Artificial intelligence-based battery swapping device anomaly identification method and device

Abstract

The application provides an artificial intelligence-based battery replacement equipment anomaly identification method and device, relates to the technical field of artificial intelligence, and can construct normalized features that reserve transient mutation features and fuse current-voltage coupling information by sliding window double-channel cooperative adaptive normalization processing of current and voltage monitoring data, so that subtle abnormal features are avoided from being weakened. Discriminative features are constructed in combination with fused time-frequency spectrum of normalized features, energy coupling and phase cooperation rules in the time-frequency joint domain can be mined, the contribution weight and representation ability of different features in the corresponding stage are quantified, high discriminative features that have time-frequency cooperation and stage frequency recognition degree are determined, and the false and missing report risks are reduced from the root. Finally, the abnormal score is quantified based on the high discriminative features, and the identification result is output, early discovery, accurate positioning and low false alarm of the battery replacement equipment anomaly can be realized, the sensitivity and robustness of the anomaly identification are improved, and the safe and efficient operation of the battery replacement process is ensured.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to an artificial intelligence-based method and apparatus for identifying anomalies in battery swapping equipment. Background Technology

[0002] With the development of the electric vehicle and electric bicycle industries, battery swapping has become an important energy replenishment solution due to its advantages such as high energy replenishment efficiency and low impact on the power grid. As the core hardware of this model, the reliability of battery swapping equipment directly affects operational safety and efficiency. However, during long-term, high-frequency operation, key components of battery swapping equipment, such as electrical connectors, mechanical locking mechanisms, and conveyor motors, are prone to malfunctions such as increased contact resistance, jamming, and stalling. If these anomalies are not identified in time, they may evolve from performance degradation into serious malfunctions, leading to battery replacement failures, equipment damage, and even safety accidents.

[0003] Currently, the industry mainly relies on simple threshold judgments or single-dimensional feature analysis of current and voltage parameters. However, the battery swapping process has obvious phased and non-stationary characteristics. The distribution characteristics of current and voltage parameters differ significantly in different stages of the preset battery swapping conditions (such as the docking stage, the power-on stage, and the charging stage). Existing technologies mostly use global fixed normalization or single-window normalization to process these parameters, which cannot adaptively adapt to the parameter distribution drift in different battery swapping stages. This can easily lead to the smoothing or amplification of subtle fault features (such as voltage spikes and current jitter caused by poor contact), reducing the sensitivity of anomaly identification.

[0004] Secondly, most existing technologies only extract the time-domain statistical features of parameters (such as mean and variance), neglecting the specific characteristic responses of typical anomalies (such as high-voltage arcs and connector oxidation) in the frequency domain during battery swapping. They also ignore the differences in the contribution of different frequency components to anomaly identification, failing to comprehensively and accurately characterize the essential features of various anomalies. This makes it difficult for existing technologies to accurately distinguish and identify various complex anomalies in battery swapping equipment, easily leading to false alarms and missed alarms, and failing to meet the actual needs of high-frequency, high-reliability operation of battery swapping equipment. Summary of the Invention

[0005] The purpose of this invention is to provide an artificial intelligence-based method and device for identifying anomalies in battery swapping equipment, which can achieve early detection, accurate location, and low false alarm of anomalies in battery swapping equipment, and significantly improve the sensitivity and robustness of anomaly identification.

[0006] In a first aspect, embodiments of the present invention provide an artificial intelligence-based method for anomaly identification of battery swapping equipment. The method includes: acquiring process monitoring data of the battery swapping equipment under preset battery swapping conditions; performing sliding window dual-channel collaborative adaptive normalization on the current and voltage parameters in the process monitoring data to construct normalized features; determining discriminative features of the process monitoring data based on the normalized features and the fused time-frequency spectrum corresponding to the normalized features; determining anomaly scores corresponding to the process monitoring data based on the discriminative features; and determining the anomaly identification result of the battery swapping equipment under the preset battery swapping conditions based on the anomaly scores.

[0007] In conjunction with the first aspect, this invention also provides a first implementation of the first aspect, wherein the step of performing sliding window dual-channel collaborative adaptive normalization on the current parameters and voltage parameters in the process monitoring data to construct normalized features includes: calculating a dual-channel adaptive scaling factor for the process monitoring data based on the local standard deviations of the current parameters and voltage parameters in the process monitoring data within a preset sliding window; determining the normalization denominators corresponding to the current parameters and voltage parameters based on the dual-channel adaptive scaling factor; and performing adaptive normalization on the current parameters and voltage parameters based on the normalization denominators to construct normalized features of the process monitoring data for the dual channels.

[0008] In conjunction with the first aspect, this embodiment of the invention also provides a second implementation of the first aspect, wherein the step of determining the discriminative features of process monitoring data based on normalized features and the fused time-frequency spectrum corresponding to the normalized features includes: determining the continuous wavelet transform complex coefficients corresponding to the current parameters and voltage parameters in the normalized features; constructing the fused time-frequency spectrum corresponding to the normalized features based on the energy information and phase coordination information of the continuous wavelet transform complex coefficients at each time-frequency point; determining the time-frequency energy moment, dual-channel coordination features, and transient anomaly features of the normalized features based on the fused time-frequency spectrum; and performing feature fusion on the time-frequency energy moment, dual-channel coordination features, and transient anomaly features to construct the discriminative features of the process monitoring data.

[0009] In conjunction with the first aspect, this embodiment of the invention also provides a third implementation of the first aspect, wherein the step of feature fusion of time-frequency energy moment, dual-channel collaborative features and transient anomaly features to construct discriminative features of process monitoring data includes: performing Gaussian kernel nonlinear weighting on time-frequency energy moment, dual-channel collaborative features and transient anomaly features to determine discriminative features of process monitoring data.

[0010] In conjunction with the first aspect, this embodiment of the invention also provides a fourth implementation of the first aspect, wherein the step of determining the abnormal score corresponding to the process monitoring data based on the discriminative feature includes: determining the feature deviation result of the discriminative feature compared with the normal benchmark value; normalizing the feature deviation result using a preset variance adjustment factor and sensitivity factor; and determining the abnormal score corresponding to the discriminative feature based on the normalized feature deviation result.

[0011] In conjunction with the first aspect, this embodiment of the invention also provides a fifth implementation of the first aspect, wherein the step of determining the time-frequency energy moment of the normalized feature based on the fused time-frequency spectrum includes: constructing adaptive weights based on the amplitude of the fused time-frequency spectrum and the local standard deviations corresponding to the current parameters and voltage parameters in the normalized feature, respectively; and constructing the time-frequency energy moment corresponding to the normalized feature by weighted aggregation of the energy distribution of the normalized feature in the scale dimension and the energy distribution in the time dimension based on the adaptive weights.

[0012] In conjunction with the first aspect, this embodiment of the invention also provides a sixth implementation of the first aspect, wherein the step of determining the dual-channel collaborative features of the normalized features based on the fused time-frequency spectrum includes: determining the joint energy magnitude of the current parameters and voltage parameters of the normalized features based on the complex coefficients of the continuous wavelet transform of the fused time-frequency spectrum; and performing target feature identification on the phase synchronization features of the current parameters and voltage parameters based on the joint energy magnitude to determine the dual-channel collaborative features corresponding to the normalized features.

