Lithium battery thermal runaway analysis method and system

Abstract

The invention belongs to the technical field of lithium batteries, and discloses a lithium battery thermal runaway analysis method and system. The method comprises the following steps: acquiring a multi-source observation sequence to form an aligned observation set; extracting response time sequence characteristics of the detector; grouping the operation time periods to form working condition groups, generating a response time sequence chain and actual peak value distribution for the working condition groups, and forming a working condition grouping response map; constructing a topological change indication quantity, and comparing the topological change indication quantity with a topological change threshold value set to determine a super-threshold detector pair to generate a spatial alias mark; retrieving a hypothesis path subset by using a working condition group identifier, and screening to obtain a candidate discharge path set; calculating a sequence consistency rate and a peak value consistency rate, and screening an effective path set; obtaining a dynamic space mapping relation, and outputting a risk source positioning result; performing differentiation processing based on the risk source positioning result; according to the method, the risk source is determined by extracting the response time sequence characteristics, screening the effective paths and generating the dynamic space mapping relation, so that the error disposal risk is reduced.

Description

Technical Field

[0001] This invention relates to the field of lithium battery technology, and more specifically, to a method and system for analyzing thermal runaway in lithium batteries. Background Technology

[0002] With the widespread deployment of lithium-ion batteries in electrochemical energy storage cabinets, energy storage containers, and electric vehicle power battery packs, the monitoring and handling of early signs of thermal runaway typically employs a combination of multiple sensors: gas detectors, temperature detectors, and pressure detectors are placed inside the cabinet or battery pack, along with air ducts, venting channels, guide vanes, and compartment structures, so that changes in decomposition gases and heat generated under abnormal operating conditions can be detected in a timely manner, thereby triggering safety actions such as isolation, power reduction, inerting, and water-based cooling suppression to reduce the risk of thermal runaway propagation.

[0003] In existing technologies, many solutions focus on optimizing the design of exhaust channels and flow guiding structures, or analyzing gas diffusion and concentration distribution based on fixed geometry and fixed ventilation conditions, in order to improve the controllability of venting at the structural design level. However, most of the above solutions assume that the spatial mapping relationship between sensors and venting paths remains stable during operation, and the location of risk sources often relies on preset fixed correspondence rules of "sampling point-spatial location" or fixed propagation sequence assumptions.

[0004] In actual operation, energy storage cabinets and battery packs may undergo transportation, handling, installation, maintenance, vibration, shock, or minor collisions. Deformation, displacement, or partial obstruction may occur in the guide vanes, air duct components, venting channels, and compartment boundaries, altering the order in which vented gas reaches the detectors and the peak distribution position. The core problem arising from this is that the sensor response timing and peak distribution will deviate from the preset mapping. If the positioning rules under the fixed topology assumption are still used, it is easy to cause distortion in the location of risk sources, misidentifying the module actually venting as another module, thus causing deviations in the target of subsequent handling actions.

[0005] The aforementioned location distortion directly amplifies the risks of handling: Under mislocation conditions, actions such as isolation, inerting, or water-based cooling suppression may be applied to non-risk modules instead of covering the actual discharge modules in time, causing heat to continue to accumulate and spread to adjacent modules; at the same time, handling resources are consumed unnecessarily, which may lead to insufficient handling capacity or unstable action effects when entering a higher risk stage, or even cause unnecessary system throttling and shutdown, resulting in additional safety and operational losses.

[0006] In view of this, the present invention proposes a method and system for analyzing the thermal runaway of lithium batteries to solve the above problems. Summary of the Invention

[0007] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, the present invention provides the following technical solution: a method for analyzing the thermal runaway of lithium batteries, comprising:

[0008] An aligned observation set is formed by acquiring multi-source observation sequences of the battery pack during the current operating period using a detector. The multi-source observation sequences include gas concentration sequences, temperature sequences, pressure sequences, fan operating condition sequences, and action execution sequences.

[0009] Based on the aligned observation set, response timing features are extracted for each detector, including arrival time, rising edge duration, peak time, and peak amplitude.

[0010] The operating time segments are grouped into operating condition groups by using the wind turbine operating condition sequence. For each operating condition group, a response time sequence chain and actual peak distribution are generated to form an operating condition group response map.

[0011] Based on the working condition group response map, a topology change indicator is constructed. The topology change indicator is compared with the topology change threshold set to determine the over-threshold detector pair and generate spatial alias tags.

[0012] When the spatial alias is marked as existing, a subset of hypothetical paths is retrieved from the pre-built hypothetical path library using the working condition group identifier as the index, and a set of candidate discharge paths is obtained by filtering based on the consistency between the actual response order of the over-threshold detector pair and the expected response order of each hypothetical path.

[0013] For each candidate leakage path, calculate the sequence consistency rate and peak consistency rate, and compare them with the corresponding thresholds to filter and obtain the effective path set;

[0014] The dynamic spatial mapping relationship is obtained by weighted summarization and normalization of the effective path set, and the risk source location result containing the target module and the location credibility marker is output accordingly.

[0015] Gating is performed on trusted markers based on the risk source location results to implement differentiated treatment.

[0016] Furthermore, methods for extracting temporal features of the response for each detector include:

[0017] Pre-configure the corresponding response threshold for each type of detector;

[0018] The observation set is aligned along the time axis. When the observation value of a certain detector first meets its response threshold and reaches the persistence constraint, the arrival time is determined.

[0019] The duration of the rising edge is obtained by statistically analyzing the time span during which the observed value remains in the response state, starting from the arrival time.

[0020] Within the continuous time interval corresponding to the response state, locate the moment when the observed value reaches its maximum value, determine the peak time, and record the maximum value as the peak amplitude.

[0021] Furthermore, the methods for obtaining the working condition group response map include:

[0022] Based on the wind turbine operating condition sequence, intervals with the same operating condition identifier and continuous time are divided into the same operating condition group.

[0023] For each operating condition group, the arrival time, rise time duration, peak time and peak amplitude of all detectors in the operating condition group are summarized, and the detectors are sorted in order of arrival time to obtain the response timing chain of the operating condition group.

