuation exceeding the limit. For example, in the above 12 total counts, if there are 3 instances where the fluctuation amplitude exceeds the upper envelope boundary or falls below the lower envelope boundary, then the frequency of exceeding the limit is 3 ÷ 12 = 25%.

[0076] S1062. Determine the over-limit intensity of the fluctuation amplitude based on the difference between the fluctuation amplitude exceeding the upper or lower envelope boundary of the fluctuation envelope for each fluctuation. After determining the frequency of fluctuations exceeding the limit, the intensity of each exceedance is analyzed to show the specific degree to which each fluctuation deviates from the normal range.

[0077] Specifically, the process involves obtaining the numerical value of the fluctuation amplitude corresponding to each out-of-limit event, as well as the values ​​of the upper or lower envelope boundaries within the preset fluctuation envelope. If the fluctuation amplitude exceeds the upper envelope boundary, the difference between the fluctuation amplitude value and the upper envelope boundary value is the out-of-limit intensity relative to the upper envelope boundary. If the fluctuation amplitude is lower than the lower envelope boundary, the difference between the fluctuation amplitude value and the lower envelope boundary value is the out-of-limit intensity relative to the lower envelope boundary. For example, if the fluctuation amplitude is 15V and the upper envelope boundary value is 12V, then the out-of-limit intensity is 15-12=3V. In this way, the corresponding out-of-limit intensity is calculated for each out-of-limit situation, thereby comprehensively understanding the degree to which each fluctuation amplitude deviates from the normal range.

[0078] S1063. The deviation value of the current operating state of the charging module is obtained by integrating the over-limit frequency and over-limit intensity with the preset weighting coefficient.

[0079] Based on the calculated over-limit frequency and over-limit intensity, the over-limit frequency and over-limit intensity are fused together with preset weighting coefficients. These preset weighting coefficients are set according to the actual operating characteristics of the charging module, historical fault data, and the importance of their impact on the operating status of the charging module.

[0080] Furthermore, the over-limit frequency is multiplied by a preset weighting coefficient to obtain a weighted value for the over-limit frequency; the over-limit intensity is multiplied by a preset weighting coefficient to obtain a weighted value for the over-limit intensity. The sum of these two weighted values ​​is the deviation value of the current operating state of the charging module.

[0081] The deviation value is calculated by integrating the over-limit frequency and over-limit intensity with a preset weighting coefficient. This fully considers the different degrees of importance of these two indicators on the operating status of the charging module, making the calculated deviation value more accurately reflect the actual operating status of the charging module.

[0082] If only one indicator such as over-limit frequency or over-limit intensity is used for evaluation, it will lead to a deviation in the judgment of the charging module's operating status.

[0083] S1064. Based on the over-limit frequency and over-limit intensity, and combined with the preset adjustment coefficient, calculate the suggested adjustment amount for the upper envelope boundary and the lower envelope boundary respectively; In this embodiment, the preset adjustment coefficient is determined based on the historical data of the charging module's long-term operation, the fluctuation characteristics under different operating conditions, and the sensitivity requirements for adjusting the envelope boundary. Different adjustment coefficients are set for the upper and lower envelope boundaries. For example, the adjustment coefficient for the upper envelope boundary is k1, and the adjustment coefficient for the lower envelope boundary is k2.

[0084] The calculation of the suggested adjustment for the upper envelope boundary mainly involves combining the frequency exceeding the upper envelope boundary with the corresponding over-limit intensity, and then multiplying this by a preset adjustment coefficient for the upper envelope boundary to obtain the suggested adjustment amount. Similarly, the suggested adjustment for the lower envelope boundary is calculated by multiplying the frequency below the lower envelope boundary with the corresponding over-limit intensity by a preset adjustment coefficient for the lower envelope boundary. For example, if the preset adjustment coefficient for the upper envelope boundary is set to k1 = 0.2, and the frequency exceeding the upper envelope boundary by 15% corresponds to an over-limit intensity of 3, then the suggested adjustment for the upper envelope boundary would be 15% × 3 × 0.2 = 0.09. This calculation method allows for more targeted adjustments to the envelope boundaries.

[0085] S1065. Determine whether the suggested adjustment amount is greater than the preset maximum allowable single adjustment threshold. After calculating the suggested adjustment amounts for the upper and lower envelope boundaries, these two suggested adjustment amounts need to be compared with the preset maximum allowable single adjustment threshold. Without this comparison, if the suggested adjustment amount is too large, a single large adjustment of the upper or lower envelope boundary will drastically change the original normal fluctuation range, leading to distortion in the subsequent judgment criteria for the charging module's operating status. This could result in misjudging normal operating conditions as abnormal, or vice versa, thus affecting the normal use and maintenance decisions of the charging module.