[0013] In conjunction with the first aspect, this embodiment of the invention also provides a seventh implementation of the first aspect, wherein the step of determining the transient anomaly features of the normalized features based on the fusion time-frequency spectrum includes: determining the morphological mutation intensity index corresponding to the normalized features based on the second-order backward difference of the normalized features; weighting and aggregating the morphological mutation intensity index with the fusion time-frequency spectrum to establish a mutation correlation between the fusion time-frequency spectrum and local morphological mutations; and determining the transient anomaly features corresponding to the normalized features based on the mutation correlation.

[0014] In conjunction with the first aspect, this embodiment of the invention also provides an eighth implementation of the first aspect, wherein the step of determining the abnormal identification result of the battery swapping equipment under a preset battery swapping condition based on the abnormal score includes: adaptively mapping the abnormal score based on the preset historical abnormal score to determine the abnormal probability corresponding to the abnormal score; and determining the abnormal identification result corresponding to the battery swapping equipment based on the abnormal probability.

[0015] Secondly, embodiments of the present invention also provide an artificial intelligence-based anomaly identification device for battery swapping equipment, wherein the device includes: a data acquisition module for acquiring process monitoring data of the battery swapping equipment under preset battery swapping conditions; a preprocessing module for performing sliding window dual-channel collaborative adaptive normalization on the current parameters and voltage parameters in the process monitoring data to construct normalized features; a feature extraction module for determining discriminative features of the process monitoring data based on the normalized features and the fused time-frequency spectrum corresponding to the normalized features; an execution module for determining the anomaly score corresponding to the process monitoring data based on the discriminative features; and an output module for determining the anomaly identification result of the battery swapping equipment under the preset battery swapping conditions based on the anomaly score.

[0016] The embodiments of this invention bring the following beneficial effects: This invention provides an artificial intelligence-based method and apparatus for anomaly identification in battery swapping equipment. By performing sliding window dual-channel collaborative adaptive normalization on current and voltage monitoring data, local dynamic adaptation and coupling feature fusion of dual-channel data can be achieved. This constructs normalized features that retain local transient change characteristics and fuse current-voltage coupling information, preventing the weakening of transient features. Furthermore, based on the normalized features and their fused time-frequency spectrum, discriminative features are determined. This can be achieved by deeply mining the energy coupling relationship and phase coordination law of the current and voltage dual channels in the time-frequency joint domain. Combined with the stage-specific characteristics of the battery swapping operation, fault-sensitive feature sets are anchored in stages, quantifying the contribution weight and representational ability of different features in corresponding stages. This enables precise allocation and screening of differentiated contributions, ultimately determining highly discriminative features that combine time-frequency joint domain coordination and stage frequency identification. This overcomes the limitations of single-dimensional feature representation, achieving a comprehensive and accurate characterization of the essential features of various anomalies, effectively improving the distinguishability of anomaly features, and fundamentally reducing the risk of false alarms and missed alarms in the anomaly identification process. Ultimately, by quantifying highly discriminative features to obtain anomaly scores and thus determine the corresponding anomaly identification results, early detection, accurate location, and low false alarms of anomalies in battery swapping equipment can be achieved, significantly improving the sensitivity and robustness of anomaly identification in battery swapping equipment and ensuring the safe and efficient operation of the battery swapping process.

[0017] Other features and advantages of the invention will be set forth in the following description, or some features and advantages may be inferred from the description or determined without doubt, or may be learned by practicing the techniques described above.

[0018] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0019] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0020] Figure 1 A flowchart illustrating an artificial intelligence-based method for identifying anomalies in battery swapping equipment, provided as an embodiment of the present invention;

[0021] Figure 2 A flowchart of another AI-based method for identifying anomalies in battery swapping equipment provided in an embodiment of the present invention;

[0022] Figure 3 This is a schematic diagram of a preprocessing process for process monitoring data provided in an embodiment of the present invention;

[0023] Figure 4 This invention provides a flowchart for multi-source feature extraction and fusion.

[0024] Figure 5 A schematic diagram illustrating the construction process of a loss function according to an embodiment of the present invention;

[0025] Figure 6 A schematic diagram of the structure of an artificial intelligence-based battery swapping equipment anomaly identification device provided in an embodiment of the present invention;

[0026] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0028] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0029] This invention provides an artificial intelligence-based method and apparatus for identifying anomalies in battery swapping equipment, which can achieve early detection, accurate location, and low false alarm of anomalies in battery swapping equipment, significantly improving the sensitivity and robustness of anomaly identification.

[0030] To facilitate understanding, the present invention will first describe an artificial intelligence-based method for identifying anomalies in battery swapping equipment, as provided in an embodiment of the invention. (Refer to...) Figure 1 The method includes the following steps:

[0031] Step S102: Obtain process monitoring data of the battery swapping equipment under preset battery swapping conditions.

[0032] The preset battery swapping operating conditions refer to the entire process of the battery swapping equipment completing a standard battery replacement, including the docking phase (mechanical docking of the battery and connector), the power-on phase (electrical circuit conduction), and the charging phase (constant current / constant voltage charging of the battery). The process monitoring data is multi-dimensional time-series data for the entire operating condition, including current parameters (transport motor current, charging circuit current) and voltage parameters (connector terminal voltage, battery terminal voltage), and can also be expanded to include auxiliary monitoring data such as position, temperature, and pressure.

[0033] Step S104: Perform sliding window dual-channel collaborative adaptive normalization on the current and voltage parameters in the process monitoring data to construct normalized features.

[0034] Conventional normalization methods typically perform global maximum-minimum normalization on current and voltage sequences separately, neglecting differences in the local dynamic range of the signal and the coupling relationship between the two channels. This leads to a weakening of key transient change features during the normalization process. This invention abandons a single global scale and adapts local signal features through a sliding window, ensuring that parameter features at each stage are accurately preserved. Combined with dual-channel collaborative constraints, it can identify coupling anomalies of current jitter and voltage spikes, ultimately constructing a normalized feature that retains transient change information and integrates current-voltage coupling anomaly features.

[0035] Step S106: Based on the normalized features and the fused time-spectrum corresponding to the normalized features, determine the discriminative features of the process monitoring data.

[0036] Existing technologies process individual channel signals individually or simply combine them, resulting in insufficient feature discriminative power. This invention constructs a current-voltage time-frequency joint matrix to mine the energy coupling and phase coordination characteristics of dual-channel signals in the time-frequency domain, capturing latent anomaly correlation information. Furthermore, by combining the phased differences of battery swapping, power-on, and charging, fault-sensitive frequency ranges under each operating condition can be anchored in stages to obtain discriminative features that combine time-frequency joint coordination, phase adaptability, and strong fault differentiation, effectively solving the problem of insufficient feature discriminative power in existing technologies.

[0037] Step S108: Based on discriminative features, determine the anomaly score corresponding to the process monitoring data.

[0038] Anomaly scores can be quantified using an anomaly detection model to measure discriminative features. The scores range from 0 to 1, with a score closer to 1 indicating a higher degree of anomaly in the current battery swapping operation, and a score closer to 0 indicating a more normal operation.

[0039] Step S110: Determine the anomaly identification result of the battery swapping equipment under the preset battery swapping conditions based on the anomaly score.

[0040] Based on the anomaly score, the final conclusion of anomaly identification can be determined by combining the classification judgment rules, such as including normal operating conditions, minor anomaly warnings, moderate anomaly alarms, and severe anomaly faults. Furthermore, the final anomaly identification result can be determined by associating specific fault types (such as poor connector contact or stalled conveyor motor) with faulty components.