[0024] The actual peak distribution is formed by indexing peak time and peak amplitude;

[0025] The response time series chain corresponding to each working condition group and the actual peak distribution are constructed into a response set. The correspondence between each working condition group and the response set is established to form a working condition group response map.

[0026] Furthermore, methods for constructing topology change indicators include:

[0027] For each operating condition group, read the set of detector pairs that match the identifier of that operating condition group from the preset detector pair configuration table;

[0028] For each pair of detectors within the same working condition group, the difference in arrival time of each pair of detectors is calculated as the arrival time difference, the difference in peak time of each pair of detectors is calculated as the peak time difference, and the ratio of peak amplitude of each pair of detectors is calculated as the peak amplitude ratio.

[0029] The arrival time difference, peak time difference, and peak amplitude ratio of all detector pairs in the working condition group are sequentially spliced ​​together in a preset order to obtain the topology change indication of the working condition group.

[0030] Furthermore, methods for identifying over-threshold detector pairs and generating spatial alias tags include:

[0031] During the calibration period, normal fluctuation ranges are formed for the three types of components of each detector pair. The normal fluctuation ranges include the normal range of arrival time difference, the normal range of peak time difference, and the normal range of peak amplitude ratio.

[0032] During the operational comparison, it is determined whether the arrival time difference of the detector pair falls within its normal range, whether the peak time difference falls within its normal range, and whether the peak amplitude ratio falls within its normal range.

[0033] When any component exceeds the corresponding normal fluctuation range, it is determined that the component meets the threshold condition, and the detector pair is marked as a threshold detector pair under this working condition group.

[0034] If any over-threshold detector pair exists in the operating condition group, then the operating condition group is considered to be inconsistent, and the output space alias is marked as present;

[0035] If no over-threshold detector pair exists, the output space alias is marked as non-existent.

[0036] Furthermore, the methods for obtaining the candidate leakage path set include:

[0037] A hypothetical path library is pre-established, which is a set of hypothetical paths pre-established when the battery pack is deployed;

[0038] Retrieve a subset of hypothetical paths that match the current working condition group from the hypothetical path library, using the current working condition group identifier as an index;

[0039] For each hypothetical path, check whether the response order of the over-threshold detector pairs is consistent with the expected response order of the hypothetical path. When the number of consistent over-threshold detector pairs is not less than the preset matching number threshold, the hypothetical path is retained, and a set of candidate discharge paths is obtained.

[0040] Furthermore, the method for obtaining the set of valid paths is as follows:

[0041] The expected response order of the candidate path is compared with the actual extracted response time sequence chain in the aligned observation set. The comparison unit is the order of the detector pairs.

[0042] When the expected order matches the actual order, it is recorded as consistent; when they do not match, it is recorded as inconsistent. The order consistency rate is obtained by comparing the number of consistent detector pairs with the total number of detector pairs included in the statistics.

[0043] The expected peak amplitude of each detector in the candidate path and the peak amplitude of each detector in the actual response are sorted from largest to smallest. The detectors in the top r% are used to form the expected peak priority detector set and the actual peak priority detector set.

[0044] The expected peak priority detector set is aligned and compared with the actual peak priority detector set. The number of detectors in the intersection of the two is counted, and the ratio of the number of detectors in the intersection to the total number of detectors in the actual peak priority detector set is used as the peak consistency rate.

[0045] When the sequence consistency rate is greater than the preset sequence consistency threshold and the peak consistency rate is greater than the preset peak consistency threshold, the candidate leakage path is determined to be a valid path. If either requirement is not met, the candidate leakage path is removed, and the valid paths are unified into a set of valid paths.

[0046] Furthermore, methods for obtaining dynamic spatial mapping relationships based on weighted summarization and normalization of the effective path set include:

[0047] A pre-maintained association table between detector and module location indexes is provided, which includes detector identifier, neighboring module location index, partition identifier, hypothetical path identifier, and pointing weight.

[0048] For each valid path in the set of valid paths, read the expected response order bound to that valid path and determine the set of detectors involved in that path;

[0049] Based on the association table, for each detector involved in the effective path, read the neighboring module location index and pointing weight corresponding to the hypothetical path identifier;

[0050] For the same module location index, the pointing weights from different detectors are accumulated to obtain the in-path pointing strength of the effective path for each module location index;

[0051] A path weight is configured for each valid path, and the path weight is obtained by equally weighting the sequential consistency rate and the peak consistency rate of the valid path;

[0052] For each module location index, the in-path pointing strength of all valid paths to that module location index is summarized, and then weighted and accumulated according to the corresponding path weights to obtain the fused pointing value of that module location index.

[0053] Normalization is performed on the fused pointer values ​​of all module location indices to obtain normalized pointer values, thereby obtaining the dynamic spatial mapping relationship between module location indices and normalized pointer values.

[0054] Furthermore, the methods for obtaining risk source location results include:

[0055] Read the normalized pointer value corresponding to the location index of each module, and determine the location index of the module with the largest normalized pointer value as the target module. Determine the location index of the module with the second largest normalized pointer value and record its normalized pointer value as the second highest normalized pointer value.

[0056] Set a trust threshold C and a discrimination threshold D. Compare the normalized pointing value corresponding to the target module with the trust threshold C. If the normalized pointing value is not less than C, continue to make a judgment. If the normalized pointing value is less than C, output the location trust mark as failed.

[0057] Under the premise of passing the trust threshold judgment, calculate the difference between the normalized pointing value and the second highest normalized pointing value corresponding to the target module, and compare the difference with the discrimination threshold D;

[0058] When the difference is not less than D, the output positioning confidence mark is passed; when the difference is less than D, the output positioning confidence mark is failed.

[0059] When both location credibility markers of the target module are passed, it is output as the risk source location result.

[0060] A lithium battery thermal runaway analysis system, implementing the aforementioned lithium battery thermal runaway analysis method, includes:

[0061] Data acquisition module: The battery pack acquires multi-source observation sequences during the current operating period through detectors to form an aligned observation set. The multi-source observation sequences include gas concentration sequence, temperature sequence, pressure sequence, fan operating condition sequence, and action execution sequence.

[0062] Response extraction module: Based on the aligned observation set, extract response timing features for each detector. The response timing features include arrival time, rising edge duration, peak time, and peak amplitude.

