Hydrogen sound print based energy storage battery state detection method and system

Abstract

The present application relates to the technical field of energy storage battery safety early warning, and discloses an energy storage battery state detection method and system based on hydrogen sound print cooperation. A sound print event is used to locate a target battery monitoring unit, and a source association verification path is generated in combination with a hydrogen monitoring point position, a gas connection relationship and a ventilation state. A downstream associated check is started point by point at an actual response time of an upstream hydrogen monitoring point, a hydrogen migration chain is formed in sequence according to a migration direction, the association is terminated when a background monitoring point responds first or the response order is reversed, and a thermal runaway risk early warning is output according to the association between the target battery monitoring unit and the hydrogen migration chain. The present application can distinguish between homologous gas release processes and common environmental hydrogen changes, and can adapt to response time differences caused by air flow.

Description

Technical Field

[0001] This invention relates to the field of energy storage battery safety monitoring and thermal runaway early warning technology, and in particular to an energy storage battery state detection method and system based on hydrogen acoustic signature coordination. Background Technology

[0002] Electrochemical energy storage systems typically consist of multiple battery modules, battery clusters, or battery racks arranged in a concentrated manner. Under conditions of overcharging, internal short circuits, mechanical damage, or localized overheating, individual batteries may experience internal gas generation, casing deformation, activation of pressure relief structures, and gas release to the outside of the battery. Temperature, voltage, and smoke monitoring can reflect battery status; however, in energy storage sites with numerous batteries, relatively enclosed chambers, and continuously changing airflow, relying on a single monitoring parameter is insufficient to simultaneously locate the source of anomalies and confirm their characteristics.

[0003] Chinese patent publication number CN118859018A, entitled "A Two-Level Early Warning Method and System for Thermal Runaway of Lithium Batteries Based on Sound Signals," discloses a scheme for identifying outliers in battery sound signals and extracting time-frequency domain features from abnormal sounds to identify thermal runaway expansion sounds. This scheme can detect battery anomalies using sound signals; however, sound within the energy storage chamber may be affected by equipment operating noise, chamber wall reflections, and sound transmission from the battery rack structure. Therefore, relying solely on sound recognition results makes it difficult to determine whether subsequently detected gas changes originate from the same battery region.

[0004] Chinese patent CN223966681U, entitled "A Thermal Runaway Early Warning Device for Battery Clusters," discloses a method that uses an acoustic fingerprint sensor to identify the sound of a safety valve rupture and a pressure sensor to detect changes in internal air pressure within the battery cluster, outputting an alarm when both signals meet certain conditions. This type of multi-signal joint judgment can reduce the limitations of single-signal judgment, but it focuses on whether the monitored signals collectively meet the alarm conditions, failing to address the spatial attribution of the gas response source under distributed gas monitoring conditions.

[0005] Chinese patent publication number CN114913660A, entitled "Fire Monitoring, Early Warning and Positioning Device for Prefabricated Cabin-Type Electrochemical Energy Storage System," discloses a temperature monitoring module, a gas analysis module, a multi-frequency acoustic positioning array, a distributed pressure sensing module, and a multi-signal fusion early warning module. The gas analysis module monitors hydrogen concentration, and the multi-frequency acoustic positioning array monitors and locates the sounds of gas evolution and safety valve activation. This solution integrates acoustic positioning and hydrogen monitoring for energy storage early warning, but the various monitoring parameters are primarily fused based on thresholds and early warning relationships.

[0006] In energy storage systems with multiple battery monitoring areas and multiple hydrogen monitoring points, hydrogen released from abnormal battery areas needs to migrate along battery rack gaps, internal passages, or air supply and exhaust paths before being detected by different hydrogen monitoring points. Different ventilation conditions can alter the response sequence and interval of hydrogen monitoring points, and changes in ambient hydrogen levels or release from other areas may also cause a particular hydrogen monitoring point to spike. Therefore, directly combining acoustic signature localization results with hydrogen exceedance results within a specific time window may still correlate acoustic signature events from different spatial sources with hydrogen changes, or may miss responses that actually belong to the same release process due to the use of a fixed waiting time.

[0007] Therefore, the technical problem to be solved is: how to determine whether the local battery anomaly obtained by acoustic fingerprint localization and the subsequent hydrogen changes at the distributed hydrogen monitoring point originate from the same gas release process, without relying on dedicated gas flow and centralized sampling devices, and how to maintain the reliability of source attribution and early warning determination when ventilation causes changes in the arrival time and migration direction of hydrogen. Summary of the Invention

[0008] To address the aforementioned problems in existing technologies, this invention provides a method and system for detecting the state of energy storage batteries based on hydrogen acoustic signature coordination, in order to solve the problems of lack of common source verification between acoustic signature localization results and distributed hydrogen response, and the inability of fixed correlation time to adapt to the actual gas migration process.

[0009] To achieve the above objectives, the present invention adopts the following technical solution.

