CN-122017594-A - Lithium battery protection board charge-discharge out-of-limit anomaly detection method based on data analysis

CN122017594ACN 122017594 ACN122017594 ACN 122017594ACN-122017594-A

Abstract

The invention discloses a method for detecting out-of-limit anomalies of charge and discharge of a lithium battery protection board based on data analysis, which relates to the technical field of operation data analysis and anomaly diagnosis of a lithium battery protection board management system and is used for solving the problems that out-of-limit anomalies of charge and discharge are difficult to stably identify and position at an event level under the conditions of working condition switching, distribution drifting and marking scarcity, collecting single cell voltage, battery pack current and temperature, aligning and cleaning and normalizing uniformly, constructing a time sequence sample by a sliding window, solving a reconstruction anomaly score by a variation self-encoder model based on an attention mechanism, combining a supervision learning fusion score and a local outlier factor method outlier score, correcting and dynamically weighting according to concentration, migration and marking adjacency to form a comprehensive score, and confirming out-of-limit events, outputting types, associated cells and records according to a threshold value and persistence rule.

Inventors

SUN TAIGUO
LV LEI
WANG FEI

Assignees

深圳市悦之彩科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260213

Claims (8)

1. The method for detecting the out-of-limit abnormality of the charge and discharge of the lithium battery protection board based on data analysis is characterized by comprising the following steps: The protection board collects the voltage of the single battery cell, the current and the temperature of the battery pack in a sampling period during charging and discharging, and builds a time window sample matrix under the unified discrete sampling moment set; taking a time window sample matrix as input, calculating attention scores and attention weights, generating attention convergence vectors, obtaining potential variables, reconstructing a window, and outputting an abnormality score and an attention weight sequence; obtaining supervised learning anomaly score and local outlier anomaly score for each time window sample, calculating the intra-window concentration by the attention weight sequence, calculating the inter-window migration amplitude by the adjacent potential variable mean value vector, and calculating the window label proximity by the neighbor label statistics; after the three correction scores are subjected to bounded mapping, index normalization weights are generated according to the intra-window concentration, inter-window migration amplitude, window mark proximity and evidence consistency, and the three correction scores are aggregated to obtain a comprehensive anomaly score; And (3) carrying out threshold value and persistence judgment on the comprehensive abnormal score to confirm out-of-limit abnormal events, determining charge out-of-limit abnormal or discharge out-of-limit abnormal and out-of-limit associated cell sets, and outputting recording results.
2. The method for detecting the out-of-limit abnormality of the charge and discharge of the lithium battery protection board based on data analysis according to claim 1 is characterized in that the protection board collects single cell voltage, battery current and temperature in a sampling period which is obtained by jointly restraining hardware sampling capacity and out-of-limit observation resolution during the charge and discharge period, and establishes a unified discrete sampling time set by a time stamp to realize time alignment of three types of channels, wherein original sampling of each channel is mapped to the nearest sampling time at each unified time, linear interpolation is adopted for the cell voltage and the temperature when a missing point occurs, and forward holding is adopted for current which possibly has a step so as to complement multidimensional observation vectors at the same time.
3. The method for detecting the abnormal charge-discharge limit crossing of the lithium battery protection board based on data analysis according to claim 2 is characterized in that impossible value rejection is carried out according to the reasonable range of the protection board range and the sensor, a neighborhood median or a neighborhood mean value and an absolute deviation median threshold value are used for judging in a neighborhood window, local peaks are replaced by neighborhood statistics to obtain purified multidimensional time sequence data, mean values and standard deviations of all dimensional features are calculated on a definitely selected stable operation statistical interval and standardized to obtain normalized multidimensional time sequence features, sliding segmentation is carried out on the normalized multidimensional time sequence features according to window length and window step length, and feature vectors corresponding to a plurality of continuous unified moments are stacked in time sequence in each window to construct a time window sample matrix.
4. The method for detecting the out-of-limit anomaly of the charge and discharge of the lithium battery protection board based on data analysis, which is characterized by taking a constructed time window sample matrix as input, representing each time window sample as a normalized multidimensional feature vector sequence arranged in time sequence, calculating attention scores at each sampling moment in a window, obtaining attention weights through exponential normalization, enabling the sum of the attention weights to be a weighted convergence of feature vectors at each moment in a window to generate an attention convergence vector, generating a mean vector and a variance vector of potential variables through coding and mapping of the attention convergence vector, obtaining a potential variable sample through heavy parameterization sampling, and generating a reconstruction window with the same shape as the input window through decoding and mapping; And then calculating square errors from moment to moment by dimension on the input window and the reconstruction window, accumulating the square errors in the window to obtain the reconstruction errors, calculating the kurbek's Brillouin divergence of the potential variable approximate posterior distribution relative to the standard normal prior distribution, and carrying out weighted summation on the reconstruction errors and the divergence according to a preset weight coefficient to obtain the variable self-coding total loss of the window, thereby defining the total loss as an abnormal score of a time window sample, and reserving an attention weight sequence corresponding to the window for tracing the position of a key sampling moment with larger contribution to the convergence vector in the window.