[0063] The graph generation module uses the wind turbine operating condition sequence to group the operating periods into operating condition groups, generates a response time sequence chain and actual peak distribution for each operating condition group, and forms an operating condition group response graph.

[0064] Alias ​​tagging module: Constructs topology change indicators based on the working condition group response map, compares the topology change indicators with the topology change threshold set to determine over-threshold detector pairs and generates spatial alias tags;

[0065] Discharge path filtering module: When the spatial alias is marked as existing, the module retrieves a subset of hypothetical paths from the pre-built hypothetical path library using the working condition group identifier as the index, and filters the candidate discharge path set based on the consistency between the actual response order of the over-threshold detector pair and the expected response order of each hypothetical path.

[0066] Effective path filtering module: Calculates the sequence consistency rate and peak consistency rate for each candidate leakage path, and compares them with the corresponding thresholds to filter and obtain the effective path set;

[0067] Risk source localization module: Based on the weighted aggregation and normalization of the effective path set, a dynamic spatial mapping relationship is obtained, and the risk source localization result containing the target module and the location credibility marker is output accordingly;

[0068] Risk Management Module: Gating is performed based on the location credibility markers of the risk source location results to implement differentiated management.

[0069] The technical effects and advantages of the lithium battery thermal runaway analysis method and system proposed in this invention are as follows:

[0070] First, by mapping multi-source data such as gas, temperature, pressure, fan operating conditions, and handling actions to the same time axis and constructing an aligned observation set, and combining threshold judgment and persistence constraints to extract response time-series features such as arrival time, peak time, and peak amplitude, the multi-source responses have consistent comparability and verifiability in the time dimension, thereby effectively suppressing false alarms and missed alarms caused by transient noise, isolated spikes, and asynchronous sampling.

[0071] Secondly, by introducing operating period grouping and condition grouping response maps based on fan operating conditions, the influence of changes in ventilation conditions on the order and intensity distribution of responses can be separated and attributed, so that the propagation sequence and intensity differences can be explained under relatively consistent operating conditions. On this basis, detectors are used to construct topological change indicators for relative quantities and identify over-threshold patterns, so that spatial mapping inconsistencies can be quantified into triggerable spatial aliasing tags, thereby enhancing the ability to identify abnormal propagation caused by factors such as changes in structural connectivity and changes in the state of the venting channel.

[0072] Furthermore, after the spatial aliasing is triggered, the search scope is limited to the candidate paths provided by the hypothetical path library. Valid paths are selected based on the consistency rate of response order and the peak consistency rate. The dynamic spatial mapping relationship is further integrated to output the target module and the location credibility marker. This transforms the risk source location from unconstrained point inference to evidence aggregation and consistency screening under path constraints, thereby improving the stability and anti-interference ability of the location results and reducing the frequent switching between similar candidates due to small fluctuations.

[0073] In summary, this invention extracts response time-series features by aligning observation sets, and uses detectors to trigger spatial aliasing for relative quantities under working condition grouping constraints. Furthermore, it filters effective paths in the hypothetical path library according to sequential consistency rate and peak consistency rate, and fuses them to generate dynamic spatial mapping relationships. It outputs target modules and location credibility markers and uses them for gating and processing, thereby improving positioning stability and interpretability and reducing the risk of misprocessing. Attached Figure Description

[0074] Figure 1 This is a flowchart of a lithium battery thermal runaway analysis method according to Embodiment 1 of the present invention;

[0075] Figure 2 This is a flowchart of the process for obtaining the risk source location results in Embodiment 1 of the present invention;

[0076] Figure 3 This is a diagram showing the module composition of a lithium battery thermal runaway analysis system according to Embodiment 2 of the present invention. Detailed Implementation

[0077] 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. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0078] Example 1

[0079] See Figure 1 As shown, this embodiment provides a method for analyzing the thermal runaway of lithium batteries, including:

[0080] Obtain the multi-source observation sequence of the battery pack during the current runtime.

[0081] The multi-source observation sequence includes: the concentration change sequence output by each gas detector over time, the temperature change sequence output by each temperature detector over time, and the pressure change sequence output by each pressure detector over time; simultaneously, the fan operating condition sequence and the action execution sequence are acquired.

[0082] The gas concentration sequence is obtained from reports from gas detectors. A fixed communication channel is established with the gas detectors to receive measurement data frames from each detector. For each data frame, the gas detector identifier and concentration measurement value are parsed, and the sampling time is written into the record, forming a concentration time sequence arranged by time.

[0083] The temperature sequence is obtained from the temperature detectors. The temperature acquisition unit performs periodic sampling at each temperature measuring point and outputs temperature channel data. After receiving the temperature channel data, it extracts the temperature detector identifier and the temperature measurement value, and writes the corresponding sampling time information. Then, it merges the temperature time sequences of each temperature detector according to the measuring point identifier.

[0084] The pressure sequence is obtained from reports by the pressure detector. The pressure detector outputs a data frame containing the pressure detector identifier, the pressure measurement value, and the sampling time, forming a pressure time series of pressure changes over time.

[0085] The wind turbine operating condition sequence is obtained from the wind turbine controller. The wind turbine controller outputs operating status data frames, which include wind turbine start / stop status, operating gear identifier, speed control status, and fault alarm status. The above operating condition fields are extracted, and corresponding time information is written for each operating condition record to form a wind turbine operating condition sequence in chronological order.

[0086] The sequence of actions to be executed is constructed from the instruction issuance records and execution receipt records of the safety control unit. When issuing instructions for actions such as isolation, power reduction, inerting, and spraying to the action execution mechanism, the action type, target identifier, instruction issuance time, and instruction number are recorded to form an action issuance record. When the action execution mechanism returns an execution receipt, the receipt carries the instruction number, execution status, action start marker, action completion marker, and receipt time. The receipt is associated with and merged with the corresponding issuance record to form a sequence of actions to be executed, including issuance, start, completion, and final status.