[0086] Furthermore, the preset maximum allowable single adjustment threshold is determined based on the stability requirements of the charging module's fluctuation envelope boundary, the safety of the charging module's operation, and historical adjustment experience. The preset maximum allowable single adjustment threshold is to prevent the fluctuation envelope boundary from changing drastically due to excessive single adjustment, thereby affecting the accuracy and stability of judging the charging module's operating status.

[0087] Specifically, the suggested adjustment amounts for the upper and lower envelope boundaries are compared one by one with the preset maximum permissible single adjustment threshold. If any suggested adjustment amount is greater than the preset maximum permissible single adjustment threshold, step S1067 is executed; if all suggested adjustment amounts are not greater than the preset maximum permissible single adjustment threshold, step S1066 is executed.

[0088] S1066. Based on the suggested adjustment amount, dynamically update the upper and lower envelope boundaries respectively.

[0089] When the suggested adjustment amount is not greater than the preset maximum allowed single adjustment threshold, the current upper envelope boundary, lower envelope boundary, and the specific values ​​of the suggested adjustment amount are obtained for updating.

[0090] Furthermore, for the upper envelope boundary, the current upper envelope boundary value is added to the suggested adjustment amount of the upper envelope boundary to obtain the updated upper envelope boundary; for the lower envelope boundary, the current lower envelope boundary value is added to the suggested adjustment amount of the lower envelope boundary to obtain the updated lower envelope boundary.

[0091] After the update is completed, the fluctuation envelope can always be synchronized with the actual operating status of the charging module, and adapt to changes in the operating characteristics of the charging module during long-term use.

[0092] S1067. The upper and lower envelope boundaries are dynamically updated using the preset maximum allowable single adjustment threshold as the suggested adjustment amount.

[0093] When any suggested adjustment exceeds the preset maximum allowable single adjustment threshold, to avoid adverse effects from a large single adjustment of the upper or lower envelope boundary, the calculated suggested adjustment amount will no longer be used. Instead, the preset maximum allowable single adjustment threshold will be used as the new suggested adjustment amount to dynamically update the corresponding upper or lower envelope boundary. In this way, while ensuring the envelope boundary can be adjusted to adapt to changes in actual conditions, the magnitude of a single adjustment is strictly controlled, ensuring the stability of the charging module.

[0094] S107. Determine whether the deviation value is higher than the preset health threshold; In order to quickly classify the operating status of the charging module, after obtaining the deviation value, the deviation value is compared with the preset health threshold to determine whether the operating status of the charging module is abnormal. The preset health threshold is determined based on the actual design standards, operating experience and fault statistics of the charging module.

[0095] Specifically, the system first determines the type of feature vector corresponding to the deviation value, and then retrieves the preset health threshold corresponding to that type for numerical comparison. For example, when the deviation value of the voltage feature vector is 18% and the preset health threshold is 15%, 18% > 15%, which means that the deviation value is higher than the preset health threshold. When the deviation value of the temperature feature vector is 15% and the preset health threshold is 20%, 15% < 20%, which means that the deviation value is not higher than the preset health threshold.

[0096] If the deviation value is higher than the preset health threshold, then step S108 is executed; if the deviation value is lower than the preset health threshold, it means that the current operating status of the charging module is within the healthy range.

[0097] S108. Call the preset dynamic twin model to simulate multi-scale feature vectors and output the fault evolution trend and diagnosis report.

[0098] When the deviation value is higher than the preset health threshold, the extracted multi-scale feature vectors are input into the preset dynamic twin model. The preset dynamic twin model simulates the operation of the charging module in the current state based on these multi-scale feature vectors. For example, when the temperature feature vector is continuously high, the aging speed of the capacitors inside the charging module and the heating of the resistors are simulated.

[0099] Furthermore, the preset dynamic twin model is a digital twin built based on the actual physical structure and operating principle of the charging module. It can accurately map the real operating state of the charging module and includes data such as the physical parameters of each component of the charging module and the fault evolution law.

[0100] During the simulation, the fault evolution rules in the preset dynamic twin model will predict the fault development trend of the charging module in the future based on multi-scale feature vectors, and generate a fault evolution trend and diagnostic report that includes the current abnormal location, fault risk level, and suggested handling measures.