[0041] In summary, this invention, through sliding window dual-channel collaborative adaptive normalization of current and voltage monitoring data, achieves local dynamic adaptation and coupling feature fusion of dual-channel data. It constructs normalized features that retain local transient change characteristics while fusing current-voltage coupling information, thus preventing the weakening of transient features. Furthermore, based on the normalized features and their fused time-frequency spectrum, discriminative features are determined. This involves in-depth mining of the energy coupling relationship and phase coordination law of the current and voltage dual channels in the time-frequency joint domain. Combined with the phased characteristics of the battery swapping operation, fault-sensitive feature sets are anchored in stages, quantifying the contribution weight and characterization ability of different features in corresponding stages. This achieves precise allocation and screening of differentiated contributions, ultimately determining highly discriminative features that combine time-frequency joint domain coordination with stage frequency identification. This overcomes the limitations of single-dimensional feature representation, achieving a comprehensive and accurate characterization of the essential features of various anomalies, effectively improving the distinguishability of anomaly features, and fundamentally reducing the risk of false alarms and missed alarms in the anomaly identification process. Ultimately, by quantifying highly discriminative features to obtain anomaly scores and thus determine the corresponding anomaly identification results, early detection, accurate location, and low false alarms of anomalies in battery swapping equipment can be achieved, significantly improving the sensitivity and robustness of anomaly identification in battery swapping equipment and ensuring the safe and efficient operation of the battery swapping process.

[0042] Furthermore, based on the above embodiments, this invention also provides another method for identifying anomalies in battery swapping equipment based on artificial intelligence, referring to... Figure 2 The method includes the following steps:

[0043] Step S202: Obtain process monitoring data of the battery swapping equipment under preset battery swapping conditions.

[0044] Based on the above embodiments, high-precision current and voltage sensors installed on key electrical circuits of the battery swapping equipment can synchronously collect raw current and voltage signals during the battery swapping process at a fixed sampling frequency to obtain high-quality monitoring data that comprehensively reflects the equipment's operating status. Each battery swapping operation is considered an independent monitoring sample, and data acquisition can begin from the issuance of the battery swapping command and continue until the battery is locked and the system enters standby mode, ensuring coverage of the complete operation process, including battery unlocking, handling, alignment, connector insertion, and locking.

[0045] Furthermore, each sample can be recorded as a dual-channel time series with 8192 time points, one channel being the current sequence and the other being the voltage sequence, thus forming the original monitoring sample.

[0046] Step S204: Perform sliding window dual-channel collaborative adaptive normalization on the current and voltage parameters in the process monitoring data to construct normalized features.

[0047] The current and voltage data obtained from the monitoring of the battery swapping equipment constitute a high-dimensional dual-channel time series with significant differences in dimensions and numerical scales. At the same time, it contains rich dynamic features from steady state to transient change. Conventional methods usually perform global maximum and minimum normalization on the current and voltage series separately, without considering the coupling correlation between the dual-channel signals or the differences in the dynamic range of different operating stages within the series. This can easily lead to the weakening of key transient change features during the normalization process.

[0048] This invention employs an adaptive dual-channel collaborative normalization method based on sliding window statistics to determine the corresponding normalization features. Specifically, the dual-channel adaptive scaling factor of the process monitoring data is calculated based on the local standard deviations of the current and voltage parameters within a preset sliding window. Based on the dual-channel adaptive scaling factor, the normalization denominators for the current and voltage parameters are determined. Based on the normalization denominators, adaptive normalization is performed on the current and voltage parameters respectively, constructing the normalized features of the process monitoring data for the dual channels.

[0049] In practical implementation, the local statistics of the current and voltage sequences of the original samples within the sliding window can be calculated separately. An adaptive scaling factor based on the ratio of the local standard deviations of the two channels is then used to dynamically adjust the denominator used for normalization of each data point. This approach adapts to the local dynamic range of the signal while also considering the normalization of inter-channel coupling relationships, as expressed below:

[0050]

[0051]

[0052] In the formula, Represents the normalized current sequence In the The values ​​at each time point are current values ​​obtained through sliding window dual-channel collaborative adaptive normalization. Represents the normalized voltage sequence In the The values ​​at each time point are voltage values ​​obtained through sliding window dual-channel collaborative adaptive normalization. This represents a point-in-time index, with a value range of [value range missing]. . Indicates the original current sequence at the 1st... The sampled values ​​at each time point are directly measured by the current sensor; Indicates the original voltage sequence at the 1st... The sampled values ​​at each time point are directly measured by the voltage sensor.

[0053] Indicates the current sequence at the th The local arithmetic mean within a sliding window is used to eliminate the DC offset component within that window; Indicates the voltage sequence at the 1st The local arithmetic mean within a sliding window is used to eliminate the DC offset component within that window.

[0054] Indicates the current sequence at the 1st The local standard deviation within a sliding window characterizes the fluctuation intensity of the current signal within that window; Indicates the voltage sequence at the 1st The local standard deviation within a sliding window characterizes the fluctuation intensity of the voltage signal within that window.

[0055] This represents the step size of the sliding window, controlling the interval distance at which the window moves. An example value is 256. This represents the floor function; This function represents the larger of the two values. This represents a very small positive integer, used to prevent the denominator from being zero and to ensure numerical stability. Examples of its values ​​are: .

[0056] This represents the coupling adjustment coefficient, used to control the degree of influence of fluctuations in another channel on the normalized denominator of this channel. An example value is 0.05.

[0057] Step S206: Determine the complex coefficients of the continuous wavelet transform corresponding to the current parameter and voltage parameter in the normalized features.

[0058] Step S208: Based on the energy information and phase coordination information of the complex coefficients of the continuous wavelet transform at each time and frequency point, construct the fused time spectrum corresponding to the normalized features.

[0059] Regarding steps S206 and S208, this embodiment of the invention further incorporates a joint time-frequency analysis method based on complex Morlet wavelets to preprocess the original monitoring data, thereby preserving key transient features and generating a more discriminative fused time-frequency representation. In specific implementation, complex Morlet wavelets can be used as the mother wavelet to perform continuous wavelet transforms on the normalized current and voltage sequences based on this mother wavelet, determining... and (Complex coefficients). See reference. Figure 3 The diagram illustrates the preprocessing procedure for process monitoring data according to an embodiment of the present invention. Correspondingly, by performing continuous wavelet transforms on the normalized current sequence and voltage sequence respectively, complex coefficients of the continuous wavelet transform for two channels are obtained. The complex coefficients of the continuous wavelet transform are time-frequency characteristic vectors of the signal at a specific time and frequency. Based on the added phase characteristics, abnormal operating conditions with the same amplitude but different timing / phase (such as normal impact vs. haphazard impact) can be distinguished. The degree of coordination between the parameters of the two channels can be quantified by calculating the cross-correlation and phase difference of the complex coefficients of the two channels. For example, it can be characterized as: , representing the normalized current sequence In scale ,time Continuous wavelet transform complex coefficients at; , representing the normalized voltage sequence In scale ,time The continuous wavelet transform complex coefficients at each time frequency point can be calculated. Furthermore, the coherence power of the continuous wavelet transform complex coefficients at each time frequency point can be calculated, and the signal energy information of the two channels and the phase consistency information between them can be fused to obtain the fused time-frequency spectrum. Specifically, this can be expressed by the following formula:

[0060]

[0061] In the formula, Represents the spectrum during fusion In scale ,time The amplitude at that point is the output after coherent power fusion, which can simultaneously characterize the energy and phase coordination change characteristics of the two channels. express The modulus characterizes the current signal at time and frequency points. The amount of energy at that location; express The modulus characterizes the voltage signal at time and frequency points. The amount of energy at that location. express The phase angle; express The phase angle; The scale index is inversely proportional to the frequency of the analysis, and its value range is [value range missing]. ; This represents a point-in-time index, with a value range of [value range missing]. ; This represents the total number of scales, with an example value of 64.

