Battery abnormity early warning method and device based on ensemble learning

Abstract

The invention provides a battery abnormity early warning method and device based on ensemble learning, relates to the technical field of artificial intelligence, can avoid feature misjudgment caused by time dislocation based on preprocessing of instantaneous change of sampling values of multi-source time series data, can effectively reserve key early warning features, and improves the early warning efficiency. Therefore, the problems of data asynchronization and feature loss can be effectively solved. Besides, the time-varying statistical magnitude is extracted for the preset charge state interval, and the time-varying statistical magnitude is spliced with the frequency-domain adaptive entropy features to generate the target feature vector, so that the dynamic association among multiple parameters can be effectively reflected, and the characterization capability of early-stage composite anomalies is remarkably improved. Furthermore, battery abnormity prediction is carried out based on an integrated learning model containing a base learning device pool, a self-adaptive weight adjustment strategy and a result fusion unit, samples of different structures can be accurately matched, key abnormity modes are focused, the output battery abnormity probability has accuracy and stability, and efficient and reliable early warning of early abnormity of the battery can be realized.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method and apparatus for early warning of battery anomalies based on ensemble learning. Background Technology

[0002] Driven by the global energy transition, the large-scale application of battery systems in electric vehicles and energy storage power stations has made safety and reliability a core requirement. Batteries undergo latent electrochemical decay and material degradation during long-term operation, initially manifesting only as slight anomalies in monitoring parameters such as voltage and current. If not warned in time, this can easily lead to thermal runaway.

[0003] In addition, the monitoring data such as voltage and current collected by multiple sensors during battery operation have an inherent asynchronous delay. Existing fixed window alignment or simple filtering methods cannot effectively correct for this asynchrony and may even smooth out weak abrupt changes in early anomalies, leading to the loss of key features. Furthermore, the multi-dimensional monitoring parameters are dynamically correlated. Conventional methods extract time-domain features independently and then simply concatenate them, ignoring parameter correlation and frequency-domain energy change analysis, making it difficult to characterize early complex anomalies. Moreover, conventional algorithms do not consider the differences in the difficulty of anomaly characterization under different states of charge of the battery and treat all samples equally, resulting in insufficient learning of key anomaly patterns.

[0004] In summary, existing methods are insufficient to accurately capture early abnormal characteristics of batteries and cannot guarantee system safety. There is an urgent need for new solutions that overcome technical bottlenecks to meet the requirements of large-scale applications for early warning accuracy and reliability. Summary of the Invention

[0005] The purpose of this invention is to provide a method and device for early warning of battery anomalies based on ensemble learning, which can achieve efficient and reliable early warning of battery anomalies.

[0006] In a first aspect, embodiments of the present invention provide a method for early warning of battery anomalies based on ensemble learning. The method includes: acquiring operational data of a target battery; the operational data includes multi-source time-series data of the target battery; based on the instantaneous changes in the sampled values ​​of each attribute data in the operational data, performing path alignment and data cleaning on each attribute data to generate preprocessed data; calculating the time-varying statistics of the preprocessed data for a preset state of charge, and concatenating the time-varying statistics with the adaptive entropy features of the preprocessed data in the frequency domain to generate a target feature vector; inputting the target feature vector into a pre-constructed battery anomaly diagnosis model, determining the battery anomaly probability corresponding to the target feature vector through the battery anomaly diagnosis model, and providing an anomaly warning for the target battery based on the battery anomaly probability; wherein the battery anomaly diagnosis model is constructed based on an ensemble learning architecture, which includes a base learner pool, an adaptive weight adjustment strategy for the base learner pool, and a result fusion unit; each base learner in the base learner pool is used to process sample data with a preset structure.

[0007] In conjunction with the first aspect, the present invention provides a first implementation of the first aspect, wherein the steps of path alignment and data cleaning for each attribute data based on the instantaneous change of the sampled values ​​of each attribute data of the running data include: determining the instantaneous change of the sampled values ​​corresponding to each attribute data of the running data based on the sampling time of the running data; calculating the optimal alignment path corresponding to the instantaneous change of the sampled values ​​using a preset dynamic time warping algorithm; aligning the sampling time of each attribute data based on the optimal alignment path to generate first preprocessed data; and cleaning the running data according to the local outlier factors in the first preprocessed data to generate preprocessed data.

[0008] In conjunction with the first aspect, this embodiment of the invention provides a second implementation of the first aspect, wherein the step of calculating the time-varying statistics of the preprocessed data for a preset state of charge includes: dividing the preprocessed data into multiple intervals according to the battery state of charge value, and calculating the current distribution characteristics of each interval; weighting and fusing the current distribution characteristics based on the state of charge center value corresponding to each interval to generate the time-varying statistics corresponding to the preprocessed data; and concatenating the time-varying statistics with the adaptive entropy features of the preprocessed data in the frequency domain to generate a target feature vector, wherein the step of concatenating the time-varying statistics with the adaptive entropy features and the temporal coupling features corresponding to the preprocessed data to generate a target feature vector; wherein the adaptive entropy features are determined by adaptive wavelet packet decomposition of the preprocessed data.

[0009] In conjunction with the first aspect, this embodiment of the invention provides a third implementation of the first aspect, wherein the step of calculating the temporal coupling features corresponding to the preprocessed data includes: calculating the derived features of the preprocessed data in multiple dimensions; reconstructing the preprocessed data into a spatiotemporal feature tensor based on the derived features; the multiple dimensions include parameter type, time evolution, and feature morphology dimension; performing tensor decomposition on the spatiotemporal feature tensor to extract the core coupling feature tensor of the spatiotemporal feature tensor; flattening the core coupling feature tensor into a one-dimensional vector to generate the temporal coupling features of the preprocessed data.

[0010] In conjunction with the first aspect, this embodiment of the invention provides a fourth implementation of the first aspect, wherein the step of performing adaptive wavelet packet decomposition on the preprocessed data to determine the adaptive entropy features of the preprocessed data in the frequency domain includes: performing wavelet packet decomposition on the preprocessed data to determine the node wavelet coefficients corresponding to multiple frequency bands respectively; using a cost function minimization to determine the optimal wavelet packet decomposition tree corresponding to the node wavelet coefficients; and extracting the entropy values ​​of all leaf nodes from the optimal wavelet packet decomposition tree to constitute the adaptive entropy features corresponding to the preprocessed data.

[0011] In conjunction with the first aspect, this invention provides a fifth implementation of the first aspect, wherein the step of determining the battery anomaly probability corresponding to the target feature vector through a battery anomaly diagnosis model includes: inputting the target feature vector into a pre-constructed battery anomaly diagnosis model; determining an initial probability sequence of the target feature vector through the base learner of the battery anomaly diagnosis model; wherein the initial probability sequence includes cluster prediction probabilities corresponding to multiple clusters of the target feature vector; modeling the temporal dependency corresponding to the initial probability sequence through a gated recurrent unit network of the battery anomaly diagnosis model to generate an initial output; and performing expert prior calibration on the initial output to determine the battery anomaly probability corresponding to the target feature vector.

[0012] In conjunction with the first aspect, this embodiment of the invention provides a sixth implementation of the first aspect, wherein the method for constructing a battery anomaly diagnosis model includes: obtaining a training sample set corresponding to the running data; performing feature clustering on the training sample set to generate multiple cluster features, and constructing a base learner for each cluster feature; using a dynamic weight update mechanism to determine the dynamic weights of the base learner, and training the base learner model; and constructing a battery anomaly diagnosis model based on the trained base learner.

[0013] In conjunction with the first aspect, the present invention provides a seventh implementation of the first aspect, wherein the step of training the base learner includes: using a cross-entropy loss function as the training optimization objective of the base learner to train the base learner.

[0014] In conjunction with the first aspect, this embodiment of the invention provides an eighth implementation of the first aspect, wherein the step of determining the dynamic weights of the base learner using a dynamic weight update mechanism includes: determining the local anomaly factor of the cluster feature based on the outlier degree corresponding to the cluster feature; determining the interval weight corresponding to the current cluster feature according to the importance of the charge state interval to which the cluster feature belongs; and dynamically updating the weight coefficients of the base learner using the interval weights and the local anomaly factor to generate the dynamic weights of the base learner.

[0015] Secondly, embodiments of the present invention provide a battery anomaly early warning device based on ensemble learning. The device includes: a data acquisition module for acquiring operational data of a target battery; the operational data includes multi-source time-series data of the target battery; a preprocessing module for performing path alignment and data cleaning on each attribute data based on the instantaneous changes in the sampled values ​​of each attribute data in the operational data, generating preprocessed data; a data processing module for calculating the time-varying statistics of the preprocessed data for a preset state of charge, concatenating the time-varying statistics with the adaptive entropy features of the preprocessed data in the frequency domain, generating a target feature vector; and an execution module for inputting the target feature vector into a pre-built battery anomaly diagnosis model, determining the battery anomaly probability corresponding to the target feature vector through the battery anomaly diagnosis model, and providing an anomaly warning for the target battery based on the battery anomaly probability; wherein the battery anomaly diagnosis model is built based on an ensemble learning architecture, which includes a base learner pool, an adaptive weight adjustment strategy for the base learner pool, and a result fusion unit; each base learner in the base learner pool is used to process sample data with a preset structure.