5. The method for detecting the out-of-limit anomaly of the charge and discharge of the lithium battery protection board based on data analysis, which is characterized in that for each constructed time window sample, on one hand, a time stamp of protection board protection action triggering record is taken as an out-of-limit anomaly event anchor point, a window class mark is established according to a time coverage relation with a sliding window to form a marked training sample pair; On the other hand, the same window feature vector is used as a feature space sample point, the distance between the sample points is calculated after the number of neighborhood points is confirmed according to normal operation data verification, a neighbor set is selected, the reachable distance is introduced to construct local reachable density, and then the ratio of the local reachable density of the neighbor to the local reachable density of the neighbor is averaged to obtain a local outlier factor which is used as a local outlier abnormal score, so that corresponding scores are respectively output on two links of a supervision evidence relying on a protection action mark and a density outlier evidence not relying on a mark, and input is provided for subsequent same-caliber normalization and comprehensive abnormal score generation.
6. The method for detecting the lithium battery protection board charge-discharge out-of-limit anomaly based on data analysis according to claim 5 is characterized in that the protection board collects single cell voltage, battery pack current and temperature in a sampling period to form time window samples, attention weight sequences, potential variable distribution parameters and corresponding anomaly scores which are output by a variable self-coding link are obtained on each time window, meanwhile, supervision learning fusion anomaly scores and local outlier anomaly scores are obtained, then, for each time window, the concentration in the window is calculated by using the attention weight sequences, the concentration in the window is determined by the ratio of the maximum value of the attention weight in the window to the average value of the attention weight in the window to represent whether anomaly contributions are concentrated at few sampling moments, the inter-window migration amplitude is calculated by using the two-norm difference of the potential variable average value vector of the adjacent time window to represent the inter-window form overall migration degree, and the out-of-limit anomaly marker proportion of the neighbor training samples is selected in a feature space by using the window feature vector to calculate the window marker proximity to represent the marker support degree near the window.
7. The method for detecting the lithium battery protection board charge-discharge out-of-limit anomaly based on data analysis, which is characterized by comprising the steps of obtaining intra-window concentration, inter-window migration amplitude and window mark adjacency, counting a mean value and a standard deviation of correlation values from a normal window sample set in a stable operation time period, standardizing the self-coding anomaly score, the intra-window concentration, the inter-window migration amplitude and the local outlier anomaly score, further carrying out logic function adjustment on the self-coding anomaly score based on the standardized intra-window concentration, carrying out logic function adjustment on the supervised learning anomaly score based on the window mark adjacency, carrying out logic function adjustment on the local outlier score based on the standardized inter-window migration amplitude, obtaining three correction scores, calculating evidence consistency by the three correction scores, wherein the evidence consistency is obtained by reverse mapping of standard deviations of the three correction scores, finally carrying out limited mapping on the three correction scores to obtain polymerizable amounts, jointly generating index normalization weights by the intra-window concentration, the inter-window migration amplitude and the local outlier score, and the polymerizable evidence consistency score, and carrying out comprehensive time score generation on the three correction scores according to the weight.
8. The method for detecting the out-of-limit anomaly of the charge and discharge of the lithium battery protection board based on data analysis according to claim 7 is characterized in that for the obtained comprehensive anomaly score of each time window sample, a normal window sample set is selected in a determined stable operation time period, the mean value and standard deviation of the comprehensive anomaly score of the set are calculated, and accordingly a comprehensive anomaly score threshold value is generated; Then, converting the start-stop window numbers of the sequences to obtain event start-stop sampling moments so as to determine event time intervals, judging charge out-of-limit abnormality or discharge out-of-limit abnormality according to the sign statistics of battery current in the time intervals, calculating interval maximum values or interval minimum values of each battery cell voltage sequence, and comparing the interval maximum values or interval minimum values with charge cut-off voltage thresholds or discharge cut-off voltage thresholds configured by a protection plate respectively so as to determine out-of-limit associated battery cell sets; and finally, outputting and recording out-of-limit abnormal results, wherein the out-of-limit abnormal results comprise event starting time, ending time, out-of-limit abnormal types, out-of-limit associated battery core sets, comprehensive abnormal score peaks and peak value window numbers in event coverage windows.