[0062] It should be noted that, The term characterizes the current and voltage channels at time and frequency points. The cosine of the phase difference has a range of values. This is used to quantify the phase synchronization degree (i.e., phase coordination information) of the two channel signals. A value of 1 indicates complete in-phase synchronization, enhancing the characteristics of phase coordination changes. In the above formula, through... Item and Multiplying terms yields It can simultaneously fuse the joint energy information of the two channels and their instantaneous phase synchronization relationship, thereby obtaining a fused time spectrum that can emphasize in-phase changes to characterize the above coherent power fusion.

[0063] Step S210: Based on the fused time-frequency spectrum, determine the time-frequency energy moment of the normalized features, the dual-channel collaborative features, and the transient anomaly features.

[0064] The fused time-frequency spectrum and normalized current and voltage sequences obtained after preprocessing contain key information about the operating status of the battery swapping equipment. However, directly using them for anomaly identification faces challenges due to their high dimensionality and insufficient discriminative power. Conventional feature extraction methods typically calculate time-domain statistics or frequency-domain energy separately, failing to fully explore the dynamic patterns and dual-channel collaborative changes in the joint time-frequency domain, and ignoring operating conditions such as load mutations during battery swapping, resulting in extracted features that are insensitive to early anomalies. This invention constructs a multi-level feature extraction module to extract feature vectors (i.e., discriminative features) with stronger discriminative power and closer to the operating mechanism of the battery swapping equipment from three perspectives: time-frequency energy distribution (time-frequency energy moment), dual-channel phase synchronization (dual-channel collaborative features), and sequence local morphology and energy correlation (used to determine the transient anomaly features corresponding to the correlation). Figure 4 The flowchart of multi-source feature extraction and fusion corresponding to this process is shown. This flowchart is used to illustrate the construction process of time-frequency energy moment, dual-channel collaborative features and transient anomaly features (determined based on the correlation strength between morphological mutation and energy).

[0065] In practical implementation, the above features can be determined through the following steps:

[0066] 1) Based on the amplitude of the fused time-frequency spectrum and the local standard deviations of the current and voltage parameters in the normalized features, an adaptive weight is constructed. Based on the adaptive weight, the energy distribution of the normalized features in the scale dimension and the energy distribution in the time dimension are weighted and aggregated to construct the time-frequency energy moment corresponding to the normalized features.

[0067] In specific implementation, this invention uses the ratio of the fused time-frequency amplitude and the local standard deviation of current and voltage as adaptive weights, and combines the central moments calculation of the scale and time dimensions to perform weighted aggregation of the energy distribution within each time-frequency block (e.g., s corresponds to the energy distribution in the scale dimension, and t corresponds to the energy distribution in the time dimension), thereby quantifying the energy distribution moment characteristics of the time-frequency block and highlighting regions of concentrated or abrupt energy changes, as expressed as:

[0068]

[0069] In the formula, The time-frequency energy moment characteristic is represented in the first... The time period and the first The eigenvalues ​​on each scale segment are used to quantify the moment characteristics of the energy distribution within that time-frequency block. The larger the value, the more the energy distribution deviates from a uniform state. This represents a time period index, with a value range of [value range missing]. ; This represents the total number of time periods; an example value is 32. This represents the scale segment index, with a value range of [value range missing]. ; This indicates the total number of scale segments, with an example value of 16.

[0070] Indicates the first The set of scale indices contained in each scale segment is used to define the range of feature computation in the scale dimension, defined by preset scale segment boundaries. and Decisions, for example, for uniform division, , ; Indicates the first The minimum scale index of each scale segment is the preset lower boundary of the scale segment; Indicates the first The maximum scale index of each scale segment is the preset upper boundary of the scale segment.

[0071] Indicates the first The set of time indices contained in each time period is used to define the range of feature calculation in the time dimension, and is defined by preset time period boundaries. and Decide; Indicates the first The minimum time index of a time period is the preset left boundary of the time period; Indicates the first The maximum time index for a given time period is the preset right boundary of that time period. Indicates the first The local standard deviation of the current series at the nth time point is defined as equal to the standard deviation of the current series at the nth time point. Local standard deviation within a sliding window This characterizes the fluctuation intensity of the current signal within the local window at that time point; Indicates the first The local standard deviation of the voltage series at the nth time point is defined as equal to the voltage series at the nth time point. Local standard deviation within a sliding window It represents the fluctuation intensity of the voltage signal within the local window at that time point. This represents the smoothing constant, used to prevent the denominator from being zero and to control the sensitivity of the weighting function to the ratio of fluctuation intensity. An example value is shown below. . This represents the energy amplification index, which contributes to the spectral amplitude during nonlinear enhancement fusion. An example value is 2.

[0072] Indicates the first The arithmetic mean of all scale indices within each scale segment is used to calculate the central moments of the scale dimension. Indicates the first The arithmetic mean of all time indices within a time period is used to calculate the central moment of the time dimension. The normalization factor representing the scale dimension is calculated by taking the standard deviation of all scale indices, and is expressed as follows: ; This represents the mean of all scale indices; The normalization factor representing the time dimension is calculated by taking the standard deviation of all time indices, and is expressed as follows: ; This represents the mean of all time indices.

[0073] 2) Based on the complex coefficients of the continuous wavelet transform of the fused time spectrum, determine the joint energy of the current and voltage parameters of the normalized feature; based on the joint energy, perform target feature identification on the phase synchronization features of the current and voltage parameters, and determine the dual-channel collaborative features corresponding to the normalized feature.

[0074] The dual-channel coordination feature is used to reflect the coordinated changes in current and voltage waveforms. The joint energy is determined by energy weighting the product of the complex coefficients of the continuous wavelet transforms of current and voltage. The exponential average of the phase difference is calculated, and then the modulus is taken to obtain the phase-locked value. This quantifies the degree of phase synchronization of the dual-channel signals within a specific time-frequency block, reflecting the coordinated changes in current and voltage waveforms, and is expressed as:

[0075]

[0076] In the formula, The dual-channel cooperative feature is represented in the first... The time period and the first The eigenvalues ​​at each scale range are used to quantify the phase synchronization degree of the dual-channel signals within that time-frequency block, with a value range of [value range missing]. A larger value indicates stronger phase synchronization. This represents the L2 norm, used to implement modulo operations on complex numbers; Represents the imaginary unit, satisfying ; This represents the adjustment index of the energy product, used to control the intensity of the influence of the dual-channel energy product term on the phase synchronization calculation. An example value is 0.5.

[0077] It should be noted that, The term is used as an energy weight to measure the current and voltage signals at time and frequency points. The magnitude of the joint energy at a given point serves as a criterion for assessing the reliability and importance of the phase difference information at that point, thereby ensuring that phase synchronization in the high-energy region contributes more significantly to the overall dual-channel coordinated eigenvalues. It should also be noted that... The term maps the phase difference at each time frequency point to a unit complex vector, the direction of which is determined by the phase difference of all time frequency points.