[0087] After acquiring the above sequences, a unified processing method is performed on each sequence to form an aligned observation set. The processing includes: performing a unified mapping on time information from different sources so that concentration data, temperature data, pressure data, fan operating condition data, and response action status data fall on the same time axis; resampling each sequence at a unified time granularity so that each time point corresponds to a set of multi-point observation values ​​at the same moment; and concatenating the gas concentration, temperature, pressure, fan operating condition, and response action status at the same time point into an aligned observation record, forming an aligned observation set in chronological order, which is used to describe the changes in multi-point response status during the operation period.

[0088] Based on the aligned observation set, response time-series features are extracted for each detector.

[0089] The method for obtaining the response timing features includes:

[0090] A corresponding response threshold is pre-configured for each type of detector. The response threshold is used to define the boundary for the detector to enter the response state from the non-response state. Among them, the concentration response threshold is used for gas detectors, the temperature response threshold is used for temperature detectors, and the pressure response threshold is used for pressure detectors. The response state refers to the stable response state after the detector observation value meets the corresponding response threshold and the persistence constraint. The persistence constraint refers to the observation value maintaining the state that meets the threshold for N (e.g., 10) consecutive alignment time points, which is used to exclude false triggering of short-term jitter and isolated spike points.

[0091] The observation set is traversed and aligned along the time axis. When an observation value of a detector first meets its response threshold and reaches the persistence constraint, that moment is determined as the arrival time. Starting from the arrival time, the time span during which the observation value remains in the response state is counted to obtain the rising edge duration. Within the continuous time interval corresponding to the response state, the moment when the observation value reaches its maximum value is located, the peak time is determined, and this maximum value is recorded as the peak amplitude. This yields the response timing characteristics of the detector, which include arrival time, rising edge duration, peak time, and peak amplitude. The arrival time characterizes the moment when the response is first confirmed at the detector; the rising edge duration characterizes the sustained intensity of the response from confirmation until it enters a relatively stable phase; the peak time characterizes the moment when the response intensity reaches its highest point; and the peak amplitude characterizes the response intensity level at that highest point.

[0092] After extracting the response timing features of each detector, the operating segments are grouped using the wind turbine operating condition sequence.

[0093] Operating condition identifiers are generated based on the fan operating condition sequence. These identifiers are combined and encoded from the fan operating condition fields, uniquely representing the operating condition category of the fan at a given moment. They ensure that the same operating condition identifier corresponds to the same combination of start / stop status, operating gear identifier, and speed control status. Subsequently, intervals with identical and temporally continuous operating condition identifiers are grouped into the same operating condition group. This operating condition group refers to a time interval within which the fan operating condition remains consistent, used to isolate the influence of different ventilation conditions on the detector arrival time and peak value distribution.

[0094] For each operating condition group, the arrival time, rise time duration, peak time, and peak amplitude of all detectors within that group are summarized. The detectors are then sorted by arrival time as the primary order to obtain the response timing chain for that operating condition group. This response timing chain refers to the sequential order in which each detector enters the response state within the same operating condition group, representing the temporal relationship of response propagation to different sampling points. An actual peak distribution is formed using peak time and peak amplitude as indices. This actual peak distribution represents the relative positional relationship and intensity differences of peak values ​​at each sampling point within the same operating condition group. The response timing chains and actual peak distributions corresponding to each operating condition group are combined to construct a response set, establishing a correspondence between each operating condition group and the response set, forming an operating condition group response map. This operating condition group response map represents the sequential relationship and peak distribution pattern of responses at each sampling point under different wind turbine operating conditions.

[0095] Construct topology change indicators based on operating condition group response maps.

[0096] For each operating condition group, the corresponding set of detector pairs is read from the detector pair configuration table. The detector pair configuration table is generated and stored during the system initialization phase.

[0097] The process of generating the configuration table for the detector includes:

[0098] The interior of the cabinet is divided into multiple candidate areas based on the connection between the compartment boundaries and the venting channel. Each candidate area is assigned a partition identifier, preferably in the form of "R + three-digit serial number", where R is the area type prefix and the three-digit serial number increments from 001 according to the order in which the candidate areas are generated. A candidate area adjacency table is established, recording each candidate area and its connected adjacent candidate areas. A detector attribution table is established, recording the partition identifier to which each detector belongs. Within each candidate area, detectors within that candidate area are paired to generate detector pairs within the area. Between each candidate area and its adjacent candidate areas, detectors within two candidate areas are paired to generate cross-area detector pairs. All detector pairs are deduplicated, and pairs containing deactivated detectors, missing detectors, or abnormal detectors are removed to obtain a set of detector pairs. The set of detector pairs is solidified into a detector pair configuration table, which records the detector pair identifier, the first detector identifier, the second detector identifier, the associated partition identifier, and the operating condition group identifier.

[0099] Among them, the detector pair identifier is used to uniquely index a pair record, preferably using the encoding form of "P + four-digit serial number" and incrementing in the order of generation; the first detector identifier and the second detector identifier are used to identify the two physical detectors corresponding to the pair record and support the subsequent reading of their arrival time difference, peak time difference and peak amplitude ratio components; the associated partition identifier is used to mark the candidate area or cross-regional association relationship to which the detector pair belongs, so as to maintain spatial consistency when sorting by region and statistically analyzing the over-threshold mode; the operating condition group identifier is used to limit the fan pair to participate in the indication calculation under which fan operating condition group, thereby avoiding the overlap of time series differences under different ventilation conditions.

[0100] It should be noted that the detector pairs are of the same type, including gas detector pairs, temperature detector pairs, and pressure detector pairs; the peak amplitude ratio is calculated only within detector pairs of the same type.

[0101] For each pair of detectors within the same operating condition group, three types of relative quantities are calculated as the topology change indication components for that detector pair under that operating condition group.

[0102] Specifically, the difference between the arrival time of the first detector and the arrival time of the second detector is calculated and recorded as the arrival time difference; the difference between the peak time of the first detector and the peak time of the second detector is calculated and recorded as the peak time difference; the ratio of the peak amplitude of the first detector to the peak amplitude of the second detector is calculated and recorded as the peak amplitude ratio. The sign of the arrival time difference is retained during the calculation process to characterize the sequential relationship of the responses of the two detectors.

[0103] The indication components of all detector pairs in the working condition group are spliced ​​together in a preset order to obtain the topology change indication of the working condition group.