[0010] In a first aspect, the present invention provides a method for energy storage battery state detection based on hydrogen acoustic signature coordination, comprising: acquiring the locations of multiple battery monitoring units, multiple acoustic signature collection points, multiple hydrogen monitoring points, gas connectivity, and ventilation status within an energy storage system, and simultaneously acquiring acoustic signature signals and hydrogen monitoring data; identifying acoustic signature events from the acoustic signature signals, determining the sound source location based on the arrival relationship of the acoustic signature events at multiple acoustic signature collection points, mapping the sound source location to a target battery monitoring unit, and determining the acoustic signature trigger time; generating a source association verification path consisting of multiple hydrogen monitoring points arranged sequentially, according to the gas migration direction extending downstream from the area where the target battery monitoring unit is located, based on the location of the target battery monitoring unit, the location of the hydrogen monitoring points, the gas connectivity, and the ventilation status at the acoustic signature trigger time, and starting from the hydrogen monitoring point not located on the source association verification path. The system identifies background monitoring points among the gas monitoring points. After the soundprint triggering time, it monitors the first hydrogen monitoring point in the source correlation verification path. When the first hydrogen monitoring point enters the hydrogen rising state, the correlation check of the next hydrogen monitoring point is initiated based on the actual response time of the first hydrogen monitoring point. Whenever the checked hydrogen monitoring point enters the hydrogen rising state, the correlation check of the next hydrogen monitoring point is initiated based on the actual response time of that hydrogen monitoring point. When the hydrogen monitoring points in the source correlation verification path respond sequentially according to the gas migration direction, a hydrogen migration chain is formed. When the background monitoring point enters the hydrogen rising state before the first hydrogen monitoring point or the downstream hydrogen monitoring point enters the hydrogen rising state before its upstream hydrogen monitoring point, the correlation between the soundprint event and the source correlation verification path is terminated. Based on the correlation result between the target battery monitoring unit and the hydrogen migration chain, a thermal runaway risk warning information is output.

[0011] Furthermore, a fixed operating noise source area is formed based on the installation location, operating status, and normal sound patterns of the fixed operating equipment. When the sound source location falls within the fixed operating noise source area, the fixed operating noise is excluded by combining the time of the sound pattern event, the status change of the operating equipment, and the boundary of the normal sound pattern. Then, the target battery monitoring unit is determined for the sound pattern events that have not been excluded.

[0012] Furthermore, ventilation status is continuously monitored until the hydrogen migration chain is completed; when ventilation status changes, the upstream hydrogen monitoring points that have responded in sequence are retained, the downstream path that has not yet been completed is reconstructed according to the new ventilation status, and the point-by-point correlation check continues from the last responded hydrogen monitoring point.

[0013] Furthermore, the event reference background level, normal fluctuation amount, and normal upper boundary of each hydrogen monitoring point are formed using hydrogen monitoring data prior to the voiceprint triggering time. The event reference state is then bound to the current voiceprint event and maintained until the association is completed or terminated.

[0014] Furthermore, the hydrogen rise status is determined based on the normalized rise of the path hydrogen monitoring point relative to its own event reference state and the net rise relative to the background monitoring point; when the path hydrogen monitoring point and the background monitoring point exhibit common environmental changes, the path hydrogen monitoring point is not used to advance the hydrogen migration chain.

[0015] Furthermore, while the hydrogen migration chain of an existing voiceprint event has not yet ended, voiceprint events that are relocated to the same target battery monitoring unit will be merged into the existing voiceprint events, retaining the earliest voiceprint trigger time, the frozen event reference state, and the path nodes that have been verified.

[0016] Furthermore, when the sound source localization cannot be uniquely mapped to a single battery monitoring unit, multiple candidate battery monitoring units located within the range of localization uncertainty are retained, source association verification paths are generated and the association states are maintained in parallel, and the target battery monitoring unit is determined based on the unique hydrogen response order in each candidate path.

[0017] Furthermore, when the hydrogen monitoring point first enters the hydrogen rising state and there are no acoustic events in the associated process, the upstream candidate battery monitoring unit is determined in reverse based on the ventilation status and gas connectivity, and the acoustic events with sound sources located in the upstream candidate battery monitoring unit are retrieved from the cached acoustic signals before the hydrogen response time.

[0018] Furthermore, when the source correlation verification paths of different candidate battery monitoring units share a downstream hydrogen monitoring point, the direct upstream hydrogen monitoring point before the shared downstream hydrogen monitoring point for each source correlation verification path is recorded separately; the shared downstream hydrogen monitoring point is only used to advance the source correlation verification path for which the direct upstream hydrogen monitoring point has already responded.

[0019] Secondly, this invention provides a hydrogen-based acoustic signature-based energy storage battery state detection system, comprising a monitoring data acquisition module, an acoustic signature event localization module, a verification path generation module, a migration chain determination module, and an early warning output module. The monitoring data acquisition module acquires acoustic signature signals, hydrogen monitoring data, ventilation status, monitoring locations, and gas connectivity. The acoustic signature event localization module identifies acoustic signature events and determines the target battery monitoring unit and the acoustic signature trigger time. The verification path generation module forms a source-related verification path and background monitoring points based on the location of the target battery monitoring unit, the location of the hydrogen monitoring point, gas connectivity, and ventilation status. The migration chain determination module initiates downstream correlation checks point by point based on the actual response time of the upstream hydrogen monitoring point, forming a hydrogen migration chain or terminating the correlation when the background monitoring point responds first or the response order is reversed. The early warning output module outputs thermal runaway risk early warning information based on the correlation result between the target battery monitoring unit and the hydrogen migration chain.

[0020] Compared with the prior art, the present invention has the following beneficial effects.

[0021] This invention uses voiceprint events to determine the battery source area that needs to be verified, and then uses the location of the target battery monitoring unit, the location of the hydrogen monitoring point, the gas connectivity and ventilation status to form a source association verification path with actual spatial direction, so that the voiceprint signal and the hydrogen signal are not simply combined as two independent over-limit quantities.

[0022] This invention initiates the correlation check of the next downstream monitoring point based on the actual response time of the upstream hydrogen monitoring point, and verifies the process step by step based on the actual gas arrival process. It does not require all hydrogen monitoring points to exceed the limit simultaneously within a unified fixed time window, nor does it rely on the same theoretical migration delay for all energy storage sites.

[0023] This invention uses background monitoring point prior response and migration order reversal to terminate association, so that changes in hydrogen in the common environment, gas release in other areas, and existing hydrogen rise outside the path are not used to confirm the current voiceprint event, forming a voiceprint and hydrogen homology verification relationship that can be traced back to the target battery monitoring unit.