[0016] The embodiments of this invention bring the following beneficial effects: This invention provides a method and apparatus for early warning of battery anomalies based on ensemble learning. By preprocessing the instantaneous changes in sampled values ​​of multi-source time-series data, it can avoid feature misjudgments caused by time misalignment and effectively retain key warning features, thereby effectively solving the problems of data asynchrony and feature loss. Furthermore, by extracting time-varying statistics for a preset state of charge (SOC) interval and concatenating them with frequency-domain adaptive entropy features to generate a target feature vector, it can effectively reflect the dynamic correlation between multiple parameters and significantly improve the characterization ability of early complex anomalies.

[0017] Furthermore, the ensemble learning model based on a base learner pool, adaptive weight adjustment strategy, and result fusion unit can accurately match samples with different structures, focus on key abnormal patterns, and output battery anomaly probabilities with both accuracy and stability, enabling efficient and reliable early warning of battery anomalies.

[0018] 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.

[0019] 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

[0020] 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.

[0021] Figure 1 A flowchart illustrating an early warning method for battery anomalies based on ensemble learning, provided as an embodiment of the present invention; Figure 2 This is a schematic diagram comparing the accuracy of battery anomaly detection according to the embodiments of the present invention; Figure 3 This is a schematic diagram illustrating the distribution of the number of abnormal samples; Figure 4 A comparison chart of the early anomaly detection capabilities of different methods; Figure 5 This is a schematic diagram illustrating the recognition performance corresponding to an embodiment of the present invention; Figure 6 A schematic diagram of a battery anomaly early warning device based on ensemble learning provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0022] 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.

[0023] 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.

[0024] This invention provides a method and apparatus for early warning of battery anomalies based on ensemble learning, which can achieve efficient and reliable early warning of early battery anomalies.

[0025] To facilitate understanding, the early warning method for battery anomalies based on ensemble learning provided in this embodiment of the invention will be described first, referring to... Figure 1 The method includes the following steps: Step S102: Obtain the operating data of the target battery.

[0026] The target battery is a specific battery cell or battery pack corresponding to the monitoring scenario, applicable to various battery systems such as electric vehicles and energy storage power stations. The operational data includes multi-source time-series data of the target battery. Operational data is a set of data reflecting the battery's operating status, collected in real-time by various sensors and data acquisition devices under actual operating conditions such as charge-discharge cycles and resting periods. Multi-source time-series data refers to sequential data continuously collected chronologically by multiple different types of data acquisition units (such as voltage sensors, current sensors, temperature sensors, and state-of-charge (SOC) estimation modules), reflecting the dynamic changes in battery status over time and used to capture early abnormal trends.

[0027] In one implementation, data acquisition can be performed in real time via a sensor network deployed on the battery module or system and a battery management system. The acquired parameters include the battery's total voltage, total current, temperature at each sampling point (located on the battery surface or at key hot spots), and the battery's state of charge (SOC) estimated using the ampere-hour integration method or a combination of the ampere-hour and open-circuit voltage methods. The data is continuously recorded at a fixed sampling frequency (e.g., 1Hz or 10Hz), covering the battery's complete operating cycle, including charging, discharging, and resting conditions, to comprehensively capture the battery's dynamic behavior under different states.

[0028] Step S104: Based on the instantaneous changes in the sampled values ​​of each attribute data in the running data, perform path alignment and data cleaning on each attribute data to generate preprocessed data.

[0029] Attribute data refers to a subdivided data type in multi-source time-series data that reflects a specific operating attribute of the battery, such as voltage attribute data, current attribute data, temperature attribute data, and state of charge attribute data. Instantaneous change of sampled values ​​refers to the difference or rate of change of a certain attribute data at adjacent sampling times. It can accurately reflect the dynamic fluctuation characteristics of the data and is used to distinguish between normal fluctuations and early abnormal weak mutations.

[0030] This invention achieves path alignment based on the instantaneous changes in attribute data sampling values. It can identify the time offset characteristics of each attribute data, realize dynamic path alignment, effectively correct the asynchronicity of multi-source data, and avoid feature misjudgment caused by time misalignment. In addition, it can distinguish noise from early anomalies and weak mutations based on the instantaneous changes in sampling values. This allows for the removal of meaningless noise during data cleaning while retaining key mutation signals related to anomalies, completely solving the defect of losing key early warning features in existing methods.

[0031] Step S106: Calculate the time-varying statistics of the preprocessed data for the preset state of charge, and concatenate the time-varying statistics with the adaptive entropy features of the preprocessed data in the frequency domain to generate the target feature vector.

[0032] The preset state of charge (SOC) refers to the abnormal characterization patterns of a battery at different operating stages. Time-varying statistics refer to the characteristic indicators that dynamically change over time, obtained by statistical analysis of preprocessed data within the preset SOC range, such as the mean, variance, peak value, and kurtosis of the voltage within the range. These can quantitatively reflect the time-domain operating characteristics of the battery within a specific SOC range. The SOC range can be pre-defined (e.g., 0-20%, 20%-80%, 80%-100%), and the electrochemical characteristics and abnormal behaviors of the battery differ significantly within different ranges. Furthermore, entropy is an indicator that measures the uncertainty or complexity of a signal. Adaptive entropy refers to the entropy value characteristic that can automatically adjust the calculation parameters according to the dynamic changes of the signal. It can accurately quantify the energy distribution complexity of the battery operating signal in the frequency domain, and its changes can effectively reflect the early electrochemical decay inside the battery.

[0033] This invention extracts time-varying statistics based on a preset SOC interval, which fully considers the differences in anomalous features under different charging states and avoids the problem of poor feature generalization caused by neglecting the influence of SOC in existing methods. Furthermore, the introduction of frequency-domain adaptive entropy features can compensate for the shortcomings of existing technologies that only focus on time-domain features, capturing early anomalous information hidden in the time-domain signal. This invention, through time-domain-frequency-domain feature concatenation, fuses multi-dimensional feature information, simultaneously reflecting the dynamic correlation between multiple parameters, significantly improving the characterization ability of early composite anomalies, and solving the problem of low accuracy in composite anomaly identification in existing methods.

[0034] Step S108: Input the target feature vector into the pre-built battery anomaly diagnosis model, determine the battery anomaly probability corresponding to the target feature vector through the battery anomaly diagnosis model, and provide an anomaly warning for the target battery based on the battery anomaly probability.

[0035] The battery anomaly diagnosis model is a model built on machine learning algorithms that can analyze target feature vectors and output the probability of battery anomalies, thereby achieving early warning of battery anomalies. In this embodiment of the invention, the battery anomaly diagnosis model is built on an ensemble learning architecture, which includes a base learner pool, an adaptive weight adjustment strategy for the base learner pool, and a result fusion unit; each base learner in the base learner pool is used to process sample data with a preset structure.

[0036] Ensemble learning architecture refers to a machine learning framework that combines the outputs of multiple independent base learners by designing reasonable weight adjustment and result fusion strategies to achieve better performance than a single learner. A base learner pool refers to a collection of base learners of different types or with different parameter settings (such as decision trees, support vector machines, neural networks, etc.). Each base learner has specific advantages in sample processing and can adapt to sample data with different structures (such as stable operating condition samples, dynamic operating condition samples, weak anomaly samples, etc.). An adaptive weight adjustment strategy refers to a dynamic adjustment mechanism that automatically adjusts the weight ratio of each base learner's output based on its diagnostic accuracy in different SOC ranges and sample types, allowing base learners with higher diagnostic accuracy to play a greater role in corresponding scenarios.

[0037] This invention's embodiment, based on the base learner pool setting, enables the model to adapt to sample data with different structures, avoiding the diagnostic blind spots of a single learner for specific samples. Combined with an adaptive weight adjustment strategy, weights can be dynamically allocated according to SOC interval differences and sample characteristics, allowing the model to focus on anomaly samples that are difficult to diagnose, thus addressing the shortcomings of existing algorithms that treat all samples equally, leading to insufficient learning of key anomalies. Furthermore, the result fusion unit comprehensively analyzes the weighted outputs of each base learner, combining time-dimension smoothing with common-sense battery operating principles (such as "the probability of anomalies is usually low in the early stages of charging") for calibration, and then outputs a stable anomaly probability. This effectively filters out instantaneous misjudgments caused by fluctuations in operating conditions, significantly reducing the false alarm rate and improving the stability of early warnings. Finally, by outputting accurate and stable anomaly probabilities, reliable early warnings of early battery anomalies are achieved, providing a guarantee for battery system safety.