[0104] The preset order is determined and fixed during the system initialization phase. The specific implementation method is as follows: First, read the detector pair records in the detector pair configuration table that match the identifier of the working condition group and are enabled; then, sort the detector pair records in ascending order of the partition identifier; within the same partition identifier group, sort in ascending order of the detector pair identifier; for each sorted detector pair record, write the arrival time difference component, peak time difference component, and peak amplitude ratio component corresponding to the detector pair in sequence; and then, according to the above sorting results, concatenate the three types of components of each detector pair to form the topology change indication of the working condition group.

[0105] Subsequently, the topology change indicator is compared item by item with the set of topology change thresholds.

[0106] In this embodiment, the set of topology change thresholds is determined by calibration period data, which corresponds to the working condition group response spectrum under normal conditions where the structure has not undergone deformation, displacement, or occlusion.

[0107] The calibration period forms normal fluctuation ranges for the three types of components of each detector pair. The normal fluctuation ranges include the normal range of arrival time difference, the normal range of peak time difference, and the normal range of peak amplitude ratio. The normal fluctuation ranges are determined by the upper and lower limits of the corresponding components of the calibration period samples. The calibration period samples cover at least the main operating conditions allowed by the system and are collected under conditions without intervention.

[0108] During the operational comparison, it is determined whether the arrival time difference of the detector pair falls within its normal range, whether the peak time difference falls within its normal range, and whether the peak amplitude ratio falls within its normal range. When any component exceeds the corresponding normal fluctuation range, the detector pair is marked as an over-threshold detector pair under this operating condition group.

[0109] If any over-threshold detector pair exists in the operating condition group, the operating condition group is considered to be inconsistent, and the output space alias is marked as present; if no over-threshold detector pair exists, the output space alias is marked as absent.

[0110] When a space alias is marked as existing, the set of candidate discharge paths is retrieved.

[0111] A hypothetical path library is pre-established, which is a set of hypothetical paths pre-established during battery pack deployment. During the deployment phase, based on the structural drawings or 3D layout data of the cabinet / battery pack, the location and connectivity of the compartment boundaries, air duct components, and venting channels are obtained, as well as the installation location and zoning assignment of each detector. On this basis, the spatial connectivity is abstracted into several hypothetical paths pointing from potential risk source areas to venting channels.

[0112] Assuming that each hypothetical path in the path library is associated with a working condition group identifier, the set of detectors involved in the path, the expected response order, and the expected peak amplitude of the detectors, it is used to provide searchable prior candidates when spatial aliasing is present. This allows the response order and peak distribution extracted from the subsequent aligned observation set to be consistently matched with the path template, thereby quickly filtering out effective paths that are more consistent with actual observations. The expected peak amplitude is obtained by clustering and statistically analyzing historical event samples.

[0113] Retrieve a subset of hypothetical paths that match the current working condition group from the hypothetical path library, using the current working condition group identifier as an index;

[0114] For each hypothetical path, check whether the response order of the over-threshold detector pairs is consistent with the expected response order of the hypothetical path. When the number of consistent over-threshold detector pairs is not less than the preset matching number threshold, the hypothetical path is retained, thus obtaining a set of candidate discharge paths.

[0115] After obtaining the set of candidate discharge paths, the consistency of response order and peak position are calculated for each candidate discharge path.

[0116] The expected response sequence of the candidate path is compared one by one with the actual response time sequence chain extracted from the aligned observation set. The comparison unit is the order relationship of the detector pairs. When the expected sequence is consistent with the actual sequence, it is recorded as consistent. When they are inconsistent, they are recorded as inconsistent. The order consistency rate is obtained by comparing the number of consistent detector pairs with the total number of detector pairs participating in the statistics.

[0117] The expected peak amplitude of each detector in the candidate path and the peak amplitude of each detector in the actual response are sorted from largest to smallest. The detectors in the top r% are selected to form the expected peak priority detector set and the actual peak priority detector set, where r is preferably 20%. The expected peak priority detector set and the actual peak priority detector set are aligned and compared. The number of detectors in the intersection of the two is counted, and the ratio of the number of detectors in the intersection to the total number of detectors in the actual peak priority detector set is used as the peak consistency rate.

[0118] When the sequence consistency rate is greater than the preset sequence consistency threshold and the peak consistency rate is greater than the preset peak consistency threshold, the candidate leakage path is determined to be a valid path. If either requirement is not met, the candidate leakage path is removed, and the valid paths are unified into a set of valid paths.

[0119] Dynamic spatial mapping relationships are generated based on the set of valid paths.

[0120] A pre-maintained association table of detector and module location indexes is stored in the form of structured records and solidified during the system deployment phase. The association table includes: detector identifier, neighboring module location index, partition identifier, hypothetical path identifier, and pointing weight field.

[0121] The module location index is used to number and index the spatial location of the battery modules within the battery pack, the partition identifier is used to describe the candidate region to which the detector belongs, and the pointing weight is used to describe the pointing contribution of the detector to a certain module location index under a specific hypothetical path.

[0122] The method for obtaining the pointing weight is as follows:

[0123] Within a certain detector's partition, the three nearest module location indices are selected from near to far in terms of spatial distance; if there are fewer than three in the partition, the nearest module location indices in adjacent partitions that are directly connected to the partition through the discharge channel are selected, until the number of candidate sets reaches three or the connectivity condition is no longer met.

[0124] Based on the detector's role in a hypothetical path, each module location index in the candidate set is assigned an initial weight, which takes one of the following discrete weight levels: When the module location index and the detector are in the same partition, and this partition is the starting partition of the hypothetical path, the weight is high (W1); when the module location index and the detector are in the same partition, but this partition is only a partition passed through by the hypothetical path, the weight is medium (W2); when the module location index and the detector are in adjacent partitions, and there is a direct discharge connection between the two partitions, the weight is low (W3); for other module location indices that do not meet the same partition or allowable connection relationship, the weight is zero. Where W1 is greater than W2, which is greater than W3, with W1 preferably taking 0.8, W2 preferably taking 0.5, and W3 preferably taking 0.2.