[0024] This invention can be implemented using distributed acoustic fingerprint collection points, hydrogen monitoring points, and existing ventilation status data of the energy storage system. It does not require dedicated sampling pipelines and gas switching mechanisms and can be adapted to energy storage systems with different battery rack layouts and ventilation methods. Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating a method for detecting the state of an energy storage battery based on hydrogen voiceprint collaboration according to the present invention. Detailed Implementation

[0026] The technical solution of the present invention will be further described below with reference to the accompanying drawings. It should be understood that the following embodiments are used to illustrate the implementation process of the present invention and are not intended to limit the scope of protection of the present invention.

[0027] Combination Figure 1 In one embodiment, the present invention provides a method for detecting the state of energy storage batteries based on hydrogen acoustic signature coordination. This method can be applied to containerized energy storage systems, prefabricated cabin energy storage systems, battery rooms, and other energy storage locations with multiple battery clusters or racks, without relying on specific battery models, acoustic sensors, or hydrogen sensors.

[0028] In this embodiment, a battery monitoring unit refers to a battery area that can be spatially distinguished from other battery areas and serves as the target for early warning. One battery monitoring unit can correspond to one battery module, one layer of battery modules within a battery cluster, or a monitoring area composed of several adjacent battery modules. The granularity of the battery monitoring unit division can be determined based on the structure of the energy storage system, the density of acoustic signature collection points, and the desired positioning range. After division, a location identifier is configured for each battery monitoring unit.

[0029] Multiple acoustic signature acquisition points and multiple hydrogen monitoring points are set up within the energy storage system. The acoustic signature acquisition points are used to obtain acoustic waveforms within the battery area; the hydrogen monitoring points directly detect the hydrogen concentration in the air at their location or the response value of the hydrogen sensor, eliminating the need for dedicated sampling pipelines, sampling pumps, or valve matrices for this method. The positions of each acoustic signature acquisition point, hydrogen monitoring point, and battery monitoring unit in the same spatial coordinate system are recorded, along with the gas connectivity between battery rack gaps, internal passages, overhead space, air supply area, and exhaust area.

[0030] All acoustic signature signals, hydrogen monitoring data, and ventilation status data are marked using a unified time reference. This unified time reference can be provided by the same controller or formed by calibrating the clocks of different acquisition devices. Therefore, the occurrence time of acoustic signature events, the response time of each hydrogen monitoring point, and the time when ventilation status changes can be compared on the same timeline.

[0031] The voiceprint referred to in this invention is a distinguishable feature of transient sound within the energy storage battery area, presented in the time domain, frequency domain, or time-frequency domain, and is not a voiceprint used to identify human voices. After continuously acquiring acoustic waveforms at the voiceprint acquisition points, the DC component of the acoustic waveforms is removed, the operating frequency band is preserved, and transient segments are extracted to form the voiceprint event to be located. The features used to identify the voiceprint event may include amplitude envelope, duration, frequency band energy distribution, and the arrival order between different voiceprint acquisition points. The identification conditions can be formed by acoustic data acquired during normal equipment operation and abnormal acoustic samples from the battery, rather than using fixed judgment values ​​that are the same for all energy storage sites.

[0032] When multiple voiceprint acquisition points are set up, the location of candidate sound sources can be determined based on the time difference of voiceprint events arriving at different acquisition points. In one example, the time difference fitting residual of the candidate sound source location is expressed as: ; The estimated location of the sound source is determined according to the following formula: ; In the formula, The coordinates of the candidate sound source are in meters. For the first The location of each voiceprint collection point is indicated by coordinates in meters. Voiceprint collection point serial number The number of voiceprint collection points participating in this positioning is a dimensionless positive integer; For the voiceprint event to reach the first The time of each voiceprint collection point is in seconds; The arrival time of the first voiceprint acquisition point used as a time reference, in seconds; Indicates the position of the candidate sound source and the first The spatial distance between each voiceprint collection point is expressed in meters. The effective sound propagation speed corresponding to the current main sound transmission path, in meters per second; The residuals are fitted to the time difference of the candidate positions, with the dimension being the square of seconds; This is a set of candidate sound source locations formed based on the positions of each battery monitoring unit; To estimate the location of the sound source that minimizes the fitting residual.

[0033] The arrival time of the aforementioned acoustic signature can be obtained from the first arrival point of the synchronously acquired waveform or the peak value of the cross-correlation. The acoustic signature acquisition point location and candidate sound source location set are obtained from the installation location data of the energy storage system. For deployments where airborne sound transmission is the primary method, the airborne sound propagation velocity corrected for the on-site environment is taken as the effective sound propagation velocity of the current main sound transmission path. For deployments where sound transmission is the primary method via the battery rack or cabin structure, the equivalent structural sound propagation velocity obtained through installation and commissioning is taken as the effective sound propagation velocity of the current main sound transmission path. The obtained estimated sound source location is mapped to the corresponding battery monitoring unit to form the target battery monitoring unit, and the acoustic signature trigger time is determined. The fitting residual is only used to determine the sound source location and is not directly used as the result of thermal runaway early warning.

[0034] After identifying the target battery monitoring unit, the ventilation status corresponding to the moment the soundprint is triggered is read. The ventilation status can come from the fan start / stop status, air supply / exhaust mode, or airflow direction feedback of the existing ventilation controller of the energy storage system, or from the existing airflow detection signal of the energy storage site. The location of the target battery monitoring unit, the location of each hydrogen monitoring point, and the current ventilation status are input into the gas connectivity relationship to form a source-related verification path corresponding to the target battery monitoring unit.

[0035] The source correlation verification path is not simply ordered according to the straight-line distance between hydrogen monitoring points and the target battery monitoring unit, but rather determined according to the actual gas channels that hydrogen can pass through as it migrates outward from the area where the target battery monitoring unit is located. Specifically, starting from the hydrogen monitoring point that can first receive the gas released from the target battery monitoring unit, downstream hydrogen monitoring points are sequentially connected along the current airflow direction to obtain an ordered path: ; in, For source association verification path; The hydrogen monitoring point is located close to the target battery monitoring unit and is initially in the receiving position under the current airflow. to Downstream hydrogen monitoring points are set sequentially along the actual gas connection direction; This represents the number of hydrogen monitoring points participating in the current path validation. Each hydrogen monitoring point in the path corresponds to an actual installation location, sensor identifier, and hydrogen monitoring data, rather than an abstract node detached from the spatial structure of the energy storage system.