[0101] This embodiment acquires multimodal big data, divides it into sub-data modules according to preset operating sample levels, and extracts the complete proportion, noise level, and coefficient of variation to provide a basis for data quality assessment. Based on the extracted complete proportion and noise level, a quality score is calculated for each sub-data module. Fault sensitivity is determined based on the coefficient of variation, and a weighted comprehensive priority is obtained to avoid homogenized input of low-quality data. After obtaining the comprehensive priority, the sub-data modules are input into a preset multi-scale feature extraction model in order of priority to accurately extract multi-scale feature vectors, further preventing low-quality data from diluting key features. The deviation value is obtained by comparing the fluctuation amplitude of the multi-scale feature vectors with a preset fluctuation benchmark. Combined with dynamic twin simulation and cross-validation, a fault evolution trend and diagnostic report are obtained. This not only filters high-quality, high-value data but also enhances feature accuracy and diagnostic reliability, improving the accuracy of graded diagnosis and fault prediction capabilities of charging modules.

[0102] The following provides a detailed description of the charging module hierarchical diagnostic system based on big data provided in this application. Please refer to [link / reference]. Figure 5 , Figure 5 One embodiment of the charging module hierarchical diagnostic system provided in this application includes: The first acquisition unit 501 is used to acquire multimodal big data of multiple charging modules in real time. The multimodal big data represents the data of each dimension when each charging module is in operation. The second acquisition unit 502 is used to divide multimodal big data into multiple sub-data modules according to a preset running sample level, and to acquire the complete proportion, noise level and coefficient of variation of each sub-data module. The determination unit 503 is used to calculate the data quality score of each sub-data module based on the integrity ratio and noise level, and to determine the fault sensitivity of each sub-data module based on the coefficient of variation. The calculation unit 504 is used to perform weighted calculation of data quality score and fault sensitivity through a preset weighted fusion algorithm to obtain the comprehensive priority of each sub-data module; Extraction unit 505 is used to input the corresponding sub-data modules into the preset multi-scale feature extraction model according to the order of comprehensive priority, and extract the multi-scale feature vectors in the sub-data modules through the preset multi-scale feature extraction model. The comparison unit 506 is used to compare the fluctuation amplitude of the multi-scale feature vector in time with the fluctuation envelope of the preset fluctuation benchmark, and obtain the deviation value of the current operating state of the charging module according to the proportion of the fluctuation amplitude exceeding the upper or lower envelope boundary of the fluctuation envelope. The judgment unit 507 is used to determine whether the deviation value is higher than the preset health threshold. Simulation unit 508 is used to call a preset dynamic twin model to simulate multi-scale feature vectors if the condition is met, and output the fault evolution trend and diagnosis report.

[0103] Optionally, the comparison unit 506 is specifically used for: The frequency of fluctuation exceeding the upper or lower envelope boundary of the fluctuation amplitude is determined based on the number of times the fluctuation amplitude exceeds the upper or lower envelope boundary of the fluctuation envelope. The over-limit intensity of the fluctuation amplitude is determined based on the difference between the fluctuation amplitude exceeding the upper or lower envelope boundary of the fluctuation envelope for each fluctuation. The deviation value of the current operating state of the charging module is obtained by combining the over-limit frequency and over-limit intensity with the preset weighting coefficient.

[0104] Optionally, it also includes a first update unit 509, specifically used for: The fluctuation envelope of the preset fluctuation benchmark is dynamically updated based on the frequency and intensity of the exceedance. The fluctuation envelope of the preset fluctuation benchmark is dynamically updated based on the frequency and intensity of the exceedance, including: Based on the over-limit frequency and over-limit intensity, and combined with the preset adjustment coefficient, the suggested adjustment amounts for the upper envelope boundary and the lower envelope boundary are calculated respectively. Determine whether the suggested adjustment amount exceeds the preset maximum allowable single adjustment threshold; If not, then dynamically update the upper and lower envelope boundaries according to the suggested adjustment amounts.

[0105] Optionally, a second update unit 510 is also included, specifically for: If so, the preset maximum allowable single adjustment threshold will be used as the suggested adjustment amount to dynamically update the upper and lower envelope boundaries.

[0106] Optionally, the second acquisition unit 502 is specifically used for: Based on the preset running sample levels, a cross-modal joint feature template is constructed for each level, wherein the cross-modal joint feature template includes each modal feature template and the cross-constraint relationship between different modalities; Synchronous time slicing is performed on multimodal big data to generate multiple cross-modal data units within the same time window, and the unimodal features and intermodal cross features in each cross-modal data unit are extracted; The single-modal features and inter-modal cross features are matched with each modal feature template and cross constraint relationship, and the matching degree is calculated. Based on the matching degree, cross-modal data units are divided into corresponding preset running sample levels, and cross-modal data units in different levels are aggregated to obtain multiple sub-data modules.