[0125] See Figure 2As shown, for each valid path in the valid path set, the expected response order bound to that valid path is read, and the set of detectors involved in that path is determined. Then, based on the association table, for each detector involved in that valid path, its corresponding neighboring module location index and pointing weight under the hypothetical path identifier are read. For the same module location index, the pointing weights from different detectors are accumulated to obtain the in-path pointing strength of that valid path for each module location index. The in-path pointing strength is used to characterize the degree to which each module location index supports the location of a risk source under the premise that the valid path is valid.

[0126] After obtaining the in-path pointing strength of each valid path, the set of valid paths is fused.

[0127] Specifically, a path weight is configured for each valid path. The path weight is obtained by equally weighting the sequential consistency rate and the peak consistency rate of the valid path, so that the path that satisfies both types of consistency contributes more in the fusion, while the path that only satisfies a single consistency contributes relatively less.

[0128] Subsequently, for each module location index, the in-path pointing strength of all valid paths to that module location index is summarized, and weighted and accumulated according to the corresponding path weights to obtain the fused pointing value of that module location index.

[0129] The fused pointing value is used to characterize the overall support of multiple valid paths for the module location index as a risk source location within the current runtime. To facilitate comparability between different runtimes, the fused pointing values ​​of all module location indices are normalized to obtain normalized pointing values, thereby obtaining a dynamic spatial mapping relationship between module location indices and normalized pointing values. This allows the detector response to be interpreted as a distributed pointing to multiple module location indices, rather than a fixed one-to-one mapping.

[0130] Based on the dynamic spatial mapping relationship, the risk source location results are output.

[0131] Specifically, the normalized pointer value corresponding to the location index of each module is read, and the module location index with the largest normalized pointer value is determined as the target module. The module location index with the second largest normalized pointer value is determined and its normalized pointer value is recorded as the second highest normalized pointer value.

[0132] Subsequently, a confidence threshold C and a discrimination threshold D are set. Preferably, C is set to 0.6 and D to 0.1. The normalized pointing value corresponding to the target module is compared with the confidence threshold C. If the normalized pointing value is not less than C, the determination continues; if the normalized pointing value is less than C, the location confidence mark is output as failed.

[0133] Under the premise of passing the confidence threshold judgment, calculate the difference between the normalized pointing value and the second highest normalized pointing value corresponding to the target module, and compare the difference with the discrimination threshold D; when the difference is not less than D, output the positioning confidence mark as passed; when the difference is less than D, output the positioning confidence mark as failed.

[0134] When both location trust markers of the target module are passed, it is output as the risk source location result, which includes the target module and the location trust markers.

[0135] In this embodiment, two threshold determinations are used to establish a hierarchical constraint between "locatable" and "confidential": the first confidence threshold is used to confirm that the current aligned observation set has formed sufficiently strong pointing evidence after effective path fusion, so as to avoid outputting the target and introducing false localization when the overall support is generally low and each candidate position lacks significant pointing; the second discrimination threshold is used to confirm that the target candidate has a sufficient advantage over other candidates, so as to avoid the target switching frequently due to small noise, missing measurement interpolation or short-term fluctuations when the target is too close to the second highest candidate, thereby reducing localization jitter and reducing the risk of false triggering caused by it.

[0136] Gating is performed on trusted markers based on the risk source location results to implement differentiated treatment.

[0137] A set of handling strategies is constructed, organized by action type and including isolation, power reduction, inerting, and spraying strategies, stored in structured record format. Each strategy record includes a strategy identifier, target identifier, action parameter set, action priority, interlocking conditions, and execution receipt requirements. The strategy identifier is generated using a fixed format encoding, preferably ACT + three-digit sequence number. The target identifier includes the target module and a corresponding partition identifier. The partition identifier is used to limit the allowed scope of the action, ensuring that actions such as isolation, inerting, and spraying only apply to execution units within the partition where the target module is located. The action parameter set describes the specific execution parameters. The isolation strategy parameter set includes the disconnection loop number and disconnection duration; the power reduction strategy parameter set includes the derating ratio and derating duration; the inerting strategy parameter set includes the medium type, injection valve number, injection flow setting, and injection duration; and the spraying strategy parameter set includes the spraying loop number, spraying intensity setting, and spraying duration. The execution receipt requirements include a receipt timeout duration and a receipt status enumeration, which includes four statuses: successful issuance, execution started, execution completed, and execution failed.

[0138] Gating is performed based on a trusted location marker, with a trigger threshold as the trigger boundary. This threshold corresponds to the trigger condition for the trusted location marker to pass. The trusted location marker is compared with this trigger condition, and the gating result is output. If the gating result is successful, a first-priority set of actions is activated, and targeted actions are performed on the target module. This first-priority set includes isolation, power reduction, and inertia strategies. The spraying strategy enters a standby state and maintains its binding to the target module.

[0139] During execution, actions are issued sequentially according to priority. The isolation strategy cuts off the energy input and electrical connection to the target module; the power reduction strategy limits the output power level of the circuit containing the target module; and the inerting strategy injects inerting medium into the corresponding discharge channel zone of the target module to suppress flammability risks. Before issuing actions, interlock conditions are checked. These interlock conditions include: the spraying action is prohibited from starting if the isolation action completion marker is not confirmed; the isolation action is prohibited from being released during inerting; and the spraying command is prohibited from being issued when the spraying action is frozen. The next action can only be issued when the interlock conditions are met. During execution, the action issuance time, action start time, action completion time, and execution receipt status are recorded to form a sequence of action executions.

[0140] When the gating result is "failed", a set of resource preservation actions is activated to reduce the risk of mishandling when the location is not reliable.

[0141] The execution rules for the set of resource preservation and disposal actions include:

[0142] The scope of the action is limited to the zone where the target module is located. Spraying commands and targeted inertization commands are prohibited for non-target modules. Only low-intensity risk mitigation actions are issued to the target module, and actions with controllable resource consumption are given priority.

[0143] The set of resource preservation and disposal actions includes ventilation adjustment strategies and conservative power reduction strategies. The ventilation adjustment strategy reduces the risk of flammable gas accumulation in the cabin by adjusting the fan operating conditions. The parameter set of the ventilation adjustment strategy includes the target fan speed and the adjustment holding time. The conservative power reduction strategy is used to reduce the system energy input level.