[0036] Without forced ventilation, the upward migration path of hydrogen can be determined based on the space above the battery monitoring unit, the gaps between adjacent battery racks, and the top connecting area. Under forced ventilation, the downstream sequence is determined based on the airflow direction from the supply end to the exhaust end. The system also identifies background monitoring points from hydrogen monitoring points that are not part of the current source association verification path but can reflect changes in the common environment of the energy storage site.

[0037] After the source association verification path is formed, the first step is to determine... Does it enter the hydrogen rising state after the voiceprint trigger moment? At any moment When entering the hydrogen rising state, instead of immediately associating all downstream hydrogen monitoring points with the voiceprint event, it uses... As The start time of the correlation check. Only when... Only after entering a hydrogen-rising state during its associated inspection period did it... Actual response time Open to The correlation check is then performed. The remaining downstream monitoring points are connected in turn. A hydrogen migration chain is formed when the hydrogen response in the path satisfies the following relationship: ; The actual response delay between adjacent monitoring points can also be expressed as: ; In the formula, For the source association verification path, the first The moment when each hydrogen monitoring point first meets the hydrogen rising state, in seconds; The order of monitoring points in the path; The number of monitoring points to form the current hydrogen migration chain; The actual observed response delay between two adjacent hydrogen monitoring points is expressed in seconds. Each response time comes from hydrogen monitoring data with a uniform time stamp, and the strictly increasing relationship indicates that the hydrogen response progresses from upstream to downstream along the source correlation verification path.

[0038] This method does not require the response delay of each segment to equal a uniform theoretical transport time. Instead, it initiates the correlation check of the next monitoring point based on the actual response time of each upstream hydrogen monitoring point. The end boundary of the correlation check of the next monitoring point can be determined based on the ventilation status of that segment, the length of the gas channel, the response characteristics of the hydrogen sensor, and the data acquisition cycle. If the next monitoring point has not entered the hydrogen rising state within this boundary, the continuation of this path is terminated; if the downstream monitoring point has already entered the hydrogen rising state before its upstream monitoring point, the response sequence is determined to be inconsistent with the current source correlation verification path, and the correlation between this acoustic event and this path is terminated.

[0039] When the background monitoring point enters the hydrogen rise state before the first hydrogen monitoring point, or when changes in the background monitoring point indicate that the current hydrogen rise is due to a change in the common intra-cabin environment, the association between the current acoustic event and the hydrogen response is terminated. Therefore, the acoustic event is responsible for narrowing the monitoring range to a specific source region, while the hydrogen migration chain is responsible for verifying whether hydrogen is indeed propagating outward from that source region along the actual gas channel. These two are not two independent thresholds triggered concurrently.

[0040] Once the acoustic fingerprint event corresponding to the target battery monitoring unit is associated with the hydrogen migration chain, a thermal runaway risk warning is output. The warning information includes at least the identifier of the target battery monitoring unit, the acoustic fingerprint trigger time, the identifiers of the hydrogen monitoring points participating in the migration chain, the response time of each hydrogen monitoring point, and the ventilation status when the migration chain is formed. This information is used to display and record the event process on the energy storage monitoring interface or to transmit it to the safety management device of the energy storage system.

[0041] Taking an energy storage compartment with multiple battery racks and ventilation channels as an example, when a transient acoustic signature appears near a battery monitoring unit, multiple acoustic signature collection points map the sound source to that battery monitoring unit based on the time difference of arrival. If the airflow flows from the battery rack area to the compartment channel and further to the ventilation area, then the hydrogen monitoring point near the battery monitoring unit, the downstream hydrogen monitoring point in the channel, and the hydrogen monitoring point in the ventilation area are selected sequentially to form a source correlation verification path. The hydrogen monitoring point near the target battery monitoring unit enters the rising state first, and then the correlation check of the downstream monitoring point in the channel is initiated; after the downstream monitoring point responds, the correlation check of the ventilation area monitoring point is initiated. When the above monitoring points respond sequentially and the background monitoring point does not respond first, a hydrogen migration chain extending from the target battery monitoring unit to the ventilation area is formed. If the ventilation area monitoring point responds earlier than the downstream monitoring point in the channel, or the background monitoring point responds first, then this group of hydrogen changes is not attributed to the acoustic signature event.

[0042] Furthermore, in one embodiment, to prevent the operating sounds generated by fixed equipment within the energy storage site from being mistaken for battery acoustic events, fixed operating noise source areas are pre-established based on ventilation fans, power conversion equipment, contactors, and other fixed operating equipment. Each fixed operating noise source area is determined by the installation location of the corresponding equipment, the spatial range in which the equipment's operating sounds can form a stable first-to-last relationship, and the sound source localization error, and is associated with the start / stop status or operation time of the equipment.

[0043] Upon detecting a soundprint event, its estimated location is first determined, and then it is checked whether this location falls within a fixed operating noise source area. If the sound source location falls within this area, and the occurrence time of the soundprint event matches the start / stop or operational state of the corresponding equipment, and its duration or frequency band distribution satisfies the soundprint boundary formed by the normal operating data of the equipment, then the soundprint event is marked as fixed operating noise, no target battery monitoring unit is generated, and hydrogen migration chain verification is not initiated. If the sound source location overlaps with the fixed operating noise source area, but the corresponding equipment does not undergo an operational state change matching the soundprint event, or the soundprint characteristics do not fall within the normal soundprint boundary, then the soundprint event is retained, and battery monitoring unit localization continues. Through continuous verification of location, operating state, and soundprint characteristics, it is possible to avoid excluding genuine abnormal battery soundprints simply because the sound source is close to fixed operating equipment.