Description

Lithium battery protection board charge-discharge out-of-limit anomaly detection method based on data analysis Technical Field The invention relates to the technical field of operation data analysis and abnormality diagnosis of a lithium battery protection board management system, in particular to a lithium battery protection board charge and discharge out-of-limit abnormality detection method based on data analysis. Background The lithium battery energy storage and power system is usually characterized in that a protection board monitors the voltage of a single battery cell, the current and the temperature of a battery pack in real time, and triggers protection when out-of-limit risks such as overcharging, overdischarging and the like occur. However, in the actual working condition, the problems of asynchronous sampling channels, communication delay, noise spikes, missing points, temperature gradients, aging drift and the like exist, so that false alarm or missing alarm is easy to generate only depending on a single threshold value or a single statistical index, meanwhile, the protection action log has hysteresis, and the out-of-limit precursor may be expressed as short-time form jump or slow accumulation across windows, and is difficult to stably identify by using a fixed rule. On the other hand, the output scale and meaning of the same window by different algorithms drift along with the change of working conditions, and if the output scale and meaning of the same window are directly fixed, weighted fusion is easy to amplify non-fault switching when multi-cluster migration is distributed such as fast plugging and fast unplugging. The present invention proposes a solution to the above-mentioned problems. Disclosure of Invention In order to overcome the defects in the prior art, the embodiment of the invention provides a method for detecting the abnormal charge and discharge limit crossing of a lithium battery protection plate based on data analysis, so as to solve the problems in the background art. In order to achieve the above purpose, the present invention provides the following technical solutions: In a preferred embodiment, the method comprises: The protection board collects the voltage of the single battery cell, the current and the temperature of the battery pack in a sampling period during charging and discharging, and builds a time window sample matrix under the unified discrete sampling moment set; taking a time window sample matrix as input, calculating attention scores and attention weights, generating attention convergence vectors, obtaining potential variables, reconstructing a window, and outputting an abnormality score and an attention weight sequence; obtaining supervised learning anomaly score and local outlier anomaly score for each time window sample, calculating the intra-window concentration by the attention weight sequence, calculating the inter-window migration amplitude by the adjacent potential variable mean value vector, and calculating the window label proximity by the neighbor label statistics; after the three correction scores are subjected to bounded mapping, index normalization weights are generated according to the intra-window concentration, inter-window migration amplitude, window mark proximity and evidence consistency, and the three correction scores are aggregated to obtain a comprehensive anomaly score; And (3) carrying out threshold value and persistence judgment on the comprehensive abnormal score to confirm out-of-limit abnormal events, determining charge out-of-limit abnormal or discharge out-of-limit abnormal and out-of-limit associated cell sets, and outputting recording results. In a preferred embodiment, the protection board collects the single cell voltage, the battery pack current and the temperature in a sampling period which is obtained by the common constraint of hardware sampling capacity and out-of-limit observation resolution during charging and discharging, and establishes a unified discrete sampling time set by a timestamp to realize the time alignment of three types of channels, wherein original sampling of each channel is mapped to the nearest sampling time at each unified time, linear interpolation is adopted for the cell voltage and the temperature when a missing point occurs, and forward holding is adopted for the current which possibly has a step so as to complement the multidimensional observation vector at the same time. In a preferred embodiment, the impossible value is removed according to the range of the protection board and the reasonable range of the sensor, the neighborhood median or the neighborhood mean and the absolute deviation median threshold are used for judging in the neighborhood window, the neighborhood statistic is used for replacing the local peak to obtain the purified multidimensional time sequence data, the mean and standard deviation of each dimensional characteristic are calculated on the specifically selected stable o