A fire risk detection method of an energy storage system and an energy storage system

Abstract

The application discloses a fire risk detection method of an energy storage system and the energy storage system, and relates to the technical field of energy storage system safety. The application acquires sound wave signals, temperature signals and voltage signals corresponding to a battery module of the energy storage system in a detection period, wherein the sound wave signals are echo signals of the internal reflection of ultrasonic signals of the battery module; sound wave signal features of abnormal frequency signals in the sound wave signals are extracted, and signal features corresponding to the temperature signals and the voltage signals are extracted; in the case that a feature combination of the sound wave signal features, the temperature signal features and the voltage signal features satisfies any one of grade conditions corresponding to at least two risk levels, at least a risk level corresponding to a grade condition satisfied by the feature combination is taken as a risk detection result. The application can realize early fire warning and improve the accuracy of the predicted risk detection result.

Description

Technical Field

[0001] This application relates to the field of energy storage system safety technology, and in particular to a fire risk detection method for an energy storage system and an energy storage system. Background Technology

[0002] Energy storage systems are systems that store energy (such as electrical, thermal, and mechanical energy) using specific technologies and release it when needed, such as battery energy storage systems based on electrochemical energy storage technology. Energy storage systems (especially electrochemical energy storage systems) pose a fire risk due to the thermal runaway characteristics of batteries, high energy density, and complex system integration. To ensure the stable operation of energy storage systems and reduce accident risks, real-time monitoring is necessary for early warning and prevention.

[0003] Traditional fire detection in energy storage systems relies on temperature and smoke sensors. However, these sensors are result-oriented and can only trigger an alarm after thermal runaway of the battery has occurred, generating a large amount of heat and visible smoke. This results in problems such as delayed response and a high false alarm rate.

[0004] Therefore, it is necessary to find detection methods that can accurately predict fire risks in order to intervene in the fire risks of energy storage systems in advance. Summary of the Invention

[0005] In view of the above problems, this application provides a fire risk detection method and energy storage system for energy storage systems, so as to accurately predict the fire risk of energy storage systems. The specific solution is as follows:

[0006] This application provides a fire risk detection method for an energy storage system, the energy storage system comprising: at least one battery module, the fire risk detection method for the energy storage system comprising:

[0007] Within a detection cycle, acquire the acoustic wave signal, temperature signal, and voltage signal corresponding to the battery module, wherein the acoustic wave signal is the echo signal reflected by the internal ultrasonic wave signal of the battery module;

[0008] Extract the acoustic signal features of the abnormal frequency signals in the acoustic signal, and extract the signal features corresponding to the temperature signal and the voltage signal respectively;

[0009] If the combination of sound wave signal features, temperature signal features, and voltage signal features satisfies any one of the preset level conditions corresponding to at least two risk levels, then at least the risk level corresponding to the level condition satisfied by the feature combination shall be taken as the risk detection result.

[0010] In one possible implementation, extracting the acoustic signal features of the abnormal frequency signals in the acoustic signal includes:

[0011] Extract abnormal frequency signals distributed in a preset abnormal frequency band from the acoustic signal;

[0012] Time-frequency analysis is performed on the abnormal frequency signal to obtain the acoustic signal characteristics of the abnormal frequency signal.

[0013] In one possible implementation, extracting the abnormal frequency signal distributed in a preset abnormal frequency band from the acoustic signal includes:

[0014] Extract a first abnormal frequency signal from the abnormal frequency band from the acoustic signal. The first abnormal frequency signal is a signal that integrates the low-frequency trend and high-frequency abnormal features of the abnormal frequency band.

[0015] The first abnormal frequency signal is subjected to bidirectional filtering to obtain the second abnormal frequency signal;

[0016] Gain adjustment is performed on the signal at the center frequency of the abnormal frequency band in the second abnormal frequency signal to obtain the second abnormal frequency signal after gain compensation.

[0017] The inherent resonance features in the second abnormal frequency signal after gain compensation are removed to obtain the abnormal frequency signal. The inherent resonance features are the vibration features of the acoustic wave signal generated by the battery module when there is no fault.

[0018] In one possible implementation, extracting the first anomalous frequency signal from the anomalous frequency band of the acoustic signal includes:

[0019] The acoustic signal is decomposed using a wavelet basis function that matches the transient characteristics of the impact signal of the battery module to obtain multiple wavelet decomposition levels.

[0020] Extract the detail coefficients of the wavelet decomposition level where the abnormal frequency band is located, and the approximation coefficients of the wavelet decomposition level where the lowest frequency band is located in the multiple wavelet decomposition levels. The detail coefficients are a digital sequence, and each number in the digital sequence represents the signal strength of its corresponding signal in the abnormal frequency band.

[0021] Based on the digital sequence, determine the noise reduction threshold for the abnormal frequency band;

[0022] The digital sequence is denoised according to the denoising threshold to obtain the denoised detail coefficients;

[0023] The signal is reconstructed using the denoised detail coefficients and the approximation coefficients to obtain the first abnormal frequency signal.

[0024] In one possible implementation, extracting the signal features corresponding to the temperature signal and the voltage signal respectively includes:

[0025] The temperature signal and the voltage signal are interpolated respectively to obtain interpolated temperature signal and interpolated voltage signal, and the sampling time points in the interpolated temperature signal and the interpolated voltage signal are aligned with the time points of the sound wave signal;

[0026] The signal features of the interpolated temperature signal and the interpolated voltage signal are extracted to obtain temperature feature signals and voltage feature signals.

[0027] One possible implementation also includes:

[0028] If the risk level in the risk detection results reaches a preset level after a preset number of consecutive detection cycles, the fault location in the battery module is located, and a fire extinguishing command is generated based on the fault location and sent to the fire extinguishing device in the energy storage system.

[0029] In one possible implementation, locating the fault location in the battery module includes:

[0030] Acquire sensor signals collected by multiple sensors in the energy storage system during the detection period;

[0031] Perform cross-correlation calculation on every two sensor signals to determine the cross-correlation peak position, and determine the time difference between the time points corresponding to the cross-correlation peak positions in every two sensor signals;

[0032] The fault location is determined based on the time difference between each pair of sensor signals and the position information of each pair of sensors.

[0033] In one possible implementation, the process of detecting whether the combination of features of the acoustic signal, temperature signal, and voltage signal satisfies any one of the preset level conditions corresponding to at least two risk levels includes:

[0034] The features of the acoustic signal, temperature signal, and voltage signal are combined and input into a pre-configured risk detection model to obtain the risk level satisfied by the feature combination.

[0035] The process of training the risk detection model includes:

[0036] Obtain feature combination samples labeled with risk level tags, the feature combination samples including: sound wave signal features, temperature signal features and voltage signal features;

[0037] The feature combination sample is input into the initial model, and the initial model is constructed based on the rule model corresponding to the level conditions of each risk level to obtain the risk level prediction result corresponding to the feature combination sample;

[0038] Calculate the target loss value based on the risk level prediction result corresponding to the feature combination sample and the risk level label;

[0039] The model parameters of the initial model are updated based on the total loss value to obtain a risk detection model that meets the training conditions. The total loss value includes at least the target loss value.

[0040] A second aspect of this application provides an energy storage system, comprising: at least one battery module, and a sensor array distributed in the gaps or on the surface of the battery module, and a signal processing module; wherein the sensor array is used to collect echo signals reflected by ultrasonic signals inside the battery module, as well as temperature signals and voltage signals of the battery module;

[0041] The signal processing module is used to implement the fire risk detection method for the energy storage system described in the first aspect or any implementation thereof.

[0042] In one possible implementation, it further includes: a fire extinguishing device, used to receive a fire extinguishing command sent by the signal processing module, and respond to the fire extinguishing command to extinguish the fire at the fault location in the battery module. The fire extinguishing command is generated by the signal processing module based on the located fault location in the battery module when the risk level in the risk detection result corresponding to a preset number of consecutive detection cycles reaches a preset level.