[0044] Furthermore, in one embodiment, ventilation is maintained while the hydrogen migration chain is not yet complete. When the fan starts or stops, the supply and exhaust mode changes, or the airflow direction changes, the moment when the new ventilation state takes effect is first determined based on ventilation control feedback and airflow detection results, and then the downstream portion of the original path that has not yet been validated is discontinued.

[0045] If the change in ventilation occurs Before responding, the first hydrogen monitoring point and its downstream monitoring points are reselected based on the updated gas flow direction, starting from the target battery monitoring unit. If the ventilation status change occurs in the path... Once each hydrogen monitoring station has responded, the status will be retained. to The verified path, The location is used as the position of the gas that has been reached, and is reconstructed based on the updated ventilation status. The subsequent path. The correlation check of the first downstream monitoring point after reconstruction, no earlier than... The later of the actual response time and the effective time of the new ventilation state. Even if downstream hydrogen monitoring points in the original path subsequently rise, they can only continue to be used to form a migration chain if they are still within the reconstructed path and meet the new response sequence. Therefore, when a wind direction change occurs during event verification, it is not necessary to discard all the upstream physical migration evidence already obtained, while avoiding continuing to interpret the downstream hydrogen response according to the failed old airflow path. Further, in one embodiment, after the acoustic fingerprint event is triggered, a reference state for this event is formed and frozen for each participating hydrogen monitoring point. The acoustic fingerprint trigger time is used as the reference state. Using previous normal monitoring data as reference data, the number of [data points] can be determined according to the following formula. Background level, normal fluctuation, and upper limit of normal for each hydrogen monitoring point: ; ; ; In the formula, This refers to the serial number of the hydrogen monitoring point; This refers to the sample number of the hydrogen monitoring data; For the first The first hydrogen monitoring point Each monitoring value can be in ppm, volume fraction or sensor response units, and the same unit is used in the same calculation process; For the first The collection time for each sample is in seconds; The time of voiceprint triggering is in seconds; For the first The length of the pre-event reference period used by each hydrogen monitoring point, in seconds, is determined by the sensor sampling period and the stability under normal conditions. The background hydrogen level before the event was triggered; This represents the normal fluctuation expressed as median absolute deviation; This refers to the dimensionless boundary coefficient formed based on the normal operation data or sensor calibration data of the hydrogen monitoring point. For the first The normal upper boundary of each hydrogen monitoring point. , and and Use the same units.

[0046] The above reference data comes from hydrogen monitoring records prior to the voiceprint event and therefore does not include hydrogen changes after the event was triggered. , and The data is saved according to the event identifier and remains unchanged until the migration chain of this voiceprint event is established or terminated, without shifting upwards due to a slow increase after the event. For newly included hydrogen monitoring points after path reconstruction, the reference state is still formed from historical data before the same voiceprint trigger time. Once the event ends and the monitoring value returns to normal, the normal reference data is updated. This process can prevent the slow gas release process from being absorbed by the constantly updated background value.

[0047] Furthermore, in one embodiment, when determining whether the hydrogen monitoring point has entered an upward trend, it is first determined whether the monitoring value has exceeded the corresponding normal upper boundary, and then its degree of change relative to the normal fluctuation is normalized. The normalized increase at each hydrogen monitoring point can be expressed as: ; In the formula, For the first The hydrogen monitoring point at the first The normalized increase at each sample time point is a dimensionless quantity. This is the current hydrogen monitoring value; This represents the frozen background level at this monitoring point during this event; This is within the normal range of fluctuations. A positive number corresponding to the resolution of this hydrogen sensor, with units of... The same applies when the normal fluctuation is zero or less than the sensor resolution, preventing the divisor from losing its practical meaning; max indicates taking the larger of the two values. When the monitored value exceeds... If the normalized increase does not return to the normal range within the continuous confirmation period determined by the sensor response characteristics and sampling period, the first sample time that satisfies this continuous state is determined as the response time of the monitoring point. Therefore, a single discrete spike does not directly form a migration chain node. To exclude common environmental changes at energy storage sites, the normalized rise is calculated using the same method for background monitoring points. The net rise of the path monitoring point relative to the background monitoring point is obtained according to the following formula: ; In the formula, This indicates the background monitoring point corresponding to the current source association verification path; For the background monitoring point at the 1st The normalized increase at each sample time point is a dimensionless quantity. For the first The net rise of each path monitoring point relative to the background monitoring point is also dimensionless. Background monitoring point data comes from a hydrogen sensor located in the same energy storage environment as the path monitoring points but not on the current gas migration path of the target battery monitoring unit. If the path monitoring points and background monitoring points rise synchronously, and If the fluctuation boundary between the path monitoring point and the background monitoring point during normal operation is not exceeded, the change is identified as a common environmental change, and no valid response time is generated. If the path monitoring point shows a continuous local net increase relative to the background monitoring point, its response time is used for migration chain verification. If the background monitoring point enters the rising state before the first path monitoring point, the association between this soundprint event and the hydrogen response is directly terminated. The normalized increase and net increase are used to determine a reliable response time, but they are not directly used as early warning results.

[0048] Furthermore, in one embodiment, when the same battery monitoring unit generates multiple intermittent acoustic events during a gas release process, the multiple acoustic events are merged based on their common source. Each time a new acoustic event is detected, its sound source location is first mapped to the battery monitoring unit, and then the battery monitoring unit is queried to determine if there are any unfinished acoustic-hydrogen-related events.