[0038] In summary, the embodiments of the present invention, based on preprocessing the instantaneous changes of sampled values ​​of multi-source time-series data, can avoid feature misjudgment caused by time misalignment and effectively retain key early warning features, thereby effectively solving the problems of data asynchrony and feature loss. Furthermore, extracting time-varying statistics for a preset state of charge (SOC) interval and concatenating them with frequency-domain adaptive entropy features to generate a target feature vector can effectively reflect the dynamic correlation between multiple parameters and significantly improve the characterization ability of early composite anomalies.

[0039] Furthermore, the ensemble learning model based on a base learner pool, adaptive weight adjustment strategy, and result fusion unit can accurately match samples with different structures, focus on key abnormal patterns, and output battery anomaly probabilities with both accuracy and stability, enabling efficient and reliable early warning of battery anomalies.

[0040] Furthermore, based on the above embodiments, this embodiment of the invention also provides another method for early warning of battery anomalies based on ensemble learning. Regarding step S104 above, this embodiment of the invention determines the instantaneous change in the sampled value corresponding to each attribute data of the running data based on the sampling time of the running data; calculates the optimal alignment path corresponding to the instantaneous change in the sampled value using a preset dynamic time warping algorithm; and aligns the sampling time of each attribute data based on the optimal alignment path to generate first preprocessed data. Further, based on the local outlier factors in the first preprocessed data, the running data is cleaned to generate preprocessed data.

[0041] In battery operation data, the timing of multiple parameters such as current, voltage, temperature, and state of charge often becomes misaligned due to sensor sampling delays and fluctuations in operating conditions. Conventional cleaning methods based on fixed windows smooth out local abrupt changes in the data, thereby weakening the weak signal characteristics related to early battery anomalies. Directly performing timing alignment or fixed-window filtering will blur the dynamic coupling relationship between current, voltage, and state of charge, failing to effectively preserve the subtle features characterizing early battery anomalies.

[0042] This invention uses a dynamic time warping algorithm to align multi-source timing sequences to eliminate inter-sensor delays, and combines an adaptive cleaning window based on local gradient mutation detection to remove noise while retaining potential abnormal mutation points.

[0043] Corresponding to the steps above, the preprocessing procedure is as follows: 1) Multi-source time-series dynamic alignment Using the current sequence as a reference, the optimal alignment path between it and the voltage sequence is calculated using a dynamic time warping algorithm. Based on this path, the temperature sequence and the state of charge sequence are synchronously interpolated to correct the sampling delay between sensors. The time-aligned multidimensional data matrix is ​​then output, as follows:

[0044] In the formula, This represents the optimal alignment path found by the dynamic time warping algorithm, which minimizes the cumulative alignment cost of all matching points on the path. This represents a candidate alignment path, which is a set of point pairs used to match the time indices of current and voltage sequences. This represents the weighting coefficient for temperature gradient differences, used to balance the synchronicity of current-voltage differences and temperature changes in path distance calculations to compensate for the hysteresis effect of the temperature sensor. The default value is [value missing]. ; This indicates the search for parameters that minimize the objective function. The operation.

[0045] The time index represents the reference current sequence; The time index represents the target voltage sequence; Indicates the optimal alignment path A matching point pair in a current sequence indicates the first matching point in the current sequence. The first time point and voltage sequence Each time point should be aligned. Represents current sequence In the Sampled values ​​at each time index; Represents voltage sequence In the The sampled value at each time index. Indicates the temperature sequence in the index The first-order forward difference at point is calculated as follows: , used to characterize the instantaneous rate of temperature change; Representing temperature sequence In the Sampled values ​​at each time index; Representing temperature sequence In the Sampled values ​​at each time index; Indicates the temperature sequence in the index The first-order forward difference at the location.

[0046] In the dynamic time warping algorithm, the path is a set of point pairs, where each point pair represents the time index of the reference current sequence. Time index of the target voltage sequence These sequences should be aligned; this set defines the temporal mapping between the two sequences, ensuring that the cumulative alignment cost is minimized. It should be noted that... The term refers to the cost function calculation method used in conventional dynamic time warping algorithms. Conventional dynamic time warping is used to align two sequences that may have different lengths. Its core is to construct a cost matrix and find a path through dynamic programming that minimizes the cumulative distance between all pairs of points on the path. This distance is usually the Euclidean distance, Manhattan distance, or other point-to-point difference measure between the two sequences at their corresponding index points. This invention improves the cost function for aligning the path. Specifically, based on the above, this invention calculates the instantaneous rate of temperature change using a first-order difference calculation method, by constructing... This method measures the difference in temperature change trends between the i-th time point of the current sequence and the j-th time point of the voltage sequence, effectively addressing the problems of "asynchronous delay between sensors" and "ignoring the dynamic interaction between electrical and thermal parameters." Temperature sensors typically have significant hysteresis. Aligning only current and voltage might find a mathematically minimum-cost path, but this path may violate the physical timing logic of temperature changes. The improved method of this invention, when searching for an alignment path, tends to select time points where not only current and voltage values ​​match, but temperature change trends are also synchronized. The resulting optimal alignment path and synchronized data matrix better reflect the true electro-thermal coupling dynamics within the battery, moving beyond simple signal alignment to intelligent synchronization based on multi-physics correlation.

[0047] Furthermore, after timing alignment, the optimal alignment path is followed. All sequences are resampled or interpolated to obtain a synchronized multidimensional data matrix. (That is, the first preprocessed data), where, , , , They represent by Aligned sequence of current, voltage, temperature and state of charge.

[0048] 2) Adaptive cleaning based on gradient mutation The first-order gradient of the aligned voltage sequence is calculated, and the degree of abrupt outlier at each gradient point is evaluated based on the local outlier factor algorithm. Then, a threshold is dynamically set according to the local outlier factor value to filter out the data indices that are determined to be non-outliers to form a cleaning window. Finally, the cleaned time series data matrix is ​​output, represented as follows:

[0049] In the formula, Indicates the first Voltage gradient values ​​at each time point The local outlier factor is used to quantify the degree of outlier of a data point relative to its local neighborhood. The larger the value, the higher the probability that the point is a mutation outlier. Indicates the first The first-order difference of the voltage sequence at each time point is calculated as follows: , used to characterize the instantaneous rate of voltage change; Represents voltage sequence In the Sampled values ​​at each time point; Represents voltage sequence In the Sampled values ​​at each time point; Indicates a point-in-time index; This represents the number of nearest neighbors considered when calculating the local outlier factor; the default value is [value to be filled in]. ; This represents the index of the nearest neighbor, used for traversal. The nearest neighbor point, ; Indicates the first The local reachability density of the nearest neighbor is used to measure the density of data around that neighbor. This represents the local reachability density of a data point, used to measure the density of data surrounding that point. Represents a data point in the feature space; This represents a data point in the feature space.

[0050] Furthermore, after the local outlier factor is calculated, the cleaning window is dynamically determined based on its value, resulting in the set of time point indices retained after cleaning, represented as:

[0051] In the formula, This represents the set of time point indices retained after cleaning; points that meet the conditions are considered to be "normal" fluctuations. This represents the outlier threshold coefficient, used to adjust the strictness of the cleaning process; the default value is [value missing]. .

[0052] Based on this, from the synchronized multidimensional data matrix Extract all column indexes belonging to The data forms a cleaned time-series data matrix. (i.e., preprocessed data), its dimensions are , for The number of elements in the middle.