[0043] In one possible implementation, the fire extinguishing device includes: a pressure relief window and a sprinkler pipe support;

[0044] The pressure relief window is located on the inner left wall of the energy storage cabinet of the energy storage system, and a smoke exhaust fan is installed inside the pressure relief window;

[0045] The spray pipe rack is installed inside the energy storage cabinet on the outside of the battery rack, and the spray pipe rack is connected to an aerosol fire extinguishing agent storage system. The battery rack is used to house the battery modules.

[0046] By means of the above technical solution, this application provides a fire risk detection method and energy storage system for an energy storage system. Considering that the internal physical state of the battery module will undergo a gradual failure process before thermal runaway (fire), this application utilizes the strong penetrating power of ultrasound to collect the echo signal of the battery module to the ultrasound signal by penetrating the outer shell of the battery module. By analyzing the acoustic signal characteristics of abnormal frequency signals in the echo signal, the changes in the internal physical state of the battery module can be detected. This can capture battery failure signs earlier than the thermal runaway stage and realize early fire warning.

[0047] Meanwhile, this application combines changes in the internal physical structure of the battery module (acoustic signal characteristics) and changes in the external environment (temperature signal characteristics, voltage signal characteristics) to jointly detect fire risks, greatly eliminating false alarms caused by environmental interference or self-fault of a single sensor, improving the accuracy of the predicted risk detection results, and greatly ensuring the safe and stable operation of the energy storage system. Attached Figure Description

[0048] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale.

[0049] Figure 1 This is a schematic diagram of the structure of an energy storage system provided in an embodiment of this application;

[0050] Figure 2 This is a schematic diagram of the structure of a battery module provided in an embodiment of this application;

[0051] Figure 3 A bottom view of the metal thermally conductive partition structure provided in the embodiments of this application;

[0052] Figure 4 A cross-sectional view of the battery module provided in an embodiment of this application;

[0053] Figure 5 This is a flowchart illustrating the fire risk detection method for an energy storage system provided in an embodiment of this application.

[0054] Figure label:

[0055] 1-Energy storage cabinet; 2-Battery module; 201-Stainless steel shell; 202-Lithium iron phosphate battery body; 3-Battery rack; 4-Metal thermally conductive partition; 5-Insulation layer; 6-Axial ultrasonic sensor; 7-Radial ultrasonic sensor; 8-Ultrasonic receiver; 9-Temperature sensor; 10-Voltage sensor; 11-Power supply box; 12-Main control cabinet; 13-Pressure relief window; 14-Smoke exhaust fan; 15-Spray pipe rack; 16-Logic expansion module. Detailed Implementation

[0056] The embodiments of this application are described below with reference to the accompanying drawings. The terminology used in the implementation section of this application is for explaining specific embodiments only and is not intended to limit the scope of this application.

[0057] The embodiments of this application will now be described with reference to the accompanying drawings. Those skilled in the art will recognize that, with technological advancements and the emergence of new scenarios, the technical solutions provided in the embodiments of this application are equally applicable to similar technical problems.

[0058] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such terms are interchangeable where appropriate; this is merely a way of distinguishing objects with the same attributes in the embodiments of this application. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, so that a process, method, system, product, or apparatus that comprises a series of elements is not necessarily limited to those elements, but may include other elements not explicitly listed or inherent to those processes, methods, products, or apparatuses.

[0059] This application can be applied to the field of safety testing of energy storage systems. Taking the fire risk detection of battery modules in an energy storage system as an example, the embodiments of this application provide a fire risk detection method and energy storage system for an energy storage system, which solves the problems of lag and high false alarm rate in fire detection of battery modules, as mentioned in the background art.

[0060] The energy storage system for implementing the fire risk detection method of the energy storage system in this application includes: at least one battery module, and a sensor array distributed in the gaps or outer surface of the battery module, and a signal processing module; wherein, the sensor array is used to collect the echo signal reflected by the ultrasonic signal inside the battery module, as well as the temperature signal and voltage signal of the battery module; the signal processing module is used to implement the fire risk detection method of the energy storage system described below.

[0061] The energy storage system utilizes at least one battery module connected in series or parallel. Sensor arrays installed between the battery modules and on the casing collect echo signals (defined as acoustic signals) from within the battery modules in response to ultrasonic signals, as well as the voltage signals of the battery modules and the temperature signals of the energy storage system influenced by the battery modules and other components. It is understood that minute mechanical deformations and internal stress changes within the battery modules generate characteristic acoustic waves; therefore, the acoustic signals provide information about changes in the internal physical state of the battery modules. The voltage signals reflect the electrochemical transformations within the battery modules. Based on this, the signal processing module, combined with the external ambient temperature of the battery modules, jointly detects fire risks in the energy storage system. This significantly reduces false alarms caused by environmental interference or malfunctions of a single signal source, improving the reliability and accuracy of the predicted risk detection results.

[0062] Reference Figure 1 , Figure 2 , Figure 3 and Figure 4 The schematic diagram of the energy storage system provided in this application illustrates one possible implementation of the energy storage system. The energy storage system consists of multiple battery modules 2 connected in series or parallel inside the energy storage cabinet 1.

[0063] Specifically, the energy storage cabinet 1 has a layered battery rack 3 fixed inside. Metal heat-conducting partitions 4 are respectively set on the inside of the battery rack 3 and between each battery module 2. The battery module 2 includes a stainless steel shell 201 and a lithium iron phosphate battery body 202 located inside the stainless steel shell 201. The inside of the stainless steel shell 201 is filled with a ceramic fiber structure heat insulation layer 5.

[0064] The sensor array may include an ultrasonic sensor array, comprising axial ultrasonic sensors 6 and radial ultrasonic sensors 7 fixedly arranged in a three-dimensional star-shaped network on the stainless steel housing 201. Two axial ultrasonic sensors 6 are located at the top and bottom centers of the stainless steel housing 201, respectively; the radial ultrasonic sensors 7 are mounted at the front center of the stainless steel housing 201. An ultrasonic receiver 8 is fixedly mounted on the inner side of the battery rack 3 via a cantilever bracket, and the ultrasonic receiver 8 is positioned between the upper and lower battery modules 2. Temperature sensors 9 are fixedly mounted on the top and left and right sides of the stainless steel housing 201, and voltage sensors 10 are fixedly mounted on the front of the stainless steel housing 201, with the voltage sensors 10 connected in parallel to a high-resistance voltage divider circuit for each individual battery cell.

[0065] In this embodiment, the signal processing module is integrated into the main control cabinet 12 of the energy storage system, and the fire risk detection method of the energy storage system is implemented by the main control cabinet 12. In one possible implementation, the energy storage system can also transmit the acoustic signals, temperature signals, and voltage signals collected by the sensor array to a server or terminal, and the server or terminal can remotely implement the fire risk detection method of the energy storage system.

[0066] Reference Figure 5 , Figure 5 A flowchart illustrating a fire risk detection method for an energy storage system provided in this application embodiment is shown below. Figure 5 As shown in the embodiment of this application, a fire risk detection method for an energy storage system may include steps S110 to S140, which are described in detail below.

[0067] Step S110: Obtain the acoustic wave signal, temperature signal and voltage signal corresponding to the battery module within a detection cycle. The acoustic wave signal is the echo signal reflected by the internal ultrasonic wave signal of the battery module.

[0068] The detection cycle can be freely set according to the operating status of the energy storage system and the detection needs of the operation and maintenance personnel. It can periodically acquire acoustic signals, temperature signals and voltage signals from the sensor array within each detection cycle. Optionally, the sensor array can transmit the various signals collected to the signal processing module or remote server in real time. The signal processing module or remote server can then acquire the acoustic signals, temperature signals and voltage signals within the detection cycle based on the timestamp of the signal.

[0069] Step S120: Extract the acoustic signal features of the abnormal frequency signals in the acoustic signal, and extract the signal features corresponding to the temperature signal and voltage signal respectively.