[0049] When a new voiceprint event corresponds to the same battery monitoring unit as a prior voiceprint event, their location areas overlap, and the hydrogen migration chain of the prior event is still in a waiting or point-by-point verification state, the new voiceprint event is merged into the prior event. The merged event continues to use the trigger time of the earliest voiceprint event and the already frozen hydrogen reference state. The occurrence time of subsequent voiceprint events is used as a supplementary record of the same source voiceprint. The hydrogen background level is not recalculated, and the hydrogen monitoring points that have been verified are not reset to a waiting state.

[0050] If the location results of a subsequent acoustic event are transferred to another battery monitoring unit, or if a prior event has ended due to migration chain completion, path mismatch, or background pre-response, a new independent event is established. This process ensures that intermittent venting, structural actions, or multiple consecutive short-duration sounds within the battery are not broken into multiple unrelated warning processes, while also preventing the background from being re-frozen with each sound, thus disrupting the temporal continuity of the same venting process.

[0051] Furthermore, in one embodiment, when the acoustic signature localization result cannot uniquely determine the target battery monitoring unit due to the influence of cabin reflection, sound transmission from adjacent battery rack structures, or the arrangement of acoustic signature collection points, a single localization result is not mandatory. Based on the aforementioned time difference fitting residual, adjacent battery monitoring units whose difference between the fitted residual and the minimum residual lies within the on-site localization uncertainty boundary are retained as candidate target battery monitoring units. The localization uncertainty boundary is determined by the synchronization error of the acoustic signature collection points, the calibration error of the effective sound propagation velocity, and the acoustic calibration data of the energy storage site.

[0052] Each candidate target battery monitoring unit is used as a starting point, and its own source-related verification path is established based on the current ventilation status. All candidate paths share the actual acquired hydrogen monitoring data, but each path separately stores its path order, background monitoring points, verified nodes, and currently waiting nodes. If a candidate path experiences a background monitoring point responding first, a downstream point responding before an upstream point, or the designated next node failing to respond, only that candidate path is terminated, without affecting the continued verification of other candidate paths.

[0053] When a candidate path first forms an upstream hydrogen response with spatial specificity for the candidate cell, and continues to form a migration chain downstream, the candidate cell monitoring cell is identified as the target cell monitoring cell. Thus, the actual migration direction of hydrogen helps eliminate ambiguity in sound source localization, avoiding forced assignment to incorrect adjacent cell regions based solely on minimum residuals under complex acoustic conditions.

[0054] Furthermore, in one embodiment, when a hydrogen monitoring point first enters the hydrogen rising state, and there are currently no voiceprint events still being verified, a reverse voiceprint retrieval is performed starting from that hydrogen monitoring point. First, based on the location of the hydrogen monitoring point and the current ventilation direction, the upstream battery monitoring unit that can deliver gas to the monitoring point is determined in reverse along the gas connectivity relationship, forming an upstream candidate region.

[0055] The acoustic signature backtracking time range is determined based on the gas channel length from the upstream candidate area to the hydrogen monitoring point, the current ventilation status, the hydrogen sensor response characteristics, and the sampling period. Short acoustic segments within this time range that were not promoted to acoustic signature events by the real-time process are then extracted from the cached raw acoustic waveforms. These acoustic segments are then re-localized using their arrival time difference and fixed operating noise is eliminated, retaining only those acoustic segments whose occurrence time is earlier than the hydrogen response time and whose sound source location is within the upstream candidate area.

[0056] When a matching acoustic segment is retrieved, it is established as a voiceprint event, and the actual occurrence time of the acoustic segment is used as the voiceprint trigger time. Hydrogen monitoring points that have entered the rising state are only considered verified monitoring points in the migration chain if they are located on the source association verification path corresponding to the voiceprint event and their response time is later than the voiceprint trigger time; subsequently, the association check of the next downstream monitoring point is initiated based on the actual response time of that monitoring point.

[0057] In this reverse retrieval scenario, the event reference state is obtained from the earlier of the acoustic segment occurrence time and the time when hydrogen first deviates from its normal state, to prevent hydrogen data that has already begun to rise from entering the background level. If no acoustic segment satisfying the positional and temporal relationships is found, the hydrogen anomaly record is retained, but it is not considered a warning event for which a voiceprint-hydrogen migration chain has been established. This process is used to supplement the detection of voiceprint events missed due to weak sound, obstructed propagation, or failure to meet real-time interception conditions.

[0058] Furthermore, in one embodiment, when the source association verification paths of different candidate battery monitoring units converge downstream and share the same hydrogen monitoring point, the direct upstream monitoring point before the shared hydrogen monitoring point and the actual response time of that upstream monitoring point are saved for each candidate path. The response of the shared hydrogen monitoring point cannot be used to determine the source independently of its upstream response.

[0059] Taking the first candidate path sequentially passing through the first branch monitoring point and the shared monitoring point, and the second candidate path sequentially passing through the second branch monitoring point and the shared monitoring point as an example, when the shared monitoring point enters the hydrogen rising state, it first determines whether the first branch monitoring point has responded and started the association check of the shared monitoring point, and then determines whether the second branch monitoring point has responded and started its corresponding association check. Only when the branch monitoring point of a candidate path responds before the shared monitoring point can the response of the shared monitoring point continue to that candidate path.

[0060] If only the shared monitoring point responds, and none of the branch monitoring points along the candidate paths generate a valid response, then the source attribution is not determined solely by the shared monitoring point. If only the first branch monitoring point responds first, the shared monitoring point only advances the first candidate path and is not used to advance the second candidate path. If two branch monitoring points respond first and both meet their respective time sequences, then two associations are retained respectively, and the shared monitoring point is not forcibly assigned to a single event. This avoids the situation where, after multiple gas migration paths converge in the exhaust area, the same response from the shared hydrogen monitoring point is incorrectly used to identify battery monitoring units that have not formed evidence of upstream migration.

[0061] In one system embodiment, the hydrogen-based voiceprint-based energy storage battery state detection system includes a monitoring data acquisition module, a voiceprint event localization module, a verification path generation module, a migration chain determination module, and an early warning output module.