[0107] Optionally, a detection unit 511 may also be included, specifically for: Sort all cross-modal data units from highest to lowest matching degree; After sorting, check whether the difference between the matching degree of each cross-modal data unit and the matching degree of the previous cross-modal data unit is greater than or equal to the preset minimum discrimination margin. If so, then proceed with the step of dividing the cross-modal data units into the corresponding preset running sample levels based on the matching degree.

[0108] Optionally, the preset multi-scale feature extraction model includes a variational mode decomposition unit, a multi-dimensional computation unit, and an attention fusion unit; Multi-scale feature vectors are extracted from sub-data modules using a pre-defined multi-scale feature extraction model, including: The time-series signal in the sub-data module is decomposed by the variational mode decomposition unit to obtain multiple eigenmode components with different center frequencies. The time-domain statistics, frequency-domain energy spectrum, and joint time-frequency distribution of each intrinsic mode component are calculated using a multi-dimensional computing unit. The attention fusion unit fuses time-domain statistics, frequency-domain energy spectrum, and time-frequency joint distribution to generate multi-scale feature vectors.

[0109] For details on the implementation method, please refer to [link / reference]. Figures 1-4 Examples will not be described in detail here.

[0110] This application also provides a charging module hierarchical diagnostic device based on big data; please refer to [link / reference]. Figure 6 , Figure 6 One embodiment of the charging module classification diagnostic device provided in this application includes: Processor 601, memory 602, input / output unit 603, bus 604; The processor 601 is connected to the memory 602, the input / output unit 603, and the bus 604; The memory 602 stores a program, and the processor 601 calls the program to execute any of the above-mentioned charging module graded diagnostic methods.

[0111] This application also relates to a computer-readable storage medium on which a program is stored, which, when run on a computer, causes the computer to perform any of the above-mentioned charging module hierarchical diagnostic methods.

[0112] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0113] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between apparatuses or units through some interfaces, and may be electrical, mechanical, or other forms.

[0114] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0115] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0116] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A method for hierarchical diagnosis of charging modules based on big data, characterized in that, include: Real-time acquisition of multimodal big data from multiple charging modules, wherein the multimodal big data represents data from each dimension when the charging module is in operation; The multimodal big data is divided into multiple sub-data modules according to the preset running sample level, and the complete proportion, noise level and coefficient of variation of each sub-data module are obtained. The data quality score of each sub-data module is calculated based on the completeness ratio and the noise level, and the fault sensitivity of each sub-data module is determined based on the coefficient of variation. The data quality score and the fault sensitivity are weighted and calculated using a preset weighted fusion algorithm to obtain the comprehensive priority of each sub-data module; According to the overall priority order, the corresponding sub-data modules are input into the preset multi-scale feature extraction model, and the multi-scale feature vectors in the sub-data modules are extracted through the preset multi-scale feature extraction model. The fluctuation amplitude of the multi-scale feature vector in time is compared with the fluctuation envelope of the preset fluctuation benchmark, and the deviation value of the current operating state of the charging module is obtained according to the proportion by which the fluctuation amplitude exceeds the upper or lower envelope boundary of the fluctuation envelope. Determine whether the deviation value is higher than a preset health threshold; If so, a preset dynamic twin model is invoked to simulate the multi-scale feature vectors, and the fault evolution trend and diagnostic report are output.

2. The charging module hierarchical diagnostic method according to claim 1, characterized in that, The step of obtaining the deviation value of the current operating state of the charging module based on the proportion by which the fluctuation amplitude exceeds the upper or lower envelope boundary of the fluctuation envelope includes: The frequency of the fluctuation amplitude exceeding the upper or lower envelope boundary of the fluctuation envelope is determined based on the number of times the fluctuation amplitude exceeds the upper or lower envelope boundary of the fluctuation envelope. The over-limit intensity of the fluctuation amplitude is determined based on the difference between each fluctuation amplitude exceeding the upper or lower envelope boundary of the fluctuation envelope. The deviation value of the current operating state of the charging module is obtained by combining the over-limit frequency and the over-limit intensity with a preset weighting coefficient.

3. The charging module hierarchical diagnostic method according to claim 2, characterized in that, After obtaining the deviation value of the current operating state of the charging module based on the proportion by which the fluctuation amplitude exceeds the upper or lower envelope boundary of the fluctuation envelope, the charging module graded diagnosis method further includes: The fluctuation envelope of the preset fluctuation benchmark is dynamically updated based on the over-limit frequency and the over-limit intensity. The step of dynamically updating the fluctuation envelope of the preset fluctuation reference based on the exceeding frequency and the exceeding intensity includes: Based on the over-limit frequency and the over-limit intensity, and combined with the preset adjustment coefficient, the suggested adjustment amounts for the upper envelope boundary and the lower envelope boundary are calculated respectively. Determine whether the suggested adjustment amount is greater than the preset maximum allowable single adjustment threshold; If not, the upper envelope boundary and the lower envelope boundary are dynamically updated according to the suggested adjustment amount.