[0070] Understandably, the core of step S120 is to perform in-depth processing on the original multimodal signals collected in step S110, and extract a set of key signal features from the sound wave, temperature and voltage signals that can accurately and efficiently characterize the current state of the battery module, especially the abnormal state, so as to provide data basis for the next step of fire risk detection.

[0071] Specifically, the original acoustic signal acquired in step S110 is a complex signal containing multiple frequency components. Under normal conditions, the battery's internal structure is stable, resulting in a relatively stable signal characteristic in the generated acoustic signal. However, when a fault occurs inside the battery, such as internal structural deformation or bubble formation, the echo contains frequency components different from those in the normal state. These frequency components in the original acoustic signal that show significant changes compared to the stable signal characteristics are called "abnormal frequency signals."

[0072] In this embodiment, signal processing algorithms, such as Fourier transform and wavelet transform, can be used to focus on the abnormal frequency components in the acoustic signal caused by battery failure, and to quantify and calculate the signal features (defined as acoustic signal features) that can describe the abnormality from these abnormal components. The acoustic signal features can include time-domain features, frequency-domain features, and time-frequency features, such as frequency offset and amplitude attenuation rate.

[0073] In parallel, signal features of both temperature and voltage signals are extracted. The temperature signal features can characterize the thermal behavior trends of the energy storage system or battery module, such as the rate of temperature rise, temperature variance, and maximum temperature. Similarly, the voltage signal features can characterize the consistency and uniformity of the voltage signal, such as voltage standard deviation, voltage range, and voltage coefficient of variation.

[0074] Step S130: Determine whether the combination of features of the acoustic signal, temperature signal, and voltage signal satisfies any one of the preset level conditions corresponding to at least two risk levels. If yes, proceed to step S140; otherwise, continue the detection.

[0075] Step S140: At least the risk level corresponding to the level conditions satisfied by the feature combination shall be taken as the risk detection result.

[0076] The system receives the results from step S120, namely the signal features corresponding to the sound wave, temperature, and voltage signals, and combines the sound wave signal features, temperature signal features, and voltage signal features as a feature combination to make a decision based on these three signal features.

[0077] Each risk level has pre-defined risk level conditions, which specify the conditions that various signal characteristics must meet under that risk level. For example, the frequency offset of the acoustic signal must be greater than 0.5%, while the temperature rise rate must be greater than 0.5°C / min or the voltage standard deviation must be greater than 0.05V.

[0078] For each risk level, the system determines whether the feature combination meets the corresponding level conditions. If each risk level indicates a fire risk in the energy storage system or battery module, and the feature combination meets any one of the level conditions, then the battery module in the current testing cycle has a fire risk. At least based on the risk level corresponding to that condition, a risk detection result for this testing cycle is generated. Conversely, if the feature combination does not meet any level condition, it indicates no fire risk, and testing continues until the risk detection result is determined to be no fire risk.

[0079] In one possible implementation, the absence of fire risk is also considered a risk level. That is, the corresponding risk-free level conditions are generated based on the characteristics of various signals under the condition of no fire risk. Therefore, even if there is no fire risk in the energy storage system or battery module, its combination of characteristics must satisfy one of the multiple level conditions, and the risk detection result is generated based on the risk-free level corresponding to the risk-free level conditions.

[0080] The following example illustrates the risk levels and their corresponding conditions. In this example, four levels are defined: a) Risk level y=0, indicating that the energy storage device or battery module is normal and there is no fire risk; b) Risk level y=1, indicating a low fire risk, and the generated risk detection results may include warning information and instructions to control the ventilation of the energy storage device; c) Risk level y=2, indicating a moderate fire risk, and the generated risk detection results may include alarm information, and if necessary, instructions to control power outage; d) Risk level y=3, indicating a high fire risk, and the generated risk detection results may include warning information about battery module malfunction and instructions to control fire extinguishing.

[0081] Specifically, the risk level y=1 can be defined as follows: among the signal characteristics of multiple detected modes (ultrasound, temperature, voltage), only the signal characteristics of a single mode exceed its corresponding threshold. For example, the frequency offset Δf of the acoustic signal > 0.5%, or the temperature rise rate dT / dt of the temperature signal > 1℃ / s.

[0082] The risk level y=2 corresponds to the following condition: among the signal characteristics of multiple detected modes (ultrasound, temperature, voltage), the signal characteristics of any two modes exceed their corresponding thresholds. For example: Δf>1.2% and dA / dt>5dB / s, σV>0.1V and dT / dt>3℃ / s.

[0083] The risk level y=3 corresponds to the following conditions: the signal characteristics of multiple detected modes (ultrasound, temperature, voltage) all exceed their corresponding thresholds, such as: detecting a 20kHz subharmonic and Δf>2%, dT / dt>5℃ / s or σV>0.2V.

[0084] The risk level y=0 corresponds to the condition that none of the detected signal characteristics of multiple modes (ultrasound, temperature, voltage) exceed their corresponding thresholds. Therefore, in this embodiment, the risk level is directly proportional to the type of signal characteristic exceeding the threshold.

[0085] In summary, the fire risk detection method for energy storage systems provided in this application takes into account that the internal physical state of the battery module will undergo a gradual failure process before thermal runaway (fire). Therefore, this application utilizes the strong penetrating power of ultrasound to collect the echo signal of the battery module to the ultrasonic signal by penetrating the battery module's shell. By analyzing the acoustic signal characteristics of abnormal frequency signals in the echo signal, the changes in the internal physical state of the battery module can be detected. This method can capture battery failure signs earlier than the thermal runaway stage and achieve early fire warning.

[0086] Meanwhile, this application combines changes in the internal physical structure of the battery module (acoustic signal characteristics) and changes in the external environment (temperature signal characteristics, voltage signal characteristics) to jointly detect fire risks, greatly eliminating false alarms caused by environmental interference or self-fault of a single sensor, improving the accuracy of the predicted risk detection results, and greatly ensuring the safe and stable operation of the energy storage system.

[0087] Next, through the following embodiments, other possible implementations of the fire risk detection method for an energy storage system provided in this application will be described in detail.

[0088] In one possible implementation, step S120, which involves extracting the acoustic signal features of abnormal frequency signals in the acoustic signal, may include:

[0089] First, the first anomalous frequency signal of the anomalous frequency band is extracted from the acoustic signal, including: decomposing the acoustic signal using a wavelet basis function that matches the transient characteristics of the impact signal of the battery module to obtain multiple wavelet decomposition levels; extracting the detail coefficients of the wavelet decomposition level where the anomalous frequency band is located, and the approximation coefficients of the wavelet decomposition level where the lowest frequency band is located among the multiple wavelet decomposition levels, wherein the detail coefficients are a digital sequence, and each number in the digital sequence represents the signal strength of its corresponding signal in the anomalous frequency band; determining the denoising threshold of the anomalous frequency band based on the digital sequence; denoising the digital sequence according to the denoising threshold to obtain the denoised detail coefficients; and reconstructing the signal using the denoised detail coefficients and approximation coefficients to obtain the first anomalous frequency signal.

[0090] The embodiments of this application employ wavelet packet decomposition algorithm to extract abnormal frequency signals (defined as the first abnormal frequency signal) from the acoustic signal in the abnormal frequency band.

[0091] First, determine the wavelet basis function that matches the transient characteristics of the battery module's impact signal. The transient characteristics can include features such as signal abrupt changes and frequency variations. Selecting the matching wavelet basis function to process the acoustic signal can better capture the transient characteristics in the acoustic signal and accurately extract abnormal frequency signals.

[0092] The acquired acoustic signal is decomposed using the selected wavelet basis function. Through wavelet decomposition, the acoustic signal is divided into multiple sub-signals of different frequency bands. These sub-signals are distributed at different wavelet decomposition levels, each level representing signal characteristics at a different scale. In this embodiment, the db4 wavelet basis function is selected to perform decomposition on the acoustic signal, resulting in five wavelet decomposition levels.