[0062] The monitoring data acquisition module connects to multiple acoustic fingerprint collection points, multiple hydrogen monitoring points, and the existing ventilation status output terminal of the energy storage system. It acquires time-stamped acoustic fingerprint signals, hydrogen monitoring data, and ventilation status data, while simultaneously reading the locations of the acoustic fingerprint collection points, hydrogen monitoring points, battery monitoring units, and gas connectivity. Each hydrogen monitoring point can directly detect the air in its area, eliminating the need for a centralized air extraction structure.

[0063] The voiceprint event localization module extracts voiceprint events from continuous voiceprint signals, determines the arrival time of each voiceprint acquisition point, and maps the estimated location of the sound source to the target battery monitoring unit or a candidate target battery monitoring unit. The event record output by the voiceprint event localization module includes the event identifier, voiceprint trigger time, estimated location of the sound source, and target battery monitoring unit identifier.

[0064] The verification path generation module receives the event records, the location of the hydrogen monitoring point, and the ventilation status corresponding to the soundprint trigger time. Based on the actual gas connectivity, it forms an ordered source-related verification path and determines the background monitoring point corresponding to the current path. When the ventilation status changes, the verification path generation module retains the upstream paths that have already been verified and regenerates the downstream paths that are not yet completed.

[0065] The migration chain determination module saves the target battery monitoring unit, voiceprint trigger time, frozen reference state, source association verification path, background monitoring point, last responded monitoring point and its response time for each event. The migration chain determination module first determines whether the first hydrogen monitoring point has entered the rising state. After its response, it initiates the association check of the next hydrogen monitoring point based on the actual response time, thus forming a hydrogen migration chain sequentially. If the background monitoring point responds first, the response order is reversed, or the next monitoring point fails to respond within the valid check period, the event is set to the association termination state.

[0066] After establishing a correlation between the acoustic fingerprint event and the hydrogen migration chain, the early warning output module outputs information about the target battery monitoring unit and the migration process. These modules can be housed within the same energy storage monitoring controller, or they can be implemented collaboratively by field acquisition devices and upper-level monitoring equipment. Their hardware and communication methods can be configured according to the existing monitoring architecture of the energy storage site, without altering the acoustic fingerprint location source area, the source correlation verification path formed according to ventilation status, or the processing relationship of sequentially forming the hydrogen migration chain based on the actual hydrogen response time.

Claims

1. A method for detecting the state of an energy storage battery based on hydrogen-based acoustic signature coordination, characterized in that, include: The system acquires the locations of multiple battery monitoring units, multiple acoustic fingerprint collection points, multiple hydrogen monitoring points, gas connectivity, and ventilation status within the energy storage system, and simultaneously collects acoustic fingerprint signals and hydrogen monitoring data. Voiceprint events are identified from the voiceprint signals, the location of the sound source is determined based on the arrival relationship of the voiceprint events at multiple voiceprint acquisition points, the location of the sound source is mapped to the target battery monitoring unit, and the voiceprint trigger time is determined. Based on the location of the target battery monitoring unit, the location of the hydrogen monitoring point, the gas connectivity and the ventilation status at the time of the soundprint triggering, a source association verification path is generated by sequentially arranging multiple hydrogen monitoring points according to the gas migration direction extending downstream from the area where the target battery monitoring unit is located, and background monitoring points are determined from the hydrogen monitoring points that are not located on the source association verification path. After the voiceprint triggering time, monitor the first hydrogen monitoring point in the source association verification path. When the first hydrogen monitoring point enters the hydrogen rising state, start the association check of the next hydrogen monitoring point with the actual response time of the first hydrogen monitoring point. Whenever the checked hydrogen monitoring point enters the hydrogen rising state, start the association check of the next hydrogen monitoring point with the actual response time of the hydrogen monitoring point. When the hydrogen monitoring points in the source association verification path respond sequentially according to the gas migration direction, a hydrogen migration chain is formed. When the background monitoring point enters the hydrogen rising state before the first hydrogen monitoring point or the downstream hydrogen monitoring point enters the hydrogen rising state before its upstream hydrogen monitoring point, the association between the voiceprint event and the source association verification path is terminated. Based on the correlation results between the target battery monitoring unit and the hydrogen migration chain, thermal runaway risk warning information is output.

2. The method for detecting the state of an energy storage battery based on hydrogen acoustic signature coordination according to claim 1, characterized in that, The step of identifying voiceprint events from the voiceprint signal and determining the location of the sound source includes: forming a fixed operating noise source area based on the installation location, operating status and normal voiceprints of the fixed operating equipment. When the location of the sound source falls within the fixed operating noise source area, and the occurrence time of the soundprint event matches the start-up or shutdown time or operation time of the operating equipment, and the soundprint event satisfies the soundprint boundary formed by the normal operating data of the operating equipment, the soundprint event is excluded. For voiceprint events that have not been excluded, perform the mapping of the target battery monitoring unit.

3. The method for detecting the state of an energy storage battery based on hydrogen acoustic signature coordination according to claim 1, characterized in that, The ventilation status is continuously monitored before the hydrogen migration chain forms. When the ventilation status changes before the first hydrogen monitoring point responds, the source association verification path is regenerated based on the changed ventilation status. When the ventilation status changes after the upstream hydrogen monitoring point in the source association verification path has responded, the upstream path that has responded in sequence is retained. Starting from the last responding hydrogen monitoring point, the downstream path is reconstructed according to the changed ventilation status. The association check of the reconstructed downstream path is started from the later of the actual response time of the last responding hydrogen monitoring point and the effective time of the new ventilation status.