[0144] For example, when the risk source location result outputs that the target module is module two and the location credibility mark passes, an isolation command and an inertia command are issued to the loop where module two is located. At the same time, the spray strategy is set to standby and bound to module two as the only target. When the risk source location result outputs that the target module is module two and the location credibility mark fails, a ventilation adjustment command and a conservative power reduction command are issued. At the same time, a spray freeze mark is generated and the treatment range is limited to the partition where module two is located to avoid applying high-intensity treatment resources to the wrong module and reduce the risk of misuse of treatment resources.

[0145] Example 2

[0146] See Figure 3 As shown, this embodiment provides a lithium battery thermal runaway analysis system, and implements the lithium battery thermal runaway analysis method, including:

[0147] Data acquisition module: The battery pack acquires multi-source observation sequences during the current operating period through detectors to form an aligned observation set. The multi-source observation sequences include gas concentration sequence, temperature sequence, pressure sequence, fan operating condition sequence, and action execution sequence.

[0148] Response extraction module: Based on the aligned observation set, extract response timing features for each detector. The response timing features include arrival time, rising edge duration, peak time, and peak amplitude.

[0149] The graph generation module uses the wind turbine operating condition sequence to group the operating periods into operating condition groups, generates a response time sequence chain and actual peak distribution for each operating condition group, and forms an operating condition group response graph.

[0150] Alias ​​tagging module: Constructs topology change indicators based on the working condition group response map, compares the topology change indicators with the topology change threshold set to determine over-threshold detector pairs and generates spatial alias tags;

[0151] Discharge path filtering module: When the spatial alias is marked as existing, the module retrieves a subset of hypothetical paths from the pre-built hypothetical path library using the working condition group identifier as the index, and filters the candidate discharge path set based on the consistency between the actual response order of the over-threshold detector pair and the expected response order of each hypothetical path.

[0152] Effective path filtering module: Calculates the sequence consistency rate and peak consistency rate for each candidate leakage path, and compares them with the corresponding thresholds to filter and obtain the effective path set;

[0153] Risk source localization module: Based on the weighted aggregation and normalization of the effective path set, a dynamic spatial mapping relationship is obtained, and the risk source localization result containing the target module and the location credibility marker is output accordingly;

[0154] Risk Management Module: Gating is performed based on the location credibility markers of the risk source location results to implement differentiated management.

[0155] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0156] In conclusion, the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for analyzing thermal runaway in lithium batteries, characterized in that, include: An aligned observation set is formed by acquiring multi-source observation sequences of the battery pack during the current operating period using a detector. The multi-source observation sequences include gas concentration sequences, temperature sequences, pressure sequences, fan operating condition sequences, and action execution sequences. Based on the aligned observation set, response timing features are extracted for each detector, including arrival time, rising edge duration, peak time, and peak amplitude. The operating time segments are grouped into operating condition groups by using the wind turbine operating condition sequence. For each operating condition group, a response time sequence chain and actual peak distribution are generated to form an operating condition group response map. Based on the working condition group response map, a topology change indicator is constructed. The topology change indicator is compared with the topology change threshold set to determine the over-threshold detector pair and generate spatial alias tags. When the spatial alias is marked as existing, a subset of hypothetical paths is retrieved from the pre-built hypothetical path library using the working condition group identifier as the index, and a set of candidate discharge paths is obtained by filtering based on the consistency between the actual response order of the over-threshold detector pair and the expected response order of each hypothetical path. For each candidate leakage path, calculate the sequence consistency rate and peak consistency rate, and compare them with the corresponding thresholds to filter and obtain the effective path set; The dynamic spatial mapping relationship is obtained by weighted summarization and normalization of the effective path set, and the risk source location result containing the target module and the location credibility marker is output accordingly. Gating is performed on trusted markers based on the risk source location results to implement differentiated treatment.

2. The lithium battery thermal runaway analysis method according to claim 1, characterized in that, The method for extracting response time-series features for each detector includes: Pre-configure the corresponding response threshold for each type of detector; The observation set is aligned along the time axis. When the observation value of a certain detector first meets its response threshold and reaches the persistence constraint, the arrival time is determined. The duration of the rising edge is obtained by statistically analyzing the time span during which the observed value remains in the response state, starting from the arrival time. Within the continuous time interval corresponding to the response state, locate the moment when the observed value reaches its maximum value, determine the peak time, and record the maximum value as the peak amplitude.

3. The lithium battery thermal runaway analysis method according to claim 1, characterized in that, The method for obtaining the working condition group response map includes: Based on the wind turbine operating condition sequence, intervals with the same operating condition identifier and continuous time are divided into the same operating condition group. For each operating condition group, the arrival time, rise time duration, peak time and peak amplitude of all detectors in the operating condition group are summarized, and the detectors are sorted in order of arrival time to obtain the response timing chain of the operating condition group. The actual peak distribution is formed by indexing peak time and peak amplitude; The response time series chain corresponding to each working condition group and the actual peak distribution are constructed into a response set. The correspondence between each working condition group and the response set is established to form a working condition group response map.

4. The lithium battery thermal runaway analysis method according to claim 1, characterized in that, The method for constructing topology change indicators includes: For each operating condition group, read the set of detector pairs that match the identifier of that operating condition group from the preset detector pair configuration table; For each pair of detectors within the same working condition group, the difference in arrival time of each pair of detectors is calculated as the arrival time difference, the difference in peak time of each pair of detectors is calculated as the peak time difference, and the ratio of peak amplitude of each pair of detectors is calculated as the peak amplitude ratio. The arrival time difference, peak time difference, and peak amplitude ratio of all detector pairs in the working condition group are sequentially spliced ​​together in a preset order to obtain the topology change indication of the working condition group.