[0093] From multiple wavelet decomposition levels, the level containing the abnormal frequency band is determined, and the detail coefficients of that level are extracted. These detail coefficients reflect the high-frequency detail information of the acoustic signal in that band. The abnormal frequency band can be determined based on prior knowledge of the battery module, such as abnormal frequency bands in the ultrasonic signals generated or reflected when experiments indicate a fire risk in the battery module, such as diaphragm damage or air bubbles. Based on this, the abnormal frequency signal of the abnormal frequency band is extracted from the original acoustic signal obtained in step S110, which greatly narrows the scope of subsequent signal analysis and eliminates signals unrelated to battery faults.

[0094] Simultaneously, the lowest frequency band is identified among multiple wavelet decomposition levels, and the approximation coefficients of that level are extracted. These approximation coefficients represent the low-frequency components of the signal, containing the main trend and energy information of the acoustic signal. In this embodiment, based on the frequency bands corresponding to the five wavelet decomposition levels, the detail coefficients of the 35-45kHz abnormal frequency band in the second level are extracted, and the approximation coefficients of the fifth level are also extracted.

[0095] The detail coefficient is actually a digital sequence. Each number in the sequence represents the signal strength of the original acoustic signal at a certain moment or location in the abnormal frequency band. By analyzing the digital sequence, we can understand the changes in the signal in the abnormal frequency band.

[0096] Based on this, a denoising threshold is determined for the digital sequence using methods such as statistical characteristics and empirical formulas. This threshold is applied to the digital sequence and compared with each digit to distinguish between valid and noise components in abnormal frequency bands. For example, signals below the threshold are considered noise and are set to zero or otherwise appropriately processed; while signals above the threshold are considered abnormal signals and are retained. This achieves the suppression of noise components in the original detail coefficients.

[0097] In this embodiment, the SUREShrink adaptive thresholding method is used for the detail coefficients of the second layer. Referring to the following formula (1), the threshold T is determined using the noise standard deviation σ. The larger the noise standard deviation σ, the larger the threshold T, which can play a role in delineating the boundary between noise and useful signal. The numbers / coefficients below the threshold T in the detail coefficients are set to zero to filter out noise in the abnormal frequency band, while the numbers / coefficients above the threshold T are retained and their amplitudes are reduced to suppress noise peaks, thus obtaining the denoised detail coefficients. .

[0098] (1)

[0099] Where σ is the noise standard deviation; N is a constant value, taking the value 2048; sign(D2) is a number / coefficient in the numerical sequence of detail coefficients of the second layer.

[0100] Furthermore, utilizing the detail coefficients after denoising The signal is reconstructed using the approximation coefficients of the fifth layer to obtain the first abnormal frequency signal of the abnormal frequency band in the acoustic signal. The first abnormal frequency signal is a signal that integrates the low-frequency trend and high-frequency abnormal characteristics of the abnormal frequency band.

[0101] Based on this, in one possible implementation, after extracting the first abnormal frequency signal of the abnormal frequency band from the acoustic signal, the first abnormal frequency signal is subjected to bidirectional filtering to obtain the second abnormal frequency signal; the gain of the signal at the center frequency of the abnormal frequency band in the second abnormal frequency signal is adjusted to obtain the second abnormal frequency signal after gain compensation; the inherent resonance feature in the second abnormal frequency signal after gain compensation is removed to obtain the abnormal frequency signal, where the inherent resonance feature is the vibration feature of the acoustic signal generated by the battery module when there is no fault.

[0102] First, a digital filter, employing a zero-phase filtering method, is used to perform bidirectional filtering on the first abnormal frequency signal to eliminate phase distortion. Details of the bidirectional filtering process are provided below:

[0103] Referring to equation (2), the first abnormal frequency signal is positively filtered using a difference equation, while simultaneously introducing a positive phase shift. At this time, the first abnormal frequency signal after forward filtering (hereinafter referred to as the output signal after forward filtering) will be phase shifted due to the forward filtering operation.

[0104] (2)

[0105] in, This represents the value of the output signal after forward filtering (the first abnormal frequency signal after forward filtering) at discrete time point n; n is the discrete time index; k is the index variable for summation; M and N are the coefficient orders of the filter, M corresponds to the order of the numerator coefficients, and N corresponds to the order of the denominator coefficients, both of which are non-negative integers; b k These are the numerator coefficients of the filter. This represents the value of the input signal (the first abnormal frequency signal) at discrete time points, i.e., the delayed sample of the input signal; a k These are the denominator coefficients of the filter; This represents the historical value of the forward filter output signal at discrete time point [nk], i.e., the delayed sample of the output signal.

[0106] Referring to the following formula (3), the output signal after forward filtering is time-reversed to prepare for subsequent reverse filtering to cancel the phase shift.

[0107] (3)

[0108] in, This represents the value of the signal at discrete time point n after time reversal. This represents the value of the forward filtered output signal at the discrete time point [-n].

[0109] Referring to equation (4), the signal after time reversal is subjected to the same filtering structure as the forward filtering (reverse filtering), introducing a reverse phase shift. The phase shift is equal in magnitude but opposite in direction to the positive phase shift, thus initially offsetting the phase shift.

[0110] (4)

[0111] Where z[n] represents the value of the output signal after inverse filtering at discrete time point n; y rev [nk] represents the value of the time-reversed signal at the discrete time point [nk], which is the delayed sample of the input signal (time-reversed signal) of the inverse filter; z[nk] represents the historical value of the output signal of the inverse filter at the discrete time point [nk], which is the delayed sample of the output of the inverse filter; k, M, and N are the inherent parameters of the filter.

[0112] Finally, the inverse-filtered signal is inverted again, referring to the following equation (5), to restore the original time sequence of the signal. The total phase shift in the signal after the second time inversion is: This ultimately achieves zero-phase filtering.

[0113] (5)

[0114] Where, x out[n] represents the value of the final output signal at discrete time point n after the second inversion; z[-n] represents the value of the reverse filtered output signal at discrete time point [-n].

[0115] The entire bidirectional filtering process uses a combination of "forward filtering - time reversal - reverse filtering - double reversal" to compensate for the phase distortion generated during the filtering process by utilizing the symmetry of phase shift, thus achieving the effect of zero-phase filtering of the first abnormal frequency signal.

[0116] Furthermore, in the first anomalous frequency signal after bidirectional filtering (defined as the second anomalous frequency signal), a gain compensation insertion loss is added at the center frequency of the frequency band, such as adding a -0.2dB gain compensation at 40kHz, the signal center of the second anomalous frequency signal (35-45kHz). Based on this, the resonance characteristics of the battery module structure are obtained with reference to the following formula (6). .

[0117] (6)

[0118] in, The spectral value of the i-th signal sample at frequency f represents the signal characteristics of the sample at a specific frequency; N represents the number of samples or the number of terms to be summed, used to determine the total number of signal samples involved in the calculation.

[0119] Furthermore, referring to the following formula (7), through baseline correction, the second abnormal frequency signal can highlight the "abnormal vibration" component superimposed under the fault state by subtracting the baseline (inherent resonance) from the remaining signal, thereby achieving accurate identification of abnormal energy in the acoustic signal and extracting the acoustic signal characteristics of the abnormal frequency signal in the acoustic signal.

[0120] (7)

[0121] Where X(f) represents the spectral value of the real-time signal (the second abnormal frequency signal) before correction at frequency f, X corrected (f) represents the spectral value of the real-time signal (abnormal frequency signal) after correction at frequency f, and α is the adaptive coefficient.

[0122] In one possible implementation, after extracting the abnormal frequency signals distributed in a preset abnormal frequency band from the acoustic signal, time-frequency analysis is performed on the abnormal frequency signals to obtain the acoustic signal characteristics of the abnormal frequency signals.