[0078] In summary, if the phase difference between two signals remains constant over a period of time (i.e., "locked"), then the average of the unit complex vectors representing these phase differences in the complex plane will have a large resultant vector; conversely, if the phase difference varies randomly, the average of these vectors will be close to zero. Based on this, the resultant vectors obtained by modulo complex calculations can be... As a phase-locking value, this value is calculated on a time-frequency local block based on energy weighting. It can extract the stability characteristics of the coordinated changes in current and voltage waveforms to quantify the degree of phase synchronization or the stability of the phase relationship between two signals.

[0079] 3) Based on the second-order backward difference of the normalized features, determine the morphological mutation intensity index corresponding to the normalized features; weight and aggregate the morphological mutation intensity index with the fusion spectrum to establish the mutation correlation between the fusion spectrum and local morphological mutations; based on the mutation correlation, determine the transient abnormal features corresponding to the normalized features.

[0080] The intensity of morphological abrupt changes can be inferred by using the second-order backward difference product of the normalized current and voltage sequences as an indicator. This product is then weighted and aggregated with the average energy of the fused time-frequency spectrum at the corresponding time point within a specific scale range to quantify the correlation between local waveform morphological abrupt changes and time-frequency energy variations, thereby revealing the characteristics of transient anomalies. This can be expressed as follows:

[0081]

[0082] In the formula, The correlation strength feature is represented in the first... The time period and the first The eigenvalues ​​on each scale segment are used to quantify the correlation between the intensity of local sequence morphological mutations and the time-frequency average energy. The larger the value, the more significant the energy change accompanied by the morphological mutation, thus effectively distinguishing the characteristics of normal operating condition mutations from transient anomalies caused by faults. Indicates the normalized current sequence at the th The second-order backward difference at each time point is used to approximate the local curvature of the current waveform at that time point to characterize the intensity of morphological abrupt changes. The calculation method is expressed as follows: ; Represents the normalized current sequence In the The values ​​at each point in time. Represents the normalized current sequence In the The values ​​at each point in time. Represents the normalized current sequence In the Values ​​at specific points in time; Indicates the normalized voltage sequence at the th The second-order backward difference at each time point is used to approximate the local curvature of the voltage waveform at that time point, and the calculation method is expressed as follows: ; Represents the normalized voltage sequence In the Values ​​at each point in time. Represents the normalized voltage sequence In the The values ​​at each point in time. Represents the normalized voltage sequence In the Values ​​at specific points in time; This represents the adjustment index of the morphological mutation product term, used to nonlinearly amplify the effect of significant morphological mutations; an example value is 0.8. Represents a set The number of elements in the middle, i.e., the first element. The total number of scales contained in each scale segment.

[0083] It should be noted that, Item passed in the first At the point in time, the first The arithmetic mean of the spectral amplitudes during fusion across each scale segment represents the value at that specific time point. Above, the joint average energy level of current and voltage signals within the frequency range of interest.

[0084] Step S212: Feature fusion is performed on time-frequency energy moment, dual-channel collaborative features and transient anomaly features to construct discriminative features of process monitoring data.

[0085] In one implementation, Gaussian kernel nonlinear weighting can be applied to time-frequency energy moments, dual-channel cooperative features, and transient anomaly features to determine the discriminative features of process monitoring data. Specifically, the time-frequency energy moment features, dual-channel cooperative features, and correlation strength features can be nonlinearly fused into a unified feature representation by applying power-law adjustment and Gaussian kernel weighting to the three types of feature values ​​respectively, and then summing them according to preset fusion coefficients. This enhances the discriminative power of the features and suppresses noise interference, as expressed below:

[0086]

[0087] In the formula, Indicates that it is located at the th The time period and the first The fused eigenvalues ​​across each scale segment are a unified feature representation after Gaussian kernel nonlinear weighted fusion. This represents the feature type index, with a value range of 100. These correspond to the time-frequency energy moment characteristics, dual-channel synergistic characteristics, and correlation strength characteristics, respectively.

[0088] Indicates the first The fusion coefficient of a class feature is used to adjust the contribution weight of that class feature in the overall fusion result, and satisfies the following conditions: Example of a value: , , ; Indicates the first The power-law scaling parameter for the class features is used to non-linearly scale the feature values ​​of that class. An example value is shown below. , , ; Indicates the first The global mean of a class feature is a statistic calculated from all normal sample data during the training phase. It is used to locate the center of the Gaussian kernel, and its calculation method is expressed as follows: ; Indicates the first The bandwidth parameter for the class of features controls the width of the Gaussian kernel to determine the rate of weighted decay, and is set based on the range of that class of features on the training set, for example... .

[0089] Represents the natural exponential function; Indicates the first Class features in the first The time period and the first Values ​​across each scale segment The range of values ​​is , respectively corresponding , and .

[0090] Step S214: Based on discriminative features, determine the anomaly score corresponding to the process monitoring data.

[0091] Furthermore, simple classifiers struggle to fully capture the complex nonlinear relationships between features and the time-scale characteristics of battery swapping equipment operation. Conventional anomaly detection methods often ignore the differences in anomaly contributions from different operating frequency regions and fail to consider adaptive adjustments based on historical states, easily leading to false alarms or missed detections. Building upon the above steps, this embodiment of the invention further improves the accuracy and robustness of anomaly detection by calculating anomaly scores, adaptive decision output, and combining weights based on operating phases and adaptive thresholds. Specifically, the deviation of the fused feature value (discriminative feature) of each time-frequency block from the normal baseline value of the corresponding time period can be calculated to determine the feature deviation result. Further normalization is performed using the variance adjustment factor of that time period and the sensitivity factor of the scale segment. The normalized deviation of all time-frequency blocks can be averaged and nonlinearly amplified to obtain an anomaly score, which more accurately reflects the degree of anomaly in the sample, expressed as:

[0092]

[0093] In the formula, The calculated anomaly score is a scalar; the larger the value, the more likely the current sample is to be in an anomalous state. Indicates the first The normal baseline value for each time period is a statistic calculated from the fused features of normal samples during the training phase. It represents the typical feature level of that time period under normal conditions, and is calculated as follows: ; Indicates the normal sample in the first... The time period and the first Fusion feature values ​​across various scale ranges.

[0094] Indicates the first The variance adjustment factor for each time period is the variance of the fused features of normal samples within that time period, used to normalize the inherent volatility of different time periods. The calculation method is expressed as follows:

[0095]

[0096] Indicates the first The sensitivity factor for each scale segment, related to the scale index, is used to assign different anomaly detection sensitivities to different scale components. The calculation method is expressed as follows: ,when hour, . This represents the amplification index, used to enhance the contribution of nonlinear enhancement to anomalies that deviate significantly from the normal baseline. An example value is 1.5.

[0097] Step S216: Determine the anomaly identification result of the battery swapping equipment under the preset battery swapping conditions based on the anomaly score.