5. The lithium battery thermal runaway analysis method according to claim 1, characterized in that, The method for determining over-threshold detector pairs and generating spatial alias tags includes: During the calibration period, normal fluctuation ranges are formed for the three types of components of each detector pair. The normal fluctuation ranges include the normal range of arrival time difference, the normal range of peak time difference, and the normal range of peak amplitude ratio. During the operational comparison, it is determined whether the arrival time difference of the detector pair falls within its normal range, whether the peak time difference falls within its normal range, and whether the peak amplitude ratio falls within its normal range. When any component exceeds the corresponding normal fluctuation range, it is determined that the component meets the threshold condition, and the detector pair is marked as a threshold detector pair under this working condition group. If any over-threshold detector pair exists in the operating condition group, then the operating condition group is considered to be inconsistent, and the output space alias is marked as present; If no over-threshold detector pair exists, the output space alias is marked as non-existent.

6. The method for analyzing thermal runaway of a lithium battery according to claim 1, characterized in that, The method for obtaining the candidate leakage path set includes: A hypothetical path library is pre-established, which is a set of hypothetical paths pre-established when the battery pack is deployed; Retrieve a subset of hypothetical paths that match the current working condition group from the hypothetical path library, using the current working condition group identifier as an index; For each hypothetical path, check whether the response order of the over-threshold detector pairs is consistent with the expected response order of the hypothetical path. When the number of consistent over-threshold detector pairs is not less than the preset matching number threshold, the hypothetical path is retained, and a set of candidate discharge paths is obtained.

7. The lithium battery thermal runaway analysis method according to claim 1, characterized in that, The method for obtaining the set of valid paths is as follows: The expected response order of the candidate path is compared with the actual extracted response time sequence chain in the aligned observation set. The comparison unit is the order of the detector pairs. When the expected order matches the actual order, it is recorded as consistent; when they do not match, it is recorded as inconsistent. The order consistency rate is obtained by comparing the number of consistent detector pairs with the total number of detector pairs included in the statistics. The expected peak amplitude of each detector in the candidate path and the peak amplitude of each detector in the actual response are sorted from largest to smallest. The detectors in the top r% are used to form the expected peak priority detector set and the actual peak priority detector set. The expected peak priority detector set is aligned and compared with the actual peak priority detector set. The number of detectors in the intersection of the two is counted, and the ratio of the number of detectors in the intersection to the total number of detectors in the actual peak priority detector set is used as the peak consistency rate. When the sequence consistency rate is greater than the preset sequence consistency threshold and the peak consistency rate is greater than the preset peak consistency threshold, the candidate leakage path is determined to be a valid path. If either requirement is not met, the candidate leakage path is removed, and the valid paths are unified into a set of valid paths.

8. The method for analyzing thermal runaway of a lithium battery according to claim 1, characterized in that, The method for obtaining dynamic spatial mapping relationships based on weighted summarization and normalization of the effective path set includes: A pre-maintained association table between detector and module location indexes is provided, which includes detector identifier, neighboring module location index, partition identifier, hypothetical path identifier, and pointing weight. For each valid path in the set of valid paths, read the expected response order bound to that valid path and determine the set of detectors involved in that path; Based on the association table, for each detector involved in the effective path, read the neighboring module location index and pointing weight corresponding to the hypothetical path identifier; For the same module location index, the pointing weights from different detectors are accumulated to obtain the in-path pointing strength of the effective path for each module location index; A path weight is configured for each valid path, and the path weight is obtained by equally weighting the sequential consistency rate and the peak consistency rate of the valid path; For each module location index, the in-path pointing strength of all valid paths to that module location index is summarized, and then weighted and accumulated according to the corresponding path weights to obtain the fused pointing value of that module location index. Normalization is performed on the fused pointer values ​​of all module location indices to obtain normalized pointer values, thereby obtaining the dynamic spatial mapping relationship between module location indices and normalized pointer values.

9. The method for analyzing thermal runaway of a lithium battery according to claim 1, characterized in that, The methods for obtaining the risk source location results include: Read the normalized pointer value corresponding to the location index of each module, and determine the location index of the module with the largest normalized pointer value as the target module. Determine the location index of the module with the second largest normalized pointer value and record its normalized pointer value as the second highest normalized pointer value. Set a trust threshold C and a discrimination threshold D. Compare the normalized pointing value corresponding to the target module with the trust threshold C. If the normalized pointing value is not less than C, continue to make a judgment. If the normalized pointing value is less than C, output the location trust mark as failed. Under the premise of passing the trust threshold judgment, calculate the difference between the normalized pointing value and the second highest normalized pointing value corresponding to the target module, and compare the difference with the discrimination threshold D; When the difference is not less than D, the output positioning confidence mark is passed; when the difference is less than D, the output positioning confidence mark is failed. When both location credibility markers of the target module are passed, it is output as the risk source location result.

10. A lithium battery thermal runaway analysis system, used to implement the lithium battery thermal runaway analysis method according to any one of claims 1-9, characterized in that, include: Data acquisition module: The battery pack acquires multi-source observation sequences during the current operating period through detectors to form an aligned observation set. The multi-source observation sequences include gas concentration sequence, temperature sequence, pressure sequence, fan operating condition sequence, and action execution sequence. Response extraction module: Based on the aligned observation set, extract response timing features for each detector. The response timing features include arrival time, rising edge duration, peak time, and peak amplitude. The graph generation module uses the wind turbine operating condition sequence to group the operating periods into operating condition groups, generates a response time sequence chain and actual peak distribution for each operating condition group, and forms an operating condition group response graph. Alias ​​tagging module: Constructs topology change indicators based on the working condition group response map, compares the topology change indicators with the topology change threshold set to determine over-threshold detector pairs and generates spatial alias tags; Discharge path filtering module: When the spatial alias is marked as existing, the module retrieves a subset of hypothetical paths from the pre-built hypothetical path library using the working condition group identifier as the index, and filters the candidate discharge path set based on the consistency between the actual response order of the over-threshold detector pair and the expected response order of each hypothetical path. Effective path filtering module: Calculates the sequence consistency rate and peak consistency rate for each candidate leakage path, and compares them with the corresponding thresholds to filter and obtain the effective path set; Risk source localization module: Based on the weighted aggregation and normalization of the effective path set, a dynamic spatial mapping relationship is obtained, and the risk source localization result containing the target module and the location credibility marker is output accordingly; Risk Management Module: Gating is performed based on the location credibility markers of the risk source location results to implement differentiated management.

Images