4. The charging module hierarchical diagnosis method according to claim 3, characterized in that, After determining whether the suggested adjustment amount is greater than the preset maximum allowable single adjustment threshold, the charging module graded diagnosis method further includes: If so, the preset maximum allowable single adjustment threshold is used as the suggested adjustment amount to dynamically update the upper envelope boundary and the lower envelope boundary.

5. The charging module hierarchical diagnostic method according to claim 1, characterized in that, The process of dividing the multimodal big data into multiple sub-data modules according to a preset running sample level includes: Based on the preset running sample level, a cross-modal joint feature template is constructed for each level, wherein the cross-modal joint feature template includes each modal feature template and the cross-constraint relationship between different modalities; The multimodal big data is synchronously sliced ​​to generate multiple cross-modal data units within the same time window, and the single-modal features and inter-modal cross features in each cross-modal data unit are extracted. The single-modal features and the inter-modal cross features are matched with the modal feature templates and the cross constraint relationships, respectively, and the matching degree is calculated. Based on the matching degree, the cross-modal data units are divided into the corresponding preset running sample levels, and the cross-modal data units in different levels are aggregated to obtain multiple sub-data modules.

6. The charging module hierarchical diagnostic method according to claim 5, characterized in that, After calculating the matching degree, the charging module graded diagnosis method further includes: Sort all the cross-modal data units from high to low according to the matching degree; After sorting, check whether the difference between the matching degree corresponding to each cross-modal data unit and the matching degree corresponding to the previous cross-modal data unit is greater than or equal to a preset minimum discrimination margin. If so, then the step of dividing the cross-modal data unit into the corresponding preset running sample level according to the matching degree is performed.

7. The method for graded diagnosis of charging modules according to any one of claims 1 to 6, characterized in that, The preset multi-scale feature extraction model includes a variational mode decomposition unit, a multi-dimensional computation unit, and an attention fusion unit. The step of extracting multi-scale feature vectors from the sub-data module using the preset multi-scale feature extraction model includes: The variational mode decomposition unit decomposes the time-series signal in the sub-data module to obtain multiple intrinsic mode components with different center frequencies. The time-domain statistics, frequency-domain energy spectrum, and joint time-frequency distribution of each intrinsic mode component are calculated by the multi-dimensional computing unit. The attention fusion unit fuses the time-domain statistics, the frequency-domain energy spectrum, and the time-frequency domain joint distribution to generate the multi-scale feature vector.

8. A hierarchical diagnostic system for charging modules based on big data, characterized in that, include: The first acquisition unit is used to acquire multimodal big data of multiple charging modules in real time, wherein the multimodal big data represents the data of each dimension when each charging module is in operation. The second acquisition unit is used to divide the multimodal big data into multiple sub-data modules according to a preset running sample level, and to acquire the complete proportion, noise level and coefficient of variation of each sub-data module. A determining unit is configured to calculate the data quality score of each sub-data module based on the completeness ratio and the noise level, and to determine the fault sensitivity of each sub-data module based on the coefficient of variation; The calculation unit is used to perform weighted calculation of the data quality score and the fault sensitivity through a preset weighted fusion algorithm to obtain the comprehensive priority of each sub-data module; An extraction unit is used to input the corresponding sub-data modules into a preset multi-scale feature extraction model according to the order of the comprehensive priority, and extract multi-scale feature vectors from the sub-data modules through the preset multi-scale feature extraction model. The comparison unit is used to compare the fluctuation amplitude of the multi-scale feature vector in time with the fluctuation envelope of the preset fluctuation benchmark, and obtain the deviation value of the current operating state of the charging module according to the proportion of the fluctuation amplitude exceeding the upper or lower envelope boundary of the fluctuation envelope. The judgment unit is used to determine whether the deviation value is higher than a preset health threshold. The simulation unit is used to call a preset dynamic twin model to simulate the multi-scale feature vector if the condition is met, and output the fault evolution trend and diagnosis report.

9. A charging module hierarchical diagnostic device based on big data, characterized in that, include: Processor, memory, input / output units, and bus; The processor is connected to the memory, the input / output unit, and the bus; The memory stores a program, which the processor calls to execute the charging module graded diagnostic method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium contains a program that, when executed on a computer, performs the charging module graded diagnostic method as described in any one of claims 1 to 7.

Images