[0123] Time-frequency analysis is performed on the abnormal frequency signal, and quantified feature values ​​that can be used for risk logic judgment are extracted from the results. These feature values ​​are used as acoustic signal characteristics for subsequent risk level assessment. The acoustic signal characteristics include at least: frequency offset, amplitude attenuation rate, and phase burst events.

[0124] Specifically, the process of calculating the frequency offset may include:

[0125] Based on the duration of the acoustic signal, 256 points are selected per frame, with 50% overlap. A continuous data stream is stored through a wake-up buffer, and the data of each frame is multiplied by a window function point by point to prevent spectral leakage. Furthermore, the number of points for the Fourier transform (FFT) is set to 4096, and the original 256 points are padded with zeros to 4096 points. The spectrum, including the amplitude spectrum, is calculated, referring to equation (8).

[0126] (8)

[0127] in, This represents the amplitude spectrum value of the signal at a discrete frequency point k; It represents the real part of the spectral component of the signal at the discrete frequency point k; This represents the imaginary part of the spectral component of the signal at a discrete frequency point k, where k is the index of the discrete frequency point.

[0128] Logarithmic transformation of the amplitude spectrum: , .in, This represents the signal value after logarithmic transformation. The reference voltage is used. Furthermore, in the logarithmically transformed L(k), a local maximum is searched within the range of 35-45kHz to obtain the amplitude V. peak With the corresponding frequency f peak By performing cubic spline interpolation near the peak, the frequency resolution is improved to the 0.1Hz level, and the frequency offset is calculated. .

[0129] Specifically, the process of extracting the amplitude attenuation rate of abnormal frequency signals may include:

[0130] Using the extracted abnormal frequency signals as the processing object, a time-frequency matrix was constructed with a window length of 256 points, a sliding step size of 128 points, and a frequency range of 0-100kHz. The matrix dimension was set to... Where Ntotal is the total number of sampling points, and the positive frequency part of the FFT has 2049 points. Furthermore, the time-frequency matrix is ​​dynamically controlled, as shown in equation (9).

[0131] (9)

[0132] in, , These represent the long-term statistical mean and standard deviation for each frequency point.

[0133] Calculate the amplitude attenuation rate by referring to the following formula (10).

[0134] (10)

[0135] Where t0 is the starting time point for calculating the amplitude decay rate, t0+T is the ending time point for calculating the amplitude decay rate, Δt is the time step for calculating the amplitude change, and L(t,f) is the time step for calculating the amplitude change. c ) represents the time t and frequency f c The amplitude value at point L(t+Δt,f) c ) represents the frequency f at time t+Δt. c The amplitude value at f c =40kHz, time window T=500ms.

[0136] It is understandable that calculating the amplitude attenuation rate and dynamically controlling the time-frequency matrix are strongly correlated in the preprocessing and core calculation of the abnormal frequency signal amplitude attenuation rate extraction process. This addresses the problem of excessive signal strength differences and noise interference overwhelming useful information in the time-frequency matrix constructed based on abnormal frequency signals. For a specific frequency within a fixed time window (T=500ms), the degree of amplitude attenuation over time (i.e., amplitude attenuation rate) is calculated. This attenuation rate is a key characteristic for judging the degree of battery module failure; for example, faster amplitude attenuation may indicate more severe damage to the internal battery structure and a higher fire risk. Therefore, the calculation of the amplitude attenuation rate depends on the amplitude data of the 40kHz frequency point in the time-frequency matrix at different times. If the time-frequency matrix is ​​not dynamically controlled, it will directly lead to distortion in the attenuation rate calculation, while dynamic range control can solve this distortion problem.

[0137] Specifically, the process of extracting phase burst events from abnormal frequency signals may include:

[0138] The analytic signal of the anomalous frequency signal is obtained by using the Hilbert transform. , where H is the Hilbert operator.

[0139] Calculate the instantaneous phase, expressed as .

[0140] Calculate the phase difference between adjacent time points, expressed as: ,when When the angle is greater than 30°, it is marked as a phase burst event. Im(z(t)) represents the imaginary part of the analytic signal z(t), which serves as the orthogonal reference signal for instantaneous phase calculation. It works in conjunction with the real part to extract the phase information of the signal through arctangent operation. Re(z(t)) represents the real part of the analytic signal z(t), which serves as the reference signal for instantaneous phase calculation and reflects the time-domain amplitude change of the abnormal frequency signal.

[0141] In one possible implementation, extracting signal features corresponding to the temperature signal and voltage signal respectively includes: performing interpolation processing on the temperature signal and voltage signal respectively to obtain interpolated temperature signal and interpolated voltage signal, wherein the sampling time points in the interpolated temperature signal and interpolated voltage signal are aligned with the time points of the sound wave signal; and extracting signal features from the interpolated temperature signal and interpolated voltage signal to obtain temperature feature signal and voltage feature signal.

[0142] Optionally, based on the IEEE 1588PTP protocol, to control the clock synchronization error between the BMS battery system and each sensor node to <1ms, a circular buffer is established, and data packets are aligned according to the maximum delay. Furthermore, interpolation processing, such as linear interpolation and nonlinear interpolation, is performed on the synchronized voltage and temperature signals respectively, to elevate the interpolated voltage and temperature signals to frequency alignment with the acoustic signal.

[0143] Furthermore, based on the CUSUM algorithm, the temperature rise rate of multiple consecutive sampling points is extracted from the interpolated temperature signal as a temperature signal feature to detect abnormal temperature changes in the energy storage system or battery module.

[0144] Optionally, referring to the following formula (11), the voltage V of different cells in the battery module can be extracted from the interpolated voltage signal. i Calculate the standard deviation of voltage As a standard value for voltage signals, it is used to analyze the voltage consistency between different individual units.

[0145] (11)

[0146] Where N represents the total number of individual voltages extracted, V i This represents the voltage value of the i-th cell. This represents the average voltage of all individual cells.

[0147] In one possible implementation, the acoustic signal features, temperature signal features, and voltage signal features are normalized, and the normalized acoustic signal features, temperature signal features, and voltage signal features are concatenated to generate a joint feature vector. Weights are assigned to the acoustic signal features, temperature signal features, and voltage signal features in the joint feature vector according to a preset weighting strategy. Based on the weights corresponding to the acoustic signal features, temperature signal features, and voltage signal features in the joint feature vector, a weighted summation is performed to obtain a fused feature vector. The fused feature vector is then subjected to dimensionality reduction to obtain a multimodal fused feature.

[0148] Procedures love you, based on Min-Max ( Normalize the various signal characteristics and combine the signal characteristics corresponding to different sensors, i.e., the acoustic signal characteristics [F] 超声 Temperature signal characteristics F 温度 and voltage signal characteristics F 电压 Concatenate the columns to obtain the joint feature vector X. concat =[F 超声 F 温度 F 电压 ].

[0149] Based on the weighted fusion strategy, the acoustic signal features, temperature signal features, and voltage signal features in the joint feature vector are dynamically assigned weights, and the fused feature X is obtained by weighted summation with reference to the following formula (12). fused .

[0150] (12)

[0151] Among them, X i For the i-th feature data in the joint feature vector, w i is the weight of the i-th feature data.

[0152] Furthermore, the dimensionality of the fused features is reduced based on principal component analysis (PCA) to obtain multimodal fused features. Specifically, the covariance matrix C is calculated according to the following equation (13) to describe the degree of linear correlation between multiple feature data in the fused features.

[0153] (13)

[0154] Among them, X i Let represent the i-th feature data, μ represent the mean vector of all feature data, and T represent the transpose operation of a matrix or vector.

[0155] After eigenvalue decomposition, the first k principal components are selected. These k principal component features are then projected from the high-dimensional space to the low-dimensional space, i.e., X. PCA =X·V k This yields low-dimensional multimodal fusion features.

[0156] Furthermore, the multimodal fusion features are input into a pre-built risk detection model to obtain the risk detection results output by the risk detection model. The risk detection model can be a rule model built based on the level conditions corresponding to each risk level. In another possible implementation, the risk detection model can also be a classification model trained based on labeled data.