[0053] Furthermore, early battery anomaly signals are weak and easily drowned out by noise in the time domain, but may exhibit specific resonant frequency shifts or energy distribution changes in the frequency domain. Meanwhile, conventional statistical feature extraction methods are easily contaminated by random noise and lack dynamic correlation with the battery's state of charge (SOC). In conventional techniques, wavelet transform with fixed basis functions is ill-suited to the complex signal characteristics of batteries under different operating conditions, while isolated statistical features cannot effectively capture the temporal correlation between anomalies and operating conditions, resulting in insufficient characterization of early anomalies. Regarding step S106, this invention extracts adaptive entropy features in the frequency domain through adaptive wavelet packet decomposition to capture the frequency domain energy redistribution caused by anomalies. Simultaneously, it calculates time-varying statistics dynamically correlated with the battery's SOC to enhance the feature's adaptability to operating conditions. Then, the two types of features are fused to form a high-dimensional derived feature set that is more sensitive to early anomalies. For the time-varying statistics, this embodiment divides the preprocessed data into multiple intervals according to the battery's SOC value, calculates the current distribution characteristics of each interval to describe the statistical law of current data within a specific SOC interval, and quantitatively reflects the stability of the battery's charge and discharge state within that interval. Furthermore, based on the center value of the state of charge (SOC) corresponding to each interval, the current distribution characteristics are weighted and fused to generate time-varying statistics corresponding to the preprocessed data. The center value of SOC is used to indicate the midpoint value of the SOC interval. Here, it serves as a representative indicator of the SOC level of that interval, reflecting the importance of different intervals in battery operation. This embodiment of the invention is based on the battery state under a specific SOC corresponding to the current distribution characteristics of the partition, and by weighting and fusing the SOC center value, the feature weights are tilted towards the normal operating interval of the battery. This can effectively amplify the weak current change characteristics corresponding to early anomalies in that interval (such as the slight increase in the current variance of the core interval under abnormal operating conditions), while reducing the interference of extreme SOC intervals and improving the sensitivity of time-varying statistics to early anomalies. In specific implementation, the cleaned current sequence (i.e., the preprocessed data mentioned above) can be divided into multiple intervals according to the battery SOC value. In each interval, the mean, standard deviation, and skewness of the current are calculated. Then, weighted fusion is performed based on the distance between the center value of the SOC of each interval and the benchmark value to generate a statistical feature vector that is dynamically related to the battery state, expressed as:

[0054] In the formula, Represents a dynamic statistical feature vector with dimension . It characterizes the distribution characteristics of current under different charging states; Indices representing the range of states of charge. ; This represents the total number of charge state intervals; the default value is [value to be filled in]. ; Indicates the first The statistical feature vector extracted within each state-of-charge interval characterizes the central tendency, dispersion, and distribution shape of the current within that interval, and is expressed as follows: ; Indicates the first The sample mean of the current sequence within each state of charge interval; Indicates the first Sample standard deviation of current sequence within each state of charge interval; Indicates the first Sample skewness of current sequence within a state-of-charge interval; Indicates the transpose operation; Indicates the first The fusion weights corresponding to each charge state interval are calculated as follows: , used to emphasize the statistical characteristics under extreme states of charge; Indicates the first The center value of a range of charged states is the midpoint of the range if the range is equally divided. Indicates the first The center value of a range of charged states is the midpoint of the range if the range is equally divided. Indicates difference from The index of the state of charge interval, ; This represents the weighting adjustment factor, used to adjust the weights of extreme states of charge, and is preferably set to 1.0; This represents the natural exponential function.

[0055] Furthermore, regarding feature concatenation, this embodiment of the invention concatenates time-varying statistics, adaptive entropy features, and time-series coupling features corresponding to the preprocessed data to generate a target feature vector. This embodiment of the invention, by integrating features from the time domain, frequency domain, and parameter correlation domain, solves the information blind spot problem caused by existing methods relying solely on single-dimensional features, and can comprehensively cover the multi-dimensional manifestations of early battery anomalies (such as time-domain parameter fluctuations, frequency-domain energy mutations, and parameter correlation imbalances). Specifically, the adaptive entropy feature is determined by adaptive wavelet packet decomposition of the preprocessed data. For example, based on the frequency fluctuation characteristics of the input signal, the number of wavelet packet decomposition layers (e.g., 2-5 layers) and the basis function type are dynamically adjusted to adaptively decompose the original time-domain signal into multiple sub-signals of different frequency bands, ensuring that the decomposition results match the non-stationary characteristics of the signal. The entropy values ​​(e.g., sample entropy, approximate entropy) of each frequency band sub-signal are further calculated to quantify the complexity and uncertainty of the sub-signals, effectively mining the early anomaly frequency domain information hidden in the time-domain signal by capturing the energy distribution and complexity changes of the preprocessed data in the frequency domain. Correspondingly, the target feature vector is represented as: This formula represents the coupling eigenvector. Adaptive wavelet packet entropy feature vector and dynamic statistical eigenvectors The vectors are concatenated to obtain the derived feature vector (i.e., the target feature vector). Its dimension is... Based on multi-dimensional data integration, a more sensitive and comprehensive characterization of early battery anomalies has been formed.

[0056] Furthermore, regarding the adaptive entropy feature, this embodiment of the invention performs wavelet packet decomposition on the preprocessed data to determine the node wavelet coefficients corresponding to multiple frequency bands; and uses the minimization of the cost function to determine the optimal wavelet packet decomposition tree corresponding to the node wavelet coefficients. In one implementation, the data can be processed by a wavelet packet decomposition algorithm (such as adaptive decomposition to determine the number of decomposition layers and basis functions) to obtain a set of wavelet coefficients under each node, constituting the signal feature representation of that frequency band. The wavelet coefficients based on the node can reflect the energy intensity of the signal in the corresponding frequency band (node ​​p) and the corresponding feature component (index d). Furthermore, the frequency domain characteristics of the signal can be quantified by Shannon entropy. The node Shannon entropy can measure the degree of disorder of the signal in the frequency band corresponding to the p-th node in the wavelet packet decomposition tree. The level of entropy directly reflects the signal energy distribution characteristics: the higher the entropy value, the more uniform and random the distribution of signal energy in each wavelet coefficient in that frequency band; the lower the entropy value, the more concentrated the energy is in a few wavelet coefficients, and the signal has obvious regularity.

[0057]

[0058] In the formula, In the wavelet packet decomposition tree, the first... The Shannon entropy of a node is used to measure the uniformity or complexity of the signal energy distribution within the frequency band corresponding to that node. The higher the entropy value, the more uniform and random the energy distribution. In the wavelet packet decomposition tree, the first... One node; Indicates the first The node in the node The proportion of the energy of each wavelet coefficient to the total energy of that node is calculated as follows: ; This represents a logarithmic function, with the default base being the natural constant. Indicates the first The corresponding node is the 1st The wavelet coefficients are obtained through wavelet packet decomposition; Indicates the first The corresponding node is the 1st The wavelet coefficients are obtained through wavelet packet decomposition; Indicates the wavelet coefficients at the th Indexes in each node; Indicates difference from The wavelet coefficients in the first The index in each node. This Shannon entropy is used to measure the uncertainty of a signal (the degree of disorder in the distribution of signal energy). When the battery is operating normally, its monitoring signals (voltage, current) exhibit stable regularity within a specific frequency band, and the energy is concentrated in a few wavelet coefficients. At this time, p... p In the 'd' parameter, some values ​​are close to 1, while the rest are close to 0, indicating a low entropy value. When an early anomaly occurs in the battery, the regularity of the signal is disrupted, and energy diffuses to multiple wavelet coefficients, leading to variations in the values ​​of each p. p The values ​​of d are relatively uniform, and the entropy value increases significantly. Therefore, abrupt changes in entropy value can serve as an important characteristic signal of early anomalies.

[0059] Furthermore, the optimal wavelet packet decomposition tree This is achieved by optimizing the following objectives:

[0060] In the formula, Indicates the first The variance of all wavelet coefficients of a node represents the degree of energy fluctuation of the signal within the frequency band corresponding to that node. This represents the optimal wavelet packet decomposition tree obtained by optimizing the objective, i.e., the decomposition tree structure that minimizes the cost function; Decomposition tree All leaf nodes; This represents the node index of the wavelet packet decomposition tree; This indicates the search for the decomposition tree that minimizes the objective function value. The operation; is the variance sensitivity coefficient, set to 0.5 by default, used to adjust the impact of nodal variance in the cost function; e is the natural constant.

[0061] Furthermore, the entropy values ​​of all leaf nodes are extracted from the optimal wavelet packet decomposition tree to form the adaptive entropy feature corresponding to the preprocessed data. In specific implementation, this embodiment of the invention performs adaptive wavelet packet decomposition on the cleaned voltage sequence, selects the optimal wavelet packet decomposition tree by minimizing the cost function that comprehensively considers the node information entropy and signal variance, and calculates the entropy values ​​of all leaf nodes in the optimal tree to form a feature vector characterizing the frequency domain complexity and energy distribution of the signal, i.e., the adaptive entropy feature, expressed as: Among them, the adaptive wavelet packet entropy feature vector is obtained by... Extract all The entropy values ​​of the leaf nodes constitute the structure. Represents the optimal wavelet packet decomposition tree The total number of leaf nodes. This represents the entropy value of the first leaf node in the optimal wavelet packet decomposition tree. In the optimal wavelet packet decomposition tree, the th The entropy value of each leaf node.

[0062] Furthermore, regarding temporal coupling characteristics, this embodiment of the invention determines them by enhancing the coupling of preprocessed data with multimodal temporal features. In specific implementation, this embodiment calculates derived features of the preprocessed data in multiple dimensions. Based on these derived features, the preprocessed data is reconstructed into a spatiotemporal feature tensor. Tensor decomposition is performed on the spatiotemporal feature tensor to extract the core coupling feature tensor. The core coupling feature tensor is flattened into a one-dimensional vector to generate the temporal coupling features of the preprocessed data. These multiple dimensions include parameter type, temporal evolution, and feature morphology dimensions.