[0098] This involves adaptively mapping anomaly scores based on preset historical anomaly scores to determine the anomaly probability corresponding to each score. Then, based on this anomaly probability, the anomaly identification result for the battery swapping equipment is determined. Specifically, the current anomaly score can be input into a Sigmoid function with the moving average and variance of historical anomaly scores as adaptive parameters, mapping the anomaly score to anomaly probability. This simulates the decision-making process of maintenance experts based on historical experience, improving the stability of anomaly identification results and adapting to the slow drift of equipment status, as expressed below:

[0099]

[0100] In the formula, This represents the probability that a sample is identified as an anomaly, with a value range of [value range missing]. The closer the value is to 1, the higher the probability of an anomaly; the closer it is to 0, the higher the probability of normality. By combining the Sigmoid function with historical state adaptation, it can adapt to the slow drift of device state. The baseline offset parameter is a trainable parameter used to adjust the baseline level of the outlier score threshold. The historical influence coefficient is a trainable parameter used to control the influence of the historical outlier score moving average on the threshold. This represents the basic scale parameter, which is a trainable parameter used to adjust the scale of the Sigmoid function and affects the sensitivity of the probability output. The historical variance influence coefficient is a trainable parameter used to control the degree of influence of historical outlier score variance on the scale. Represents the number of samples in the historical sample queue. The anomaly score of each sample, used to provide a reference benchmark for state adaptation, is calculated from historical monitoring data and stored in a fixed-length queue for adaptive adjustment. Represents a historical abnormal score sequence, i.e., containing The set of abnormal scores, including those in the range; This represents the historical sample index, with a value range of [value range missing]. ; This indicates the number of samples used for historical reference, i.e., the length of the historical queue, with an example value of 100. This represents the variance function, used to calculate the variance of a historical outlier score sequence.

[0101] Furthermore, the system can make a final judgment based on this anomaly probability value. For example, a decision threshold (e.g., 0.5) can be set. When the anomaly probability is greater than this threshold, it is determined that there is an anomaly in the current battery swapping process, triggering an alarm and prompting maintenance personnel to conduct an inspection; otherwise, it is determined to be normal. Furthermore, this embodiment of the invention can also operate continuously, constantly incorporating the identification results of new samples into the historical state to adaptively adjust the decision threshold for subsequent identifications, thereby dynamically adapting to the slow changes in equipment performance and achieving continuous, stable, and accurate anomaly identification and early warning functions for battery swapping equipment.

[0102] Furthermore, the trained model can be deployed in an actual battery swapping equipment monitoring system for online or offline anomaly identification to determine the aforementioned anomaly probabilities. The model can be trained by defining a loss function. Conventional cross-entropy loss or mean squared error loss only focuses on the degree of matching between the predicted probability and the label, failing to fully utilize the extracted intermediate information and effectively modeling the prevalent characteristics of battery swapping equipment anomaly data—that normal samples constitute the vast majority while anomaly samples are scarce and variable—as well as the differences in the contribution of different operational stages to anomalies.

[0103] This invention employs a dual-supervised loss function combining dynamic hard example mining and normal state spectrum consistency constraints. This not only forces the model to classify correctly but also mines sample segments that are difficult for the current model to judge and constrains the state feature distribution of normal samples to remain compact and consistent across different operating stages. This improves the model's generalization ability and robustness in scenarios with imbalanced data and changing operating conditions. (Refer to...) Figure 5 The diagram illustrates the process of constructing the loss function.

[0104] Specifically, for dynamic hard case mining loss based on time period and scale sensitivity:

[0105] By calculating the weighted cross-entropy loss and introducing prior importance weights to amplify the loss contribution of difficult-to-classify samples, the model is forced to focus on and improve its ability to distinguish ambiguous samples. The prior importance weights integrate the prior importance of time periods and the sensitivity factor of scale segments, making the loss function more consistent with the operating mechanism of battery swapping equipment, as expressed as:

[0106]

[0107] In the formula, This represents the dynamic hard example mining loss. It is a scalar, and the smaller the value, the better the model's classification performance and the stronger its ability to distinguish difficult-to-classify samples. This represents the total number of samples in the training batch, controlling the range of the average loss. An example value is 32. Indicates the first The probability that a sample is identified as an anomaly; This represents the sample index within the batch, used to iterate through all samples in the batch, with a value range of [value missing]. ; Indicates the first The prior importance weights for each time period are set based on prior knowledge of the battery swapping equipment or historical operational statistics. These weights are used to emphasize critical operational phases and satisfy the following conditions: The constraints, with examples of values, are as follows: during critical stages such as battery locking and connector insertion / removal. During other stable operating phases ; This represents the anomaly category compensation factor, used to alleviate the problem of imbalance between the number of normal and abnormal samples in the training data. It is usually set to the reciprocal of the ratio of the number of normal samples to the number of abnormal samples, with an example value of 5.0. This indicates an indicator function; the function value is 1 if the condition within the parentheses is true, and 0 otherwise. Indicates the first The true category labels for each sample are obtained through manual annotation or historical fault records. Represents an abnormal sample. This represents a normal sample; This represents a logarithmic function, with the default base being the natural constant. This represents the difficulty sample mining intensity coefficient, which controls the degree to which the dynamic weight term amplifies the difficult-to-classify samples. The example value is 0.1.

[0108] It should be noted that, The term is used as a dynamic weight term when When it is close to 0.5, it becomes difficult to judge. Approaching 0, its reciprocal causes the dynamic weights to increase dramatically, thus applying a stronger gradient signal to the difficult-to-classify sample in the loss calculation.

[0109] Furthermore, regarding the consistency constraint loss for the normal state spectrum:

[0110] By calculating the deviation of the fused feature values ​​of normal samples from the normal baseline in the corresponding time period, and applying kurtosis constraints and proportional consistency penalties, the feature distribution of all normal samples is forced to remain compact and consistent with preset statistical characteristics, thereby enhancing the model's ability to model the internal consistency of the normal state. This can be expressed as:

[0111]

[0112] In the formula, The loss represents the consistency constraint loss of the normal state spectrum. It is a scalar. The smaller the value, the more consistent the feature distribution of the normal samples is and the more it conforms to the preset statistical characteristics. This represents the set of indices of all normal samples in the current training batch; Represents a set The number of elements in the batch, i.e., the total number of normal samples in the current batch; This represents the index of a normal sample, used for traversing the collection. All normal samples; Indicates the first The normal sample at the ... The time period and the first Fusion feature values ​​across each scale segment; Indicates the first The normal baseline values ​​for a given time period are used as learnable model parameters for optimization during the training phase; these are trainable parameters. This represents the weighting coefficient of the proportional consistency penalty term, used to balance the relative strength of the kurtosis constraint term and the proportional consistency penalty term in the loss function. An example value is 0.01. Indicates the first Variance adjustment factor for each time period Indicates the first Sensitivity factor for each scale segment.

[0113] It should be noted that the loss function consists of two parts, the first part... This is a strong kurtosis constraint term. Using a fourth-power penalty can more severely suppress the situation where the feature values ​​of normal samples deviate significantly from the baseline, resulting in a more concentrated distribution. (Part Two) It is a proportional consistency penalty term, which aims to constrain the ratio of the squared feature offset of each normal sample to the overall variance of the time period to maintain a relatively stable ratio, thereby stabilizing the shape of the feature distribution of normal samples.

[0114] Furthermore, the above-mentioned dynamic hard example mining loss and the above-mentioned normal state spectrum consistency constraint loss can be weighted and summed to obtain the total loss function for model training. Through joint optimization, the model is forced to learn both accurate classification boundaries and feature representations that are highly discriminative and conform to physical laws.

[0115] The total loss function is expressed as:

[0116]

[0117] In the formula, This represents the total loss function, which is a scalar. The backpropagation algorithm minimizes this value to optimize all parameters of the model. Indicates spectral consistency constraint loss The balance coefficient is an adjustable hyperparameter used to control the strength of the normal state consistency constraint in the total loss. An example value is 0.5.