[0157] Specifically, the training process of the risk detection model may include: acquiring feature combination samples labeled with risk level tags, including: sound wave signal features, temperature signal features, and voltage signal features; inputting the feature combination samples into the initial model, which is a rule model constructed based on the level conditions corresponding to each risk level, to obtain the risk level prediction result corresponding to the feature combination sample; calculating the target loss value based on the risk level prediction result corresponding to the feature combination sample and the risk level tag; updating the model parameters of the initial model according to the total loss value to obtain a risk detection model that meets the training conditions, wherein the total loss value includes at least the target loss value.

[0158] In this embodiment, supervised learning is used to train a mapping model from multimodal signal feature groups to risk levels using labeled risk level data. This model is constructed based on risk level judgment rules, and during training, the model parameters are automatically adjusted according to the loss value, thereby improving the accuracy of the risk detection model.

[0159] Specifically, firstly, a batch (e.g., no less than 1000) of feature samples already labeled with risk levels needs to be collected. Each feature sample group contains three types of features: acoustic signal features, temperature signal features, and voltage signal features. Simultaneously, each sample has a risk level label based on risk level judgment rules, such as labeling each sample with "y=0", "y=1", "y=2", etc., according to the level conditions corresponding to each risk level described in step S140 above.

[0160] Furthermore, an initial model is constructed based on the risk level conditions used when labeling. Optionally, in this embodiment, the initial model is constructed using an LSTM-SVM hybrid architecture, wherein the LSTM layer retains the ability to extract temporal features (for dynamic changes in acoustic signals), and its output dimension is adjusted to a feature vector (e.g., 64-dimensional) that matches the subsequent SVM input. The binary SVM is optimized into a multi-class SVM using a one-to-one strategy, such as constructing a C4 model for four classification labels with equal risk levels. 2 =Six binary SVMs are used to distinguish between “y0-y1”, “y0-y2”, “y0-y3”, “y1-y2”, “y1-y3”, and “y2-y3” respectively. The risk level prediction result is finally determined by voting.

[0161] Based on this, all feature combination samples are input into the initial model. The model processes these samples according to the current grading rules and outputs a risk level prediction result for each feature combination sample. For example, LSTM is used to extract dynamic features from acoustic signals (time series data), which are then concatenated with static features of temperature / voltage. The concatenated features are then normalized and dimensionality reduced. The processed features are then input into a multi-class SVM to obtain the output risk level prediction result.

[0162] Referring to the following formula (14), the risk level prediction result output by the risk detection model is compared with the risk level label, and the target loss value L is calculated. The model parameters of the initial model, such as the penalty coefficient of SVM and kernel function parameters, are optimized based on the target loss value until the initial model meets the training conditions, such as the loss value converges or the training times reach the upper limit. The initial model that meets the training conditions is then used as the risk detection model and put into application.

[0163] (14)

[0164] Where N is the total number of samples of the feature combination, y ic Let p be the risk level label for the i-th feature combination sample. It is 1 if it belongs to level c, and 0 otherwise; ic The probability that sample i belongs to level c is predicted for the risk detection model, which is the risk level prediction result.

[0165] In one possible implementation, the fire risk detection method for the energy storage system further includes: when the risk level in the risk detection results corresponding to a preset number of consecutive detection cycles reaches a preset level, locating the fault location in the battery module, generating a fire extinguishing command based on the fault location, and sending it to the fire extinguishing device in the energy storage system.

[0166] It is understandable that a single sensor may generate false alarms due to environmental interference or its own malfunction. Similarly, the risk detection results of a single detection cycle may also be subject to chance. Therefore, to improve the accuracy of fire risk detection, the fire risk of the energy storage system is determined by comprehensively considering the detection results of multiple consecutive detection cycles (e.g., 3 cycles). For example, if the risk level is Level 1 for 3 consecutive detection cycles, then the energy storage system is determined to have a Level 1 fire risk.

[0167] If a fire risk is identified in the energy storage system, targeted fire extinguishing or prevention measures must be implemented to prevent the fire from occurring or spreading. Specifically, refer to the fire prevention measures corresponding to each fire level in step S140 above, such as early warning to maintenance personnel, power outage, and fire extinguishing. Similarly, if the risk level in the risk detection results corresponding to a preset number of consecutive detection cycles reaches a preset level, a fire extinguishing command is generated based on the pre-set fire extinguishing measures corresponding to that preset level, and the fire extinguishing command is sent to the fire extinguishing device to extinguish the fire.

[0168] Optionally, the energy storage system also includes: a fire extinguishing device, used to receive fire extinguishing instructions sent by the signal processing module, and respond to the fire extinguishing instructions to extinguish the fire at the fault location in the battery module. The fire extinguishing instructions are generated by the signal processing module based on the fault location in the battery module when the risk level in the risk detection results corresponding to a preset number of consecutive detection cycles reaches a preset level.

[0169] Among them, fire extinguishing devices can be used to spray liquids, powders, etc. for fire extinguishing.

[0170] In one possible implementation, the fire extinguishing device may include: a pressure relief window 13 and a sprinkler pipe rack 15; wherein, the pressure relief window is located on the inner left wall of the energy storage cabinet of the energy storage system, and a smoke exhaust fan 14 is installed inside the pressure relief window; the sprinkler pipe rack is installed inside the energy storage cabinet on the outside of the battery rack, and the sprinkler pipe rack 15 is externally connected to an aerosol fire extinguishing agent storage system, and the battery rack is used to place battery modules.

[0171] The energy storage cabinet 1 has a power supply box 11 and a main control cabinet 12 fixedly installed inside. A pressure relief window 13 is provided on the inner left wall of the energy storage cabinet 1, and a smoke exhaust fan 14 is movably installed inside the pressure relief window 13. Inside the energy storage cabinet 1, a sprinkler pipe rack 15 is fixedly installed on the outside of the battery rack 3. An aerosol fire extinguishing agent storage system is connected to the sprinkler pipe rack 15.

[0172] In one possible implementation, refer to Figure 1 The inner rear wall of the energy storage cabinet can also be fixedly installed with a logic expansion module 16. The logic expansion module is electrically connected to the main control cabinet 12, and the main control cabinet 12 transmits fire extinguishing instructions to the fire extinguishing device through the logic expansion module 16. If a sudden increase of 30% in sound wave amplitude is detected, accompanied by a high-frequency component (greater than 35kHz) and lasts for 10 seconds, the main control cabinet 12 starts the smoke exhaust fan 14 through the logic expansion module 16 to force air cooling and upload an alarm signal.

[0173] It is understandable that extinguishing a fire requires locating the source of the fire, and putting out the fire is much more effective. However, if the spraying is not precise and covers a wide area, the fire source may be missed, resulting in poor fire extinguishing results. Therefore, in this embodiment, after determining the risk level, the location of the abnormal signal is located based on the speed at which the abnormal signal reaches each sensor, and the fire is extinguished specifically at that location.

[0174] In one possible implementation, locating the fault location in the battery module includes: acquiring sensor signals collected by multiple sensors in the energy storage system during the detection period; performing cross-correlation calculation on every two sensor signals to determine the cross-correlation peak position, and determining the time difference between the time points corresponding to the cross-correlation peak positions of every two sensor signals; and determining the fault location based on the time difference corresponding to every two sensor signals and the position information of every two sensors.

[0175] First, perform cross-correlation calculations on the sensor signals collected by each sensor (ultrasonic sensor, temperature sensor, voltage sensor, etc.), expressed as: , where x i (t) represents the signal collected by the i-th sensor at time t, x j (t-τ) represents the signal acquired by the j-th sensor at time t-τ. Based on the cross-correlation calculation results between the sensors, the peak position of the cross-correlation is found, and the time difference between the peak positions of the cross-correlation between the two sensors is calculated. This measures the time difference between the propagation of the same anomalous signal to the two sensors. Optionally, the visual resolution can be improved to 1 ns using interpolation, and adaptive filtering can be used to eliminate multipath interference.