[0063] There are complex nonlinear coupling relationships between parameters such as current, voltage, temperature and state of charge of a battery. For example, an increase in temperature causes a change in internal resistance, which in turn leads to a decrease in terminal voltage. Single-parameter feature extraction methods cannot model this dynamic interaction across parameters. Conventional techniques extract the features of each parameter independently and then simply splice them together, ignoring the coupling effect between parameters, thereby reducing the sensitivity of the model to early composite anomaly signals.

[0064] This invention extracts low-dimensional fusion features that characterize the internal state interactions of a battery by constructing a high-order spatiotemporal feature tensor and performing constrained tensor decomposition, and jointly learning the coupling patterns between parameters, time points, and different feature types. The specific steps are as follows: 1) Constructing the spatiotemporal feature tensor The cleaned time-series data matrix is ​​combined with pre-derived features to reconstruct a third-order spatiotemporal feature tensor, which simultaneously encapsulates information in three dimensions: parameter type, temporal evolution, and feature morphology. This is represented as:

[0065] In the formula, The third-order spatiotemporal feature tensor represents the multimodal coupling characteristics of the battery's operating state, and its dimension is... ; Representing the spatiotemporal feature tensor In parameter index Time Index and feature index The element value at that position; Represents the third-order spatiotemporal feature tensor The first dimension index corresponds to the parameter type. These represent current, voltage, temperature, and state of charge, respectively. This indicates a time point index that is distinct from t; Represents the third-order spatiotemporal feature tensor The third dimension index corresponds to the feature type. These represent the original observation, the first-order difference, and the local energy characteristics, respectively. Represents the cleaned time series data matrix (i.e., preprocessed data) in the first The parameter in the first... The original observations at each time point; Represents the cleaned time series data matrix The Middle The parameter in the first... The original observations at each time point.

[0066] Represents the cleaned time series data matrix The Middle The parameter in the first... The first difference at each time point, i.e. ; Represents the cleaned time series data matrix The Middle The parameter in the first... The original observations at each time point. Indicates the first The parameter sequence is based on time points. The sum of energy of the data within a local window centered on the data is used to characterize the signal strength or volatility near that time point. Both represent the derived characteristics of the preprocessed data in their respective dimensions. This function represents the larger of the two values. This represents a function that takes the smaller of the two values.

[0067] 2) Coupled Feature Tensor Decomposition In one implementation, the spatiotemporal feature tensor can be decomposed using Tucker decomposition (adapted to tensors of any order). By projecting and reducing the dimensionality of the tensor onto three factor matrices in different dimensions, the core coupled feature tensor, which encapsulates multidimensional coupled information, can be extracted and flattened into a feature vector, represented as follows: .

[0068] In the formula, Represents the core coupled feature tensor, with dimension . It condenses the joint information between multiple parameters, long time spans, and multiple feature types; This represents the product operation between a tensor and a matrix on the first modulus. This represents the product operation between a tensor and a matrix on the second modulus. This represents the product operation between a tensor and a matrix on the third modulus; This represents the parameter coupling factor matrix, with dimension 1. Used to learn and compress four primitive parameter types into one A dimensional coupling parameter space; Indicates the parameter coupling rank, with a default value of ; This represents the time-dependent factor matrix, with dimension 1. Used to learn evolutionary patterns over time and compress them to dimension; Represents the time rank, with a default value of ; This represents the feature interaction factor matrix, with dimension 1. It is used to learn the interaction relationships among three feature types: original value, first-order difference, and local energy, and compress them to... dimension; Indicates the feature interaction rank, with a default value of Furthermore, the core coupling feature tensor Flattened into a one-dimensional vector, the coupled feature vector is obtained. Its dimensions are .

[0069] Based on the above data processing steps, this embodiment of the invention also analyzes the performance of various anomaly detection methods under different state-of-charge intervals, as well as the distribution of anomaly samples in different state-of-charge intervals. (Refer to...) Figure 2 This diagram illustrates a comparison of battery anomaly detection accuracy in embodiments of the present invention. The experiment compares the methods of the present invention, the convolutional neural network (LSW) method, and conventional statistical methods. In the experimental setup, the battery state of charge was divided into five equally wide intervals, and the detection accuracy of each method was independently tested within each interval. Figure 2 The line graph shows that the method of the present invention maintains a high detection accuracy in all states of charge ranges, and although there is a slight decrease in the extreme states of charge range, it is still significantly better than other methods.

[0070] Furthermore, refer to Figure 3 The provided diagram illustrating the distribution of abnormal sample numbers shows the actual number of abnormal samples within each state of charge interval, expressed in units of individuals. It can be seen that abnormal samples are more abundant in extreme state of charge intervals. Figure 2 , Figure 3 The analysis of the two figures shows that the method of the present invention exhibits stable high performance in different charge state ranges, indicating that it effectively enhances the adaptability to abnormal modes under different operating conditions through the dynamic weight integration mechanism of charge state perception and the extraction of statistical features dynamically associated with charge state. In contrast, the conventional method shows a significant performance degradation under extreme conditions, highlighting the robustness advantage of the method of the present invention under complex operating conditions.

[0071] Furthermore, early battery anomaly patterns are sparse, weak, and strongly coupled with complex operating conditions. Directly using a single model easily leads to overfitting on most "normal" samples, failing to fully exploit feature combinations with high discriminative power for the few "early anomaly" samples. Simultaneously, the model's adaptability to the cumulative effects of battery aging and different operating stages is insufficient, resulting in delayed warnings or false alarms. Regarding step S108 above, this embodiment of the invention inputs the target feature vector into a pre-constructed battery anomaly diagnosis model. The base learner of the battery anomaly diagnosis model determines the initial probability sequence of the target feature vector; the gated recurrent unit network of the battery anomaly diagnosis model models the temporal dependencies corresponding to the initial probability sequence, generating an initial output; the initial output is calibrated by expert priors to determine the battery anomaly probability corresponding to the target feature vector. The initial probability sequence includes the cluster prediction probabilities corresponding to multiple clusters of the target feature vector. This embodiment of the invention utilizes a multi-base learner to perform parallel predictions on different feature clusters of the target feature vector, obtaining a multi-perspective initial probability sequence. Furthermore, the temporal dependencies of the probability sequence can be captured through a gated recurrent unit (GRU) network, achieving dynamic information fusion. Furthermore, the fusion results are calibrated using expert rules in the battery field to eliminate unreasonable predictions, thereby outputting anomaly probabilities that closely reflect the actual operating state of the battery. These anomaly probabilities not only ensure the comprehensiveness of feature interpretation but also improve the stability and reliability of the probability output.

[0072] Corresponding to the above steps, this invention adopts a three-level ensemble learning architecture: First, a base learner pool with structural differences is constructed based on derived features to capture diverse anomaly discrimination patterns. Then, a novel adaptive boosting strategy based on battery state of charge evolution and sample local anomaly factor weighting is adopted to dynamically adjust the base learner weights, strengthen the focus on difficult-to-classify, potential early anomaly samples, and fuse the temporal prediction sequences of all base learners through a gated recurrent unit network. This is combined with expert priors based on operational phase divisions for decision calibration, outputting an anomaly probability sequence with temporal consistency and operational condition adaptability. In one implementation, the target feature vector corresponding to the test data is input into the pre-trained ensemble learning model. The model sequentially executes feature-guided heterogeneous base learner prediction, state of charge-aware dynamic weight integration (for online incremental learning or historical data analysis weight reference), and temporal gated fusion and expert prior decision calibration, outputting an anomaly probability value between 0 and 1. This probability value quantitatively represents the probability that the battery is in an abnormal state at the current moment or during the current analysis period. The early warning system continuously receives the anomaly probability sequence output by the model and applies preset early warning strategies. For example, a probability threshold (such as 0.7) can be set. When the anomaly probability exceeds this threshold at multiple consecutive time points, or when the probability value rises sharply in a short period of time, the system triggers different levels of early warning signals (such as alert, warning, and critical alarm). Simultaneously, the system can record and visualize the anomaly probability change curve, helping maintenance personnel trace the onset time and evolution trend of anomalies. This enables early warning of battery anomalies, from capturing weak signs to clear alarms, providing crucial decision support for preventative maintenance and safety management of the battery system.