[0118] In summary, a pre-built anomaly identification architecture (e.g., feature fusion, anomaly score calculation, and adaptive decision output) and a total loss function can be used for model training and parameter updates. Specifically, training and validation sets can be constructed based on the collected data to optimize all trainable parameters in the model in an end-to-end manner. A large dataset can be constructed by collecting numerous battery swapping process samples covering different equipment, operating conditions, and health states during long-term actual operation. Data annotation can be performed manually by domain experts using equipment operation logs, maintenance records, and subsequent fault diagnosis reports. The annotation categories are divided into two main categories: "normal" and "abnormal." "Normal" samples refer to monitoring data showing that the battery swapping equipment completed battery replacement smoothly without any faults throughout the process. "Abnormal" samples are further subdivided according to common fault modes, mainly including electrical connection anomalies (e.g., excessive connector contact resistance, loose connections), mechanical action anomalies (e.g., locking mechanism jamming, transport motor stall), and control timing anomalies. Each abnormal sample is not only labeled with an anomaly category but also associated with a specific anomaly stage or type. Furthermore, all collected samples and their annotation information can be divided into three mutually exclusive sets: a training set, a validation set, and a test set. The training set is used for learning model parameters, the validation set is used for hyperparameter tuning and monitoring the training process, and the test set is used to finally evaluate the model's generalization performance.

[0119] Furthermore, the training process can employ a gradient descent-based optimization algorithm to minimize the aforementioned total loss function. In each iteration, a batch of samples is randomly sampled from the training set, for example, a batch size of 32. Normalization and feature extraction are sequentially performed on these samples to obtain corresponding feature vectors, which are then input into the corresponding recognition architecture to calculate the anomaly probability of each sample. Further, based on the true labels of the batch of samples and the model output, the dynamic hard example mining loss and the normal state spectrum consistency constraint loss are calculated according to the loss function formula, and then weighted and summed to obtain the total loss function. The gradient of the total loss with respect to all trainable parameters of the model is calculated using the backpropagation algorithm. The optimizer uses these gradients to update the values ​​of all trainable parameters, thereby enabling the model to produce lower loss values ​​in the next iteration, thus improving its ability to distinguish between normal and abnormal samples and making the feature distribution of normal samples more consistent with the consistency constraints.

[0120] Furthermore, the training process is not infinite; explicit stopping conditions need to be set to prevent overfitting. Typically, training monitors model performance on a validation set. When the anomaly detection accuracy or overall performance metric (such as the F1 score) on the validation set no longer improves over several consecutive training epochs, an early stopping mechanism is triggered, terminating training and preserving a snapshot of the optimal model parameters on the validation set. Additionally, a maximum number of training epochs is often set as a backup stopping condition. After training, a set of optimal model parameters is obtained. These parameters encapsulate complete knowledge from data preprocessing and feature extraction to anomaly decision-making, which is used for subsequent anomaly detection tasks.

[0121] In summary, the embodiments of the present invention are innovative compared to the prior art in the following aspects:

[0122] 1. A sliding window dual-channel collaborative adaptive normalization method is adopted. By dynamically adjusting the normalization denominator and introducing an inter-channel coupling adjustment factor, it can adapt to changes in the local dynamic range of the signal while preserving key transient features, thus avoiding the weakening of abrupt features by traditional global normalization.

[0123] 2. A dual-channel fusion time-frequency spectrum generation method based on complex Morlet wavelets is adopted. By fusing the time-frequency energy and phase consistency information of current and voltage through coherent power fusion, the generated time-frequency representation can more comprehensively reflect the cooperative change characteristics of the two channels.

[0124] 3. A multi-level feature extraction module is adopted to extract discriminative features from three complementary perspectives: time-frequency energy distribution, dual-channel phase synchronization, and the correlation between local sequence morphology and time-frequency energy. This overcomes the limitations of traditional time-domain or frequency-domain features being singular and insensitive to early anomalies.

[0125] 4. By adopting a dual-supervised loss function that combines dynamic hard example mining and normal state spectrum consistency constraints, as well as an anomaly probability output mechanism based on historical state adaptation, the challenges of data imbalance, variable operating conditions, and slow equipment state drift are effectively addressed, thereby improving the model's generalization ability and decision stability.

[0126] Based on the above system embodiments, this invention also provides an artificial intelligence-based battery swapping equipment anomaly identification device, referring to... Figure 6 The device includes: a data acquisition module 10 for acquiring process monitoring data of the battery swapping equipment under preset battery swapping conditions; a preprocessing module 20 for performing sliding window dual-channel collaborative adaptive normalization on the current and voltage parameters in the process monitoring data to construct normalized features; a feature extraction module 30 for determining discriminative features of the process monitoring data based on the normalized features and the fused time-frequency spectrum corresponding to the normalized features; an execution module 40 for determining the anomaly score corresponding to the process monitoring data based on the discriminative features; and an output module 50 for determining the anomaly identification result of the battery swapping equipment under preset battery swapping conditions based on the anomaly score.

[0127] The artificial intelligence-based battery swapping equipment anomaly identification device provided in this embodiment of the invention has the same implementation principle and technical effects as the aforementioned method embodiment. For the sake of brevity, any parts not mentioned in the device embodiment can be referred to the corresponding content in the aforementioned method embodiment.

[0128] This invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the above-described... Figures 1-2 The steps of any of the methods shown. Embodiments of the present invention also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the above-described steps. Figures 1-2 The steps of any of the methods shown. Embodiments of the present invention also provide a structural schematic diagram of an electronic device, such as... Figure 7 The diagram shows the structure of the electronic device, which includes a processor 101 and a memory 100. The memory 100 stores computer-executable instructions that can be executed by the processor 101. The processor 101 executes the computer-executable instructions to implement the above-mentioned... Figures 1-2 Any of the methods shown.

[0129] exist Figure 7In the illustrated embodiment, the electronic device further includes a bus 102 and a communication interface 103, wherein the processor 101, the communication interface 103, and the memory 100 are connected via the bus 102. The memory 100 may include high-speed random access memory (RAM) or non-volatile memory, such as at least one disk drive. Communication between this system network element and at least one other network element is achieved through at least one communication interface 103 (which can be wired or wireless), using the Internet, wide area network, local area network, metropolitan area network, etc. Bus 102 can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, or an AMBA (Advanced Microcontroller Bus Architecture) bus. AMBA defines three types of buses: APB (Advanced Peripheral Bus), AHB (Advanced High-performance Bus), and AXI (Advanced eXtensible Interface). Bus 102 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 7The diagram uses only a single double-headed arrow, but this does not imply a single bus or a single type of bus. Processor 101 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of processor 101 or by instructions in software form. Processor 101 can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the embodiments of this application can be directly manifested as execution by a hardware decoding processor, or execution by a combination of hardware and software modules in the decoding processor. The software module can reside in a mature storage medium in the field, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory, and the processor 101 reads information from the memory and, in conjunction with its hardware, completes the aforementioned tasks. Figures 1-2 Any of the methods shown.

[0130] The computer program product of the artificial intelligence-based battery swapping equipment anomaly identification method and device provided in this embodiment of the invention includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the methods described in the preceding method embodiments. For specific implementation, please refer to the method embodiments, which will not be repeated here. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system described above can be referred to the corresponding process in the preceding method embodiments, which will not be repeated here. If the function is implemented in the form of 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 the present invention, or the part that contributes to the prior art, 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, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, mobile hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0131] In the description of this invention, it should be noted that the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Finally, it should be noted that the above embodiments are merely specific implementations of this invention, used to illustrate the technical solutions of this invention, and not to limit it. The scope of protection of this invention is not limited thereto. Although this invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in this invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this invention, and should all be covered within the scope of protection of this invention. Therefore, the scope of protection of this invention should be determined by the scope of the claims.