[0176] Furthermore, the geometric equations of the sensor are constructed, for example, the following equation (15).

[0177] (15)

[0178] Where (x,y) represents the two-dimensional coordinates of the fault location to be solved, (x0,y0), (x1,y1), and (x2,y2) represent the two-dimensional coordinates of the three sensors, and (x0,y0) is the coordinate of the reference sensor. This represents the ratio of the distance from the fault location to the first sensor to the distance to the reference sensor 0. This represents the ratio of the distance from the fault location to the second sensor to the distance to the reference sensor 0. c represents the effective propagation speed of the abnormal signal (such as ultrasonic signals, sound / pressure wave signals generated by the fault, etc.) in the relevant propagation medium of the energy storage system. This represents the time difference between the propagation of the same abnormal signal to the sensors at positions (x1, y1) and (x0, y0). This represents the time difference between the propagation of the same abnormal signal to the sensors at positions (x2, y2) and (x0, y0). This time difference can be determined based on the above. Calculated, or based on (t) i t j The values ​​represent the times when sensor i and sensor j received the abnormal signal, respectively.

[0179] Equation (15) with only 3 sensors is merely an example. In practice, a geometric equation set can be constructed based on the number of sensors in the energy storage system, using Equation (15) as an example. The geometric equation set is then solved using nonlinear optimization to obtain the two-dimensional coordinates of the fault location. Furthermore, the calculated fault location is used to perform targeted fire suppression.

[0180] It should be noted that the two-dimensional coordinates of the fault location calculated in this embodiment are based on a reasonable simplification scheme that balances the physical structural characteristics of the energy storage system and the positioning requirements. That is, by combining the logic of pre-locking in the vertical direction (three-dimensional) with precise calculation in the plane (two-dimensional), the complexity brought by three-dimensional positioning technology can be avoided, while meeting the accuracy requirements of fault location.

[0181] Understandably, although energy storage systems are three-dimensional devices (as shown in the attached diagram), Figure 1 The system consists of layered battery racks and multi-layered battery modules. However, faults (such as precursors to battery thermal runaway or localized damage) tend to be concentrated in the plane within each layer and can be preset in the vertical direction. Therefore, two-dimensional coordinate calculations are used instead of three-dimensional coordinate calculations. Specifically, battery modules in an energy storage system are typically arranged in a hierarchical manner on a battery rack, and sensors (ultrasonic sensors, temperature sensors) on the same layer are only responsible for collecting signals from that layer (as shown in the attached diagram). Figure 1 The sensor array, which is distributed in the gaps between battery modules or on the surface of the casing, is deployed in layers. Therefore, the vertical height (z coordinate) of the fault location can be directly locked through the sensor's belonging layer. That is, the vertical direction (Z axis) of the fault location in the three-dimensional coordinate system is known and fixed.

[0182] Furthermore, the logic for locating the fault location based on the obtained two-dimensional coordinates can be described below: First, determine the layer where the fault occurs based on the deployment hierarchy of the sensor array and the source of the abnormal signal. For example, in an energy storage system with three layers, each layer deploys three ultrasonic sensors (numbered 1_1, 1_2, 1_3 for the first layer; 2_1, 2_2, 2_3 for the second layer; and 3_1, 3_2, 3_3 for the third layer). If only sensors 2_1, 2_2, and 2_3 in the second layer collect abnormal signals (other layer sensors do not detect abnormalities), then the z-coordinate of the fault is directly determined to be the preset height of the second layer (e.g., 1.0m, fixed by the battery rack design parameters). During this process, if the sensor in the same layer is closer to the fault point, and the intensity and signal-to-noise ratio of the abnormal signal are much higher than those of sensors in other layers, the fault location layer can be quickly filtered out using a signal strength threshold.

[0183] Furthermore, three sensors are selected from the layer where the fault location is located (e.g., 2_1(x0,y0), 2_2(x1,y1), and 2_3(x2,y2) in the second layer), and one of them is used as the reference sensor (e.g., 2_1(x0,y0)). A system of equations is constructed as shown in equation (16). Based on the distance ratio R between the two sensors determined above... 10 =d1 / d0、R 20 =d2 / d0, solve this system of equations to obtain the specific location of the fault point on the second plane.

[0184] (16)

[0185] Where d0 is the distance from the fault point to sensor 2_1, d1 is the distance from the fault point to sensor 2_2, and d3 is the distance from the fault point to sensor 2_3. Δt is the time difference between the arrival of the abnormal signal at each sensor, calculated through cross-correlation; that is, the abnormal signal arrives at sensor 2_2 Δt1 later than at sensor 2_1, and arrives at sensor 2_3 Δt2 later than at sensor 2_1. v represents the speed of sound propagation in the battery module, such as v = 340 m / s.

[0186] Optionally, several modules can be set up to form a positioning subnet to reduce computational complexity, assign higher weights to sensors with high signal-to-noise ratios, and combine the position data from the first 10 seconds for Kalman filtering smoothing to optimize positioning.

[0187] Referring to the above examples of locating the fault location of the battery module and using the fault location to pinpoint the fire extinguishing point, we can also apply this principle to locate the fault location of the sprinkler pipes used for fire extinguishing, so as to conduct timely safety inspections of the fire extinguishing equipment, remind maintenance personnel to carry out timely repairs, and avoid fire extinguishing failure due to fire extinguishing equipment failure.

[0188] In one possible implementation, ultrasonic transceiver probes are symmetrically mounted along the outer wall of the battery fluid pipe of the lithium iron phosphate battery body 202 at a preset interval (e.g., 50 mm), covering the diameter range of the battery fluid pipe. Furthermore, a set of probes is arranged along the fluid direction within the pipe at a preset interval (e.g., 1 m) to detect the distribution of cavitation (e.g., air bubbles, cavities, or voids) throughout the pipe. Optionally, a transmission method is used, with the transmitting probe T and receiving probe R arranged in pairs, a sampling rate of 1 MHz, capturing the sound wave time-of-flight (ToF) and amplitude attenuation.

[0189] Install a preset number (e.g., 1) of piezoresistive pressure sensors at preset intervals (e.g., 0.5m) upstream, midstream, and downstream of the battery fluid pipeline, aligning them with the ultrasonic probe positions. Set the sampling rate to 10kHz to capture the high-frequency components of pressure pulsations, and eliminate low-frequency interference through a 0.1Hz~1kHz bandpass filter.

[0190] In this example, the PXIe-6612 timing module is used to synchronize the sampling clocks of the ultrasonic sensor and the pressure sensor with an accuracy of 1ms, and to embed a GPS synchronization timestamp into the data packet.

[0191] Furthermore, the sound wave signal acquired by the ultrasonic sensor is processed. First, the sound velocity is corrected according to the following formula (17).

[0192] (17)

[0193] Where K is the fluid bulk modulus, ρ is the density, and α is the initial estimate of the cavitation rate.

[0194] Referring to the following formula (18), the cavitation rate α of the pipeline is calculated based on the sound attenuation method.

[0195] (18)

[0196] Where A0 is the received amplitude without cavitation, A is the measured amplitude, β is the attenuation coefficient, and L is the length of the propagation path, i.e. the straight-line propagation distance of the ultrasonic wave in the fluid inside the pipe.

[0197] It is understandable that the presence of cavitation bubbles (gas cavities) in the fluid (such as battery electrolyte) within battery fluid pipes significantly alters the propagation speed of sound waves. The calculation of the cavitation rate α based on the sound attenuation method depends on the degree of sound wave attenuation within the pipe. However, the degree of attenuation is not only related to the cavitation rate but also directly related to the medium properties (i.e., sound velocity) along the sound wave propagation path. Inaccurate sound velocity will lead to deviations in the calculation of sound wave propagation time and the attenuation coefficient β, thus distorting the calculated α result. Therefore, the cavitation rate determines the sound velocity, and the sound velocity, in turn, affects the calculation of the cavitation rate.