[0073] The base learner clusters features from the training sample set corresponding to the running data, generating multiple cluster features, and then constructs a cluster feature for each cluster feature. The K-means clustering algorithm can be used to cluster the standardized feature matrix, with the number of clusters K determined by the elbow rule (usually K=3-5, corresponding to time domain, frequency domain, correlation domain, and other feature types). After clustering, the features within each cluster are sorted according to their variance contribution, retaining the top 80% of features by variance contribution, forming multiple cluster features with clear structures and prominent commonalities (e.g., cluster 1 is a time domain feature cluster, cluster 2 is a frequency domain feature cluster), reducing feature redundancy. Furthermore, a dedicated base learner is matched according to the characteristics of each cluster feature. For example, a heterogeneous learner combination strategy is used to improve model diversity: time domain feature clusters (numerical, highly regular) are matched with random forest learners; frequency domain feature clusters (high-dimensional, nonlinear) are matched with lightweight neural networks (e.g., 3-layer MLP); and time-coupled feature clusters (correlation-type features) are matched with support vector machines (SVM). The initial parameters of each base learner can be initially optimized through grid search (e.g., the number of decision trees in a random forest is set to 100, and the number of hidden layer neurons in an MLP is set to 64). A dynamic weight update mechanism is then used to determine the dynamic weights of the base learners, and the base learners are used for model training. A battery anomaly diagnosis model is then constructed based on the trained base learners.

[0074] A dynamic weight update mechanism is used to ensure the accuracy of the battery anomaly diagnosis model. This embodiment of the invention combines the essential characteristics of feature clusters with the importance of the operational scenario, enabling the model to focus on the diagnostic results corresponding to high-value features. Specifically, based on the outlier degree of the cluster features, local anomaly factors of the cluster features are determined; according to the importance of the state of charge (SOC) interval to which the cluster features belong, the interval weight corresponding to the current cluster feature is determined; the interval weight and local anomaly factors are used to dynamically update the weight coefficients of the base learner, generating the dynamic weights of the base learner. This embodiment of the invention uses a two-dimensional index of cluster feature outlier degree and state of charge (SOC) interval importance, ensuring that the base learner weights both align with the essential attributes of the features and match actual operational needs, thus overcoming the limitations of fixed weights or single-dimensional weights.

[0075] In practice, the model construction steps are as follows: 1) Feature-guided heterogeneous base learner generation By leveraging the nonlinear relationships and statistical differences among the features of each dimension of the derived feature vector, a set of complementary base learners is automatically generated. This avoids biases introduced by manual model selection and ensures the diversity of the ensemble system. Specifically, this is achieved through feature clustering and model type mapping, as follows:

[0076] In the formula, This represents the generated heterogeneous base learner set, which captures diverse anomaly detection patterns in derived feature vectors through different model architectures, thereby improving the robustness and generalization ability of the ensemble model. The union operation represents the set operation; Indicates the index of the base learner. ; This indicates the total number of base learners generated; the default value is 15. Indicates the first Each base learner derives feature vectors. Anomaly probability estimation of its output The mapping function; Indicates a mapping relationship; Indicates the first The anomaly probability prediction value of each base learner for the current sample. ; This indicates that it follows a distribution; This represents the generation probability distribution of the base learner. Indicates the first The set of hyperparameters for each base learner, such as the maximum depth of a decision tree, the number of hidden layer neurons in a neural network, etc. The model type function assigns a specific model architecture to the base learner based on the feature clustering results, such as gradient boosting tree, multilayer perceptron, linear discriminant analysis, etc. Represents the derived feature vector The performed feature clustering divides the features into Each cluster is specifically divided into several clusters using an unsupervised clustering algorithm. Taking all dimensional features as input, and dividing them according to feature similarity. There are several clusters, each with high similarity in features within each cluster and significant differences in features between clusters. Unsupervised clustering algorithms can employ K-means clustering or hierarchical clustering. This indicates the number of clusters for feature clustering, with a default value of 3.

[0077] It's important to note that the base learner is the fundamental classifier in ensemble learning, and can be a gradient boosting tree, multilayer perceptron, linear discriminant analysis, support vector machine, or random forest, etc. It's also worth noting that the derived feature vectors typically contain three main types of features: coupling features, frequency domain entropy features, and dynamic statistical features. Setting the number of clusters for feature clustering to three clusters corresponds to these three types of features, thus better matching feature characteristics when assigning model types. Furthermore, the base learner on the training set... The above minimizes its specific loss function Training was conducted, including... For the first The true labels for each training sample are 0 for normal and 1 for abnormal. The number of training samples, For the first Derived feature vectors of each training sample For the index of the training samples, .

[0078] For the training dataset, the collected raw time-series data can be segmented and labeled. For example, the long-term series is first divided into sliding windows of fixed duration or fixed number of data points. Each window constitutes a sample segment to be analyzed. Then, each sample segment is labeled, with the labeling category determined based on the battery's actual operating state and known fault or anomaly records. It is mainly divided into two categories: "normal" and "abnormal." The "normal" category indicates that the battery performance parameters are within the safe threshold and no known faults have occurred during this time period. The "abnormal" category corresponds to periods when the battery exhibits early performance degradation or minor faults. The criteria for this can be combined with features such as abnormal voltage drops, abnormal temperature increases, accelerated capacity decay curves, and internal resistance growth exceeding thresholds, or aligned with the time points of early fault symptoms recorded in actual maintenance records. Furthermore, the constructed training dataset consists of a large number of labeled samples. Each sample contains a multi-dimensional time-series data segment (current, voltage, temperature, state of charge) and its corresponding category label (0 indicates normal, 1 indicates abnormal).

[0079] 2) Dynamic weighted integration of state-of-charge sensing The conventional AdaBoost algorithm assigns the same weight increase to all misclassified samples, without considering the impact of the state of charge during battery operation on the difficulty of anomaly characterization or the anomaly potential of the samples themselves.

[0080] This invention employs a dynamic weight update mechanism, which correlates the weight increase with the charge state interval to which the sample belongs and its local anomaly factor values. This makes the integration process more focused on potential anomalous samples under key operating states, as follows:

[0081] In the formula, Indicates the first In the first round of integration iteration The weight of each training sample represents the importance of that sample in the current ensemble learning. The higher the weight, the more difficult or crucial the sample is to classify, and the more attention the subsequent base learner will pay to these samples. Indicates the first In the first round of integration iteration The weights of each training sample; This represents the index of the iteration rounds in ensemble learning. ; This represents the total number of iterations in the ensemble learning, preferably set to 50 to 100. The specific value depends on the number of training samples and the model complexity, and is usually selected through cross-validation to ensure that the ensemble model converges and does not overfit. Represents the natural exponential function; Indicates the first The current weak learner in the round of iteration The weighting coefficients are calculated as follows: ; for The weighted error rate is obtained by calculating the classification error rate of the current weak learner on the training samples, specifically the sum of the weights of all misclassified samples; It is an indicator function that returns 1 if the condition inside the parentheses is true, and 0 otherwise. Indicates the first The true label for each sample is either 0 or 1, where 0 indicates normal and 1 indicates abnormal. Indicates as of the date In round 1, the current ensemble model is for the 1st... The predicted category of each training sample is specifically determined by analyzing the previous... The threshold is determined by weighted voting results of all base learners. If the weighted probability is greater than 0.5, the prediction is 1; otherwise, it is 0. Indicates the first The charge state interval of each sample The corresponding weights are used to emphasize the importance of samples in extreme charge state intervals, and the calculation method is the same as that of the first... Fusion weights corresponding to each charge state interval same; Indicates based on sample The state of charge value determines the index of the state of charge interval to which it belongs; This represents the adjustment coefficient for local anomaly factors, used to control the intensity of the influence of local anomaly factors on weight updates. The default value is 0.5. Indicates the first The local anomaly factor value corresponding to the nth sample represents the degree of outlier of that sample point in its original time series context. The calculation method is the same as that of the nth sample. Voltage gradient values ​​at each time point Local outlier same; and and represent the minimum and maximum values ​​of the local outlier factor for all training samples, respectively. According to this rule, for misclassified samples, if they are in an extreme state of charge or the original signal exhibits high outlier characteristics, they receive a greater weight increase in the next iteration, driving subsequent base learners to focus on learning these "critical and difficult-to-distinguish" sample patterns.

[0082] In practical implementation, when determining whether a sample is in an extreme state of charge range, its state of charge value is compared with a threshold. For example, a state of charge close to 0 or 1 is considered an extreme range. Furthermore, the high outlier of the original signal is determined by the value of the local anomaly factor; the larger the value of the local anomaly factor, the higher the outlier.

[0083] 3) Temporal gating fusion and expert prior decision calibration go through After rounds of iteration, a set of weighted basis learners is obtained. The weighted voting result is ;in, This represents the total number of base learners, i.e., the number of base learners in ensemble learning. As the index of the base learner, ; For the first A base learner is a classification model; For the first The weights of each base learner are calculated using the AdaBoost algorithm; This is the outlier probability after weighted voting, which is the weighted average of the prediction probabilities of all base learners; For the first The anomaly probability prediction of a sample by each base learner.