Claims

1. A method for anomaly identification of battery swapping equipment based on artificial intelligence, characterized in that, The method includes: Acquire process monitoring data of the battery swapping equipment under preset battery swapping conditions; The current and voltage parameters in the process monitoring data are subjected to sliding window dual-channel collaborative adaptive normalization to construct normalized features; Based on the normalized features and the fused time-spectrum corresponding to the normalized features, the discriminative features of the process monitoring data are determined. Based on the discriminative features, the anomaly score corresponding to the process monitoring data is determined; Based on the anomaly score, the anomaly identification result of the battery swapping equipment under the preset battery swapping condition is determined; The step of performing sliding window dual-channel collaborative adaptive normalization on the current and voltage parameters in the process monitoring data to construct normalized features includes: Based on the local standard deviations of the current and voltage parameters in the process monitoring data within a preset sliding window, calculate the dual-channel adaptive scaling factor of the process monitoring data; based on the dual-channel adaptive scaling factor, determine the normalized denominator corresponding to the current and voltage parameters respectively; based on the normalized denominator, perform adaptive normalization on the current and voltage parameters respectively to construct the normalized feature of the process monitoring data for the dual channels; The step of determining the discriminative features of the process monitoring data based on the normalized features and the fused time-spectrum corresponding to the normalized features includes: The complex coefficients of continuous wavelet transform corresponding to the current and voltage parameters in the normalized feature are determined. Based on the energy information and phase coordination information of the complex coefficients of continuous wavelet transform at each time-frequency point, the fused time-frequency spectrum corresponding to the normalized feature is constructed. Based on the fused time-frequency spectrum, the time-frequency energy moment, the dual-channel coordination feature, and the transient anomaly feature of the normalized feature are determined. The time-frequency energy moment, the dual-channel coordination feature, and the transient anomaly feature are fused to construct the discriminative features of the process monitoring data. The step of determining the time-frequency energy moment of the normalized feature based on the fused time-frequency spectrum includes: constructing adaptive weights based on the amplitude of the fused time-frequency spectrum and the local standard deviations corresponding to the current parameters and voltage parameters in the normalized feature; and constructing the time-frequency energy moment corresponding to the normalized feature by weighted aggregation of the energy distribution of the normalized feature in the scale dimension and the energy distribution in the time dimension based on the adaptive weights. The step of determining the dual-channel collaborative features of the normalized features based on the fused time-frequency spectrum includes: determining the joint energy magnitude of the current and voltage parameters of the normalized features based on the complex coefficients of the continuous wavelet transform of the fused time-frequency spectrum; and performing target feature identification on the phase synchronization features of the current and voltage parameters based on the joint energy magnitude to determine the dual-channel collaborative features corresponding to the normalized features. The step of determining the transient anomalous features of the normalized features based on the fused time-frequency spectrum includes: determining the morphological mutation intensity index corresponding to the normalized features based on the second-order backward difference of the normalized features; weighting and aggregating the morphological mutation intensity index with the fused time-frequency spectrum to establish a mutation correlation between the fused time-frequency spectrum and local morphological mutations; and determining the transient anomalous features corresponding to the normalized features based on the mutation correlation.

2. The method according to claim 1, characterized in that, The step of fusing the time-frequency energy moment, the dual-channel collaborative features, and the transient anomaly features to construct discriminative features of the process monitoring data includes: Gaussian kernel nonlinear weighting is applied to the time-frequency energy moment, the dual-channel collaborative characteristics, and the transient anomaly characteristics to determine the discriminative features of the process monitoring data.

3. The method according to claim 1, characterized in that, The step of determining the anomaly score corresponding to the process monitoring data based on the discriminative features includes: The discriminative feature is determined to be a feature deviation from the normal baseline value. The feature deviation results are normalized using preset variance adjustment factors and sensitivity factors; Based on the feature deviation results of the normalization process, the abnormal scores corresponding to the discriminative features are determined.

4. The method according to claim 1, characterized in that, The step of determining the anomaly identification result of the battery swapping equipment under the preset battery swapping condition based on the anomaly score includes: Based on preset historical anomaly scores, an adaptive mapping is performed on the anomaly scores to determine the anomaly probability corresponding to each anomaly score: Based on the anomaly probability, the anomaly identification result corresponding to the battery swapping equipment is determined.

5. An artificial intelligence-based anomaly identification device for battery swapping equipment, characterized in that, The device includes: The data acquisition module is used to acquire process monitoring data of the battery swapping equipment under preset battery swapping conditions; The preprocessing module is used to perform sliding window dual-channel collaborative adaptive normalization on the current and voltage parameters in the process monitoring data to construct normalized features; The feature extraction module is used to determine the discriminative features of the process monitoring data based on the normalized features and the fused time-frequency spectrum corresponding to the normalized features. An execution module is used to determine the anomaly score corresponding to the process monitoring data based on the discriminative features; The output module is used to determine the anomaly identification result of the battery swapping equipment under the preset battery swapping condition based on the anomaly score. The preprocessing module is further configured to: calculate the dual-channel adaptive scaling factor of the process monitoring data based on the local standard deviations of the current and voltage parameters in the process monitoring data within a preset sliding window; determine the normalized denominator corresponding to the current and voltage parameters based on the dual-channel adaptive scaling factor; and perform adaptive normalization on the current and voltage parameters based on the normalized denominator to construct the normalized features of the process monitoring data for the dual channels. The feature extraction module is further configured to: determine the continuous wavelet transform complex coefficients corresponding to the current parameters and voltage parameters in the normalized features; construct the fused time-frequency spectrum corresponding to the normalized features based on the energy information and phase coordination information of the continuous wavelet transform complex coefficients at each time-frequency point; determine the time-frequency energy moment, dual-channel coordination features, and transient anomaly features of the normalized features based on the fused time-frequency spectrum; and perform feature fusion on the time-frequency energy moment, the dual-channel coordination features, and the transient anomaly features to construct the discriminative features of the process monitoring data. The step of determining the time-frequency energy moment of the normalized feature based on the fused time-frequency spectrum includes: constructing adaptive weights based on the amplitude of the fused time-frequency spectrum and the local standard deviations corresponding to the current parameters and voltage parameters in the normalized feature; and constructing the time-frequency energy moment corresponding to the normalized feature by weighted aggregation of the energy distribution of the normalized feature in the scale dimension and the energy distribution in the time dimension based on the adaptive weights. The step of determining the dual-channel collaborative features of the normalized features based on the fused time-frequency spectrum includes: determining the joint energy magnitude of the current and voltage parameters of the normalized features based on the complex coefficients of the continuous wavelet transform of the fused time-frequency spectrum; and performing target feature identification on the phase synchronization features of the current and voltage parameters based on the joint energy magnitude to determine the dual-channel collaborative features corresponding to the normalized features. The step of determining the transient anomalous features of the normalized features based on the fused time-frequency spectrum includes: determining the morphological mutation intensity index corresponding to the normalized features based on the second-order backward difference of the normalized features; weighting and aggregating the morphological mutation intensity index with the fused time-frequency spectrum to establish a mutation correlation between the fused time-frequency spectrum and local morphological mutations; and determining the transient anomalous features corresponding to the normalized features based on the mutation correlation.

Images