[0198] In the actual detection process, it can be executed in the order of "initial sound velocity correction → calculation of cavitation rate → iterative optimization and speed increase". Specifically, the first step is to use the initial estimated value of cavitation rate α0 (such as the common cavitation rate preset according to the type of pipeline fluid) to substitute into formula (17) to correct the sound velocity c and obtain a sound velocity c0 close to the actual sound velocity, avoiding the deviation caused by the calculation of pure fluid sound velocity. The second step is to substitute the corrected sound velocity c0 into the sound attenuation calculation model, and then combine the measured amplitude A, the amplitude without cavitation A0, the path length L, and calculate a more accurate cavitation rate α1 through formula (18). If it is necessary to further improve the accuracy, the cavitation rate α1 can be used as the new initial estimated value of cavitation rate α0, and return to the first step. The cavitation rate calculated again is the iterative optimization relationship between the approximate initial value and the accurate calculated value. The former is the calculation starting point, and the latter is the optimization result.

[0199] During the correction of the sound velocity, the pressure signal acquired by the pressure sensor is analyzed simultaneously. First, pressure pulsation characteristics are extracted. For example, frequency domain analysis of the pressure signal is performed: such as FFT (4096 points) to extract characteristic frequencies; cavitation collapse characteristic peaks (usually located at 1-5kHz) are extracted; and pipe resonant frequencies are extracted to detect abnormal frequency drift. Time domain analysis of the pressure signal is also performed: pressure fluctuation amplitude (P) is extracted. pp =P max -P min ), pressure gradient (dP / dx).

[0200] Furthermore, based on the pre-set leak location algorithm, the location of electrolyte leak is determined by observing the pressure wave generated by cavitation reaching each sensor. For details, please refer to the following formula (19).

[0201] (19)

[0202] By eliminating pressure or cavitation fluctuations caused by normal operations such as pump start-up and shutdown, and valve operation, the location of pipeline leaks can be accurately located, and maintenance personnel can be alerted to carry out timely repairs to reduce safety hazards.

[0203] In summary, the fire risk detection method and energy storage system of this energy storage system capture microscopic changes such as electrolyte boiling and electrode deformation inside the battery module using ultrasonic waves, providing early warning 5-10 minutes earlier than traditional detection methods. Furthermore, the system integrates temperature and voltage data with ultrasonic features through multimodal data fusion processing, effectively improving detection accuracy and reducing the false alarm rate to <0.1%. It also utilizes time-difference positioning technology of sensor arrays to adaptively and accurately locate faulty battery modules.

Claims

1. A method for detecting fire risk in an energy storage system, characterized in that, The energy storage system includes at least one battery module, and the fire risk detection method for the energy storage system includes: Within a detection cycle, acquire the acoustic wave signal, temperature signal, and voltage signal corresponding to the battery module. The acoustic wave signal is the echo signal formed after the internal part of the battery module reflects the ultrasonic wave signal emitted from the outside. Extract the acoustic signal features of the abnormal frequency signals in the acoustic signal, and extract the signal features corresponding to the temperature signal and the voltage signal respectively; If the combination of sound wave signal features, temperature signal features and voltage signal features satisfies any one of the preset level conditions corresponding to at least two risk levels, the risk level corresponding to the level condition satisfied by the feature combination shall be taken as the risk detection result. Each level condition corresponding to the risk level includes a preset threshold corresponding to each signal feature type. The risk level is proportional to the number of signal feature types that exceed the threshold. The step of extracting acoustic signal features of abnormal frequency signals in the acoustic signal includes: Extracting a first anomalous frequency signal from the anomalous frequency band of the acoustic signal, the first anomalous frequency signal incorporating the low-frequency trend and high-frequency anomalous characteristics of the anomalous frequency band, including: The acoustic signal is decomposed using a wavelet basis function that matches the transient characteristics of the impact signal of the battery module, resulting in multiple wavelet decomposition levels. The detail coefficients of the wavelet decomposition level containing the abnormal frequency band and the approximation coefficients of the wavelet decomposition level containing the lowest frequency band among the multiple wavelet decomposition levels are extracted. The detail coefficients reflect the high-frequency detail information of the acoustic signal in that frequency band, and the approximation coefficients represent the low-frequency components of the signal, containing the main trend and energy information of the acoustic signal. The detail coefficients are a digital sequence, where each number in the digital sequence represents the signal strength of its corresponding signal in the abnormal frequency band. Based on the digital sequence, determine the noise reduction threshold for the abnormal frequency band; The digital sequence is denoised according to the denoising threshold to obtain the denoised detail coefficients; The signal is reconstructed using the denoised detail coefficients and the approximation coefficients to obtain the first abnormal frequency signal; The first abnormal frequency signal is subjected to bidirectional filtering to obtain the second abnormal frequency signal; Gain adjustment is performed on the signal at the center frequency of the abnormal frequency band in the second abnormal frequency signal to obtain the second abnormal frequency signal after gain compensation. The inherent resonance features in the second abnormal frequency signal after gain compensation are removed to obtain the abnormal frequency signal. The inherent resonance features are the vibration features of the acoustic wave signal generated by the battery module when there is no fault. Time-frequency analysis is performed on the abnormal frequency signal to obtain the acoustic signal characteristics of the abnormal frequency signal.

2. The fire risk detection method for an energy storage system according to claim 1, characterized in that, The step of extracting the signal features corresponding to the temperature signal and the voltage signal respectively includes: The temperature signal and the voltage signal are interpolated respectively to obtain interpolated temperature signal and interpolated voltage signal, and the sampling time points in the interpolated temperature signal and the interpolated voltage signal are aligned with the time points of the sound wave signal; The signal features of the interpolated temperature signal and the interpolated voltage signal are extracted to obtain temperature feature signals and voltage feature signals.

3. The fire risk detection method for an energy storage system according to any one of claims 1-2, characterized in that, Also includes: If the risk level in the risk detection results reaches a preset level after a preset number of consecutive detection cycles, the fault location in the battery module is located, and a fire extinguishing command is generated based on the fault location and sent to the fire extinguishing device in the energy storage system.

4. The fire risk detection method for an energy storage system according to claim 3, characterized in that, Locating the fault location in the battery module includes: Acquire sensor signals collected by multiple sensors in the energy storage system during the detection period; Perform cross-correlation calculation on every two sensor signals to determine the cross-correlation peak position, and determine the time difference between the time points corresponding to the cross-correlation peak positions in every two sensor signals; The fault location is determined based on the time difference between each pair of sensor signals and the position information of each pair of sensors.

5. An energy storage system, characterized in that, include: At least one battery module, and a sensor array distributed in the gaps or on the surface of the casing of the battery module, and a signal processing module; wherein the sensor array is used to collect echo signals formed by reflecting ultrasonic signals emitted from the outside by the inside of the battery module, as well as temperature signals and voltage signals of the battery module; The signal processing module is used to implement the fire risk detection method for the energy storage system according to any one of claims 1-4.

6. The energy storage system according to claim 5, characterized in that, Also includes: A fire extinguishing device is used to receive a fire extinguishing command sent by the signal processing module and respond to the fire extinguishing command to extinguish the fire at the fault location in the battery module. The fire extinguishing command is generated by the signal processing module based on the located fault location in the battery module when the risk level in the risk detection results corresponding to a preset number of consecutive detection cycles reaches a preset level.

7. The energy storage system according to claim 6, characterized in that, The fire extinguishing device includes: a pressure relief window and a sprinkler pipe frame; The pressure relief window is located on the inner left wall of the energy storage cabinet of the energy storage system, and a smoke exhaust fan is installed inside the pressure relief window; The spray pipe rack is installed inside the energy storage cabinet and located outside the battery rack. The spray pipe rack is connected to an aerosol fire extinguishing agent storage system. The battery rack is used to house the battery modules.

Images