[0084] Furthermore, a time-gated fusion network is adopted, and decision calibration is performed by combining expert prior knowledge during the operation phase, outputting a smooth and common-sense abnormal probability sequence, which is represented as:

[0085] In the formula, Indicates at time step The output calibration anomaly probability is made smoother and more consistent with common sense about battery operation through time series fusion and expert prior calibration. For time step index, ; To predict the total length of the time series; This represents the Sigmoid activation function, which maps the output to... interval; represents the weight matrix of the output layer of the gated recurrent unit network, which are trainable parameters; This represents the bias vector of the output layer of the gated recurrent unit network, which is a trainable parameter; This represents a gated recurrent unit network used to model the temporal dependencies of the predicted probability sequences of the base learners; Indicates from time step 1 to time step 2. Weighted voting probability of all base learners The sequence formed; This indicates that the gated recurrent unit network is at time step The hidden state; This indicates the total number of preset battery operation phase types, for example... It can represent three typical stages: constant current charging, constant power discharging, and resting. Indexes representing runtime phase types, ; Indicates at time step The sample was determined to belong to the first The probability of each operational phase is given by a pre-defined lightweight phase classifier, such as a shallow network classifier based on current, voltage sign, and rate of change. Indicates at time step The battery's operating phases, such as charging, discharging, or resting; Indicates at time step The anomaly probability after weighted voting of the base learners, i.e. At time step The value; Indicates the first The "expert prior" calibration function corresponding to the class's runtime phase outputs a calibration offset. For example, during the charging phase, if the voltage continues to rise slowly but... A sudden increase, A negative offset may be applied to suppress false alarms caused by the rise in normal constant current charging voltage.

[0086] It should be noted that the specific structure of the gated recurrent unit network includes update gates and reset gates, which are used to capture long-term dependencies in the time series. The gated recurrent unit network receives the weighted voting probability sequence of the base learner as input and outputs the hidden state for final prediction.

[0087] Furthermore, refer to Figure 4 This invention also demonstrates the early anomaly detection capabilities of different methods during actual battery operation through time series analysis, particularly the timeliness of early warning at the time of anomaly occurrence. The experiment used battery voltage time-series data including normal fluctuations and abnormal drops, with the time unit being minutes. The anomaly probability curves output by the method of this invention, the convolutional neural network long short-term memory network method, and conventional statistical methods were compared. The conventional statistical method employs a threshold determination method based on sliding window statistical features and logistic regression. The raw time-series data of four parameters—current, voltage, temperature, and state of charge—are segmented using a fixed-length sliding window. Within each window, the mean, standard deviation, maximum value, minimum value, and range of each parameter are independently calculated. These statistics are then simply concatenated and input into a logistic regression classifier for normal and abnormal judgment.

[0088] The battery voltage time series plot illustrates the voltage variation over time, with the red shaded area indicating the period when the actual anomaly occurred. A time series comparison plot of the early anomaly detection probabilities of different methods shows the anomaly probability outputs for the three methods. In the experimental setup, all methods received the same voltage sequence as input and output anomaly probability values ​​between zero and one. The probability curves show that the method of this invention begins to show a probability increase approximately 5 minutes before the actual anomaly occurs, maintains a high probability output during the anomaly's duration, and gradually decreases after the anomaly ends, with a relatively smooth and stable curve throughout the process. The convolutional neural network (LSM) method only reaches a high probability after the anomaly occurs, exhibiting a significant detection lag. Conventional statistical methods frequently show probability spikes throughout the time period, generating numerous false alarms, and also exhibit large probability fluctuations during the actual anomaly. Experimental results show that the method of the present invention captures early abnormal signs in the frequency domain through adaptive wavelet packet entropy extraction, smooths the probability sequence through a time-gated fusion network, and reduces false alarms caused by operating condition fluctuations by combining expert prior decision calibration, thereby achieving earlier, more accurate, and more stable early warning. Conventional technologies either have delayed warnings or excessively high false alarm rates, which cannot meet the dual requirements of timeliness and reliability in practical engineering applications.

[0089] Furthermore, the cross-entropy loss function can be used as the training optimization objective for the base learners to train the model. In each training iteration, in order to quantify the gap between the predicted output of the ensemble learning model (including all base learners and the temporally gated fusion network) and the true label, it is necessary to calculate the loss function. This invention uses the cross-entropy loss function as the optimization objective for model training. Specifically, for each sample in the training dataset, based on the calibration anomaly probability output after temporally gated fusion and expert prior calibration, and the known true class label of the sample, the cross-entropy loss value of the sample is calculated. The magnitude of this loss value directly reflects the uncertainty or error degree of the model's prediction for the sample: when the model predicts a high anomaly probability for normal samples or a low anomaly probability for abnormal samples, a large loss value will be generated; conversely, when the predicted probability is highly consistent with the true label, the loss value tends to be zero. After each complete training iteration (i.e., all training samples are processed by the model once), the cross-entropy loss values ​​of all training samples are summed or averaged to obtain the total training loss of the current model in that iteration. Then, using the backpropagation algorithm, the gradients of all trainable parameters in the model with respect to the loss function are calculated based on the total loss value. The gradient represents the direction and magnitude of parameter adjustment to reduce the loss. Then, the stochastic gradient descent optimization algorithm is used to update all parameters in the model at once according to the calculated gradient and the set learning rate. This parameter update process completes one model iteration. By repeatedly executing the loop of "forward propagation to calculate prediction and loss, backpropagation to calculate gradient, and optimizer to update parameters", the model continuously adjusts its internal representation and decision boundary. The goal is to make the total cross-entropy loss function value continuously decrease, thereby improving the model's ability to distinguish between normal battery states and early abnormal states.

[0090] The iterative process of model training is not infinite; specific stopping conditions need to be set to prevent overfitting or resource waste. The stopping iteration judgment in this invention is mainly based on the following two criteria: Criterion 1 is based on a preset maximum number of iterations. When the number of training iterations reaches the preset upper limit, training is forcibly stopped regardless of model performance, ensuring the training process is completed within a controllable time. Criterion 2 is based on an early stopping mechanism, which monitors the model's performance on a validation dataset independent of the training set. After each training iteration, in addition to calculating the training loss, the model with the current parameters is evaluated on the validation set, calculating performance metrics such as cross-entropy loss or classification accuracy. The early stopping mechanism continuously tracks the changing trend of the validation set performance metrics. If the validation set performance no longer improves or even begins to decline in consecutive iterations (e.g., 10 or 20 iterations), while the training loss continues to decrease, it indicates that the model may be overfitting to noise or specific patterns in the training data, and its generalization ability no longer improves. In this case, the training process will terminate early and roll back to the model parameters corresponding to the iteration with the best validation set performance, using these parameters as the final trained model. By combining the maximum number of iteration rounds with an early stopping mechanism based on validation set performance, overfitting can be effectively avoided while ensuring that the model fully learns effective patterns, thus obtaining a battery anomaly early warning model with stronger generalization ability.

[0091] Furthermore, this embodiment of the invention also compares the F1 values ​​of different methods at various operating stages to evaluate the identification performance of different anomaly detection methods in three typical battery operating stages (charging, discharging, and resting). (Refer to...) Figure 5 The comparison methods include: fixed window cleaning + logistic regression: data is cleaned using a fixed time window and then classified using logistic regression; independent features + support vector machine: features of each parameter are extracted independently and then input into a support vector machine; dynamic time warping + random forest: only time alignment is used but feature enhancement is not coupled; wavelet transform + gradient boosting: frequency domain features are extracted based on fixed basis wavelet transform; the technology of this invention (integrated early warning): fully realizes dynamic alignment, feature coupling and integrated learning.

[0092] The vertical axis of the graph represents the F1 score (a comprehensive evaluation index), and the horizontal axis represents the operational stage. Experimental results show that the technology of this invention achieves the highest F1 score during the charging stage, significantly outperforming other methods, demonstrating the effectiveness of multi-stage adaptability and feature enhancement.

[0093] In summary, the embodiments of the present invention are innovative compared to the prior art in the following aspects: 1. To address the issues of asynchronous data and noise interference from battery sensors, a dynamic time warping algorithm is innovatively adopted to correct timing misalignments. Combined with an adaptive cleaning strategy based on local outlier factors, this approach avoids smoothing out weak abrupt change signals related to early anomalies while denoising.

[0094] 2. By constructing a high-order spatiotemporal feature tensor and performing constrained tensor decomposition, the coupling relationships between parameters, time points, and feature types are jointly learned; at the same time, by combining adaptive wavelet packet entropy extraction and dynamic statistical features related to the state of charge, a comprehensive feature representation that is more sensitive to early anomalies is formed.