4. The method for detecting the state of an energy storage battery based on hydrogen acoustic signature coordination according to claim 1, characterized in that, The process of forming the reference state of the rising hydrogen state includes: using hydrogen monitoring data before the soundprint triggering time to determine the event reference background level, normal fluctuation amount and normal upper boundary of each hydrogen monitoring point; The event reference background level, the normal fluctuation amount, and the normal upper boundary are bound to the voiceprint event and remain unchanged until the association between the voiceprint event and the source association verification path is completed or terminated. For newly added hydrogen monitoring points after path reconstruction, the hydrogen monitoring data before the soundprint triggering time is used to form the reference state of the hydrogen monitoring point.

5. The method for detecting the state of an energy storage battery based on hydrogen acoustic signature coordination according to claim 4, characterized in that, The process of determining the rising state of hydrogen includes: forming a normalized rise based on the monitoring values ​​of hydrogen monitoring points, the event reference background level, and normal fluctuations; The net increase is calculated by comparing the normalized increase at the hydrogen monitoring points along the path with the normalized increase at the background monitoring points. When the monitored value at the hydrogen monitoring point along the path continuously exceeds the normal upper boundary, and the net increase exceeds the difference fluctuation boundary formed by the normal operation data, it is determined that the hydrogen monitoring point along the path has entered the hydrogen rising state. When the hydrogen monitoring point along the path and the background monitoring point both rise and the net increase does not exceed the difference fluctuation boundary, the hydrogen migration chain is not advanced with respect to the hydrogen monitoring point along the path.

6. The method for detecting the state of an energy storage battery based on hydrogen acoustic signature coordination according to claim 1, characterized in that, While the association between existing voiceprint events and source-related verification paths has not yet ended, when a new voiceprint event is mapped to the target battery monitoring unit of the existing voiceprint event, the new voiceprint event is merged into the existing voiceprint event, retaining the earliest voiceprint trigger time, the reference state of the existing voiceprint event, and the hydrogen monitoring point that has been verified, and updating the source-related voiceprint record with the occurrence time of the new voiceprint event without restarting the source-related verification path.

7. The method for detecting the state of an energy storage battery based on hydrogen acoustic signature coordination according to claim 1, characterized in that, When the location of the sound source is within the range of uncertainty in the location of multiple adjacent battery monitoring units, the multiple adjacent battery monitoring units are identified as candidate battery monitoring units. Each candidate battery monitoring unit generates a source association verification path and independently maintains the association status. If the background monitoring point responds first, the response order is reversed, or the next hydrogen monitoring point fails to enter the hydrogen rising state in the candidate source association verification path, the candidate source association verification path will be terminated. The target battery monitoring unit is determined based on the hydrogen response with spatial specificity of the candidate battery monitoring unit in the candidate source association verification path that has not been terminated.

8. The method for detecting the state of an energy storage battery based on hydrogen acoustic signature coordination according to claim 1, characterized in that, When the hydrogen monitoring point enters the hydrogen rising state and there are no voiceprint events in the correlation process, the upstream candidate battery monitoring unit is determined in the opposite direction to the gas migration direction based on the location of the hydrogen monitoring point, the ventilation state and the gas communication relationship. Retrieve from the cached voiceprint signals prior to the response time of the hydrogen monitoring point voiceprint events whose sound source location is located in the upstream candidate battery monitoring unit and whose occurrence time is earlier than the response time; When the voiceprint event is retrieved, a source association verification path is generated, and the downstream association check of the hydrogen monitoring point is started at the actual response time of the hydrogen monitoring point that has already responded. If the voiceprint event is not found, no association is established between the hydrogen monitoring point and the voiceprint event.

9. The method for detecting the state of an energy storage battery based on hydrogen acoustic signature coordination according to claim 7, characterized in that, When the source association verification paths of different candidate battery monitoring units share the downstream hydrogen monitoring point, the directly upstream hydrogen monitoring point located before the shared downstream hydrogen monitoring point in each source association verification path is recorded respectively. After the shared downstream hydrogen monitoring point enters the hydrogen rising state, only its directly upstream hydrogen monitoring point has already responded and started the source correlation verification path of the shared downstream hydrogen monitoring point correlation check. If none of the upstream hydrogen monitoring points respond first, the target battery monitoring unit is not determined based on the shared downstream hydrogen monitoring points. When multiple directly upstream hydrogen monitoring points respond first, their respective correlations are retained.

10. A hydrogen-based acoustic signature-based energy storage battery state detection system, used to execute the hydrogen-based acoustic signature-based energy storage battery state detection method as described in any one of claims 1 to 9, characterized in that, include: The monitoring data acquisition module is used to connect multiple acoustic fingerprint collection points, multiple hydrogen monitoring points and ventilation status data sources to obtain acoustic fingerprint signals, hydrogen monitoring data, ventilation status, the location of each battery monitoring unit, the location of each acoustic fingerprint collection point, the location of each hydrogen monitoring point and gas connectivity. The voiceprint event localization module is used to identify voiceprint events from the voiceprint signal, determine the sound source location based on the arrival relationship of the voiceprint event at multiple voiceprint acquisition points, map the sound source location to the target battery monitoring unit, and determine the voiceprint trigger time. The verification path generation module is used to generate a source association verification path including a first hydrogen monitoring point and a downstream hydrogen monitoring point arranged sequentially along the gas migration direction, based on the location of the target battery monitoring unit, the location of the hydrogen monitoring point, the gas connectivity relationship, and the ventilation status at the time of the soundprint triggering, and to determine the background monitoring point. The migration chain determination module is used to start the association check of the next hydrogen monitoring point at the actual response time of the first hydrogen monitoring point after the first hydrogen monitoring point enters the hydrogen rising state, and form a hydrogen migration chain point by point. It is also used to terminate the association when the background monitoring point enters the hydrogen rising state before the first hydrogen monitoring point or the downstream hydrogen monitoring point enters the hydrogen rising state before its upstream hydrogen monitoring point. The early warning output module is used to output thermal runaway risk early warning information based on the correlation results between the target battery monitoring unit and the hydrogen migration chain.

Images