[0095] 3. An improved version of the conventional AdaBoost algorithm is adopted, which dynamically correlates the sample weight update with the importance of the charge state interval and the value of local outlier factors. This makes the ensemble process more focused on difficult-to-classify samples under extreme conditions and potential outliers, thereby improving the ability to learn sparse outlier patterns.

[0096] 4. The temporal prediction results of the gated recurrent unit network are fused with the base learner, and the decision calibration is performed by combining the expert prior rules based on the operation stage classification. This achieves smooth output of the abnormal probability sequence and suppression of false alarms, thereby enhancing the temporal consistency and adaptability of the early warning results.

[0097] Based on the above system embodiments, this invention also provides a battery anomaly early warning device based on ensemble learning, referring to... Figure 6 The device includes: a data acquisition module 10 for acquiring operational data of the target battery; the operational data includes multi-source time-series data of the target battery; a preprocessing module 20 for performing path alignment and data cleaning on each attribute data based on the instantaneous changes of the sampled values ​​of each attribute data in the operational data, generating preprocessed data; a data processing module 30 for calculating the time-varying statistics of the preprocessed data for a preset state of charge, and concatenating the time-varying statistics with the adaptive entropy features of the preprocessed data in the frequency domain to generate a target feature vector; and an execution module 40 for inputting the target feature vector into a pre-built battery anomaly diagnosis model, determining the battery anomaly probability corresponding to the target feature vector through the battery anomaly diagnosis model, and providing an anomaly warning for the target battery based on the battery anomaly probability; wherein, the battery anomaly diagnosis model is built based on an ensemble learning architecture, which includes a base learner pool, an adaptive weight adjustment strategy for the base learner pool, and a result fusion unit; each base learner in the base learner pool is used to process sample data with a preset structure.

[0098] The battery anomaly early warning device based on ensemble learning provided in this embodiment of the invention has the same implementation principle and technical effect 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.

[0099] 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... Figure 1The steps of the method are 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. Figure 1 The steps of the method are shown. This invention also provides a schematic diagram of the structure of an electronic device, as shown. 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... Figure 1 The method shown.

[0100] exist Figure 7 In 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. Figure 1 The method shown.

[0101] The computer program product of the battery anomaly early warning method and device based on ensemble learning provided in this 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. Specific implementations can be found in the method embodiments and 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, and will not be repeated here. Furthermore, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances. 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 a 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 the present invention. 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.

[0102] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, 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 the invention, used to illustrate the technical solutions of the invention, and not to limit it. The scope of protection of the invention is not limited thereto. Although the 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 battery abnormality early warning method based on ensemble learning, characterized in that, The method comprises: acquiring operation data of a target battery; the operation data comprises multi-source time series data of the target battery; based on the instantaneous change of the sampling value of each attribute data of the operation data, path alignment and data cleaning are performed on each attribute data to generate preprocessed data; calculating the time-varying statistics of the preprocessed data for a preset state of charge, and performing feature splicing on the time-varying statistics and the adaptive entropy feature of the preprocessed data in the frequency domain to generate a target feature vector; inputting the target feature vector into a pre-constructed battery anomaly diagnosis model, determining the battery anomaly probability corresponding to the target feature vector through the battery anomaly diagnosis model, and performing anomaly warning on the target battery based on the battery anomaly probability; wherein the battery anomaly diagnosis model is constructed based on an ensemble learning architecture, the ensemble learning architecture comprises a base learner pool, an adaptive weight adjustment strategy of the base learner pool, and a result fusion unit; each base learner included in the base learner pool is used to process sample data of a preset structure.

2. The method of claim 1, wherein, Based on the instantaneous change of the sampling value of each attribute data of the operation data, path alignment and data cleaning are performed on each attribute data to generate preprocessed data, comprising: based on the sampling time of the operation data, determining the instantaneous change of the sampling value of each attribute data of the operation data respectively; calculating the optimal alignment path corresponding to the instantaneous change of the sampling value through a preset dynamic time warping algorithm; based on the optimal alignment path, the sampling time of each attribute data is aligned to generate first preprocessed data; based on the local outlier factor in the first preprocessed data, the operation data is cleaned to generate preprocessed data.

3. The method of claim 1, wherein, The step of calculating the time-varying statistics of the preprocessed data for a preset state of charge comprises: dividing the preprocessed data into multiple intervals according to the battery state of charge value, and calculating the current distribution characteristics of each interval; based on the state of charge center value corresponding to each interval respectively, the current distribution characteristics are weighted and fused to generate the time-varying statistics corresponding to the preprocessed data; The step of splicing the time-varying statistics and the adaptive entropy feature of the preprocessed data in the frequency domain to generate a target feature vector comprises: splicing the time-varying statistics and the adaptive entropy feature, and the time series coupling feature corresponding to the preprocessed data to generate a target feature vector; wherein the adaptive entropy feature is determined by adaptively decomposing the preprocessed data into wavelet packets.

4. The method of claim 3, wherein, The step of calculating the time series coupling feature corresponding to the preprocessed data comprises: calculating the derived features of the preprocessed data in multiple dimensions, and reconstructing the preprocessed data into a spatio-temporal feature tensor based on the derived features; the multiple dimensions include parameter type, time evolution and feature morphology dimension; performing tensor decomposition on the spatio-temporal feature tensor to extract the core coupling feature tensor of the spatio-temporal feature tensor; flattening the core coupling feature tensor into a one-dimensional vector to generate the time series coupling feature of the preprocessed data.

5. The method of claim 3, wherein, The step of performing adaptive wavelet packet decomposition on the pretreated data to determine adaptive entropy features of the pretreated data in the frequency domain comprises: Performing wavelet packet decomposition on the pretreated data to determine node wavelet coefficients corresponding to a plurality of frequency bands respectively; Determining an optimal wavelet packet decomposition tree corresponding to the node wavelet coefficients by minimizing a cost function; Extracting entropy values of all leaf nodes from the optimal wavelet packet decomposition tree to form adaptive entropy features corresponding to the pretreated data.

6. The method of claim 1, wherein, The step of determining a battery abnormality probability corresponding to the target feature vector by the battery abnormality diagnosis model comprises: Inputting the target feature vector into a pre-constructed battery abnormality diagnosis model to determine an initial probability sequence of the target feature vector by a base learner of the battery abnormality diagnosis model; wherein the initial probability sequence comprises cluster prediction probabilities corresponding to a plurality of clusters of the target feature vector respectively; Modeling time sequence dependence corresponding to the initial probability sequence by a gated recurrent unit network of the battery abnormality diagnosis model to generate an initial output; Calibrating the initial output by expert priori to determine the battery abnormality probability corresponding to the target feature vector.

7. The method of claim 6, wherein, The method for constructing the battery abnormality diagnosis model comprises: Obtaining a training sample set corresponding to the operation data; Performing feature clustering on the training sample set to generate a plurality of cluster features and constructing a base learner for each cluster feature; Determining dynamic weights of the base learner by a dynamic weight updating mechanism to perform model training on the base learner; Constructing a battery abnormality diagnosis model based on the trained base learner.

8. The method of claim 7, wherein, The step of performing model training on the base learner comprises: Using a cross-entropy loss function as a training optimization target of the base learner to perform model training on the base learner.

9. The method of claim 7, wherein, The step of determining dynamic weights of the base learner by a dynamic weight updating mechanism comprises: Determining a local anomaly factor of the cluster feature based on a degree of outlying of the cluster feature; Determining an interval weight corresponding to the current cluster feature according to importance of a state of charge interval to which the cluster feature belongs; Dynamically updating weight coefficients of the base learner by the interval weight and the local anomaly factor to generate dynamic weights of the base learner.

10. An integrated learning-based early warning device for battery abnormalities, characterized in that, The device comprises: A data acquisition module configured to acquire operation data of a target battery; the operation data comprises a plurality of time sequence data of the target battery; A preprocessing module configured to perform path alignment and data cleaning on each attribute data of the operation data based on instantaneous changes of sampling values of the attribute data to generate pretreated data; A data processing module configured to calculate time-varying statistics of the pretreated data for a preset state of charge, perform feature splicing on the time-varying statistics and adaptive entropy features of the pretreated data in the frequency domain, and generate a target feature vector; and A battery abnormality diagnosis module configured to determine a battery abnormality probability corresponding to the target feature vector by a battery abnormality diagnosis model. The execution module is configured to input the target feature vector into a pre-constructed battery abnormality diagnosis model, determine a battery abnormality probability corresponding to the target feature vector through the battery abnormality diagnosis model, and perform an abnormality early warning on the target battery based on the battery abnormality probability. The battery abnormality diagnosis model is constructed based on an ensemble learning architecture. The ensemble learning architecture includes a base learner pool, an adaptive weight adjustment strategy of the base learner pool, and a result fusion unit. Each base learner included in the base learner pool is configured to process sample data of a preset structure.

Images