A method and system for monitoring thermal runaway protection of an energy storage power station

Abstract

The present application relates to the technical field of intelligent monitoring and control of energy storage power stations, and particularly relates to a monitoring method and system for thermal runaway protection of energy storage power stations; the method comprises synchronous acquisition and processing of thermal runaway characteristic signals of liquid-cooled battery units through a photoacoustic module and an infrared imager to obtain CO concentration time series data and temperature cloud map spatial data; through filtering, feature extraction and multi-modal fusion processing of an AI intelligent controller, a sparse feature vector of fused gas concentration and temperature features is obtained; the sparse feature vector of fused gas concentration and temperature features is processed through compressed sensing and a risk decision model to form a thermal runaway early warning signal and positioning result of the liquid-cooled battery unit. The system comprises a liquid-cooled box body, an AI intelligent controller, a coolant reservoir, a condenser, a circulating pump, an electric valve, a pressure relief valve, an infrared imager, a photoacoustic module and a liquid-cooled battery unit. The present application realizes rapid extraction and reliable identification of early thermal runaway features.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring and control technology for energy storage power stations, and in particular to a monitoring method and system for thermal runaway protection of energy storage power stations. Background Technology

[0002] As the scale of energy storage power stations continues to expand and the number of batteries surges, the risk of thermal runaway increases accordingly. Traditional detection technologies are insufficient to meet the demand for early and accurate warnings. Currently, the industry's requirements for the accuracy and timeliness of early thermal runaway detection are increasingly stringent, and the shortcomings of traditional technologies have become a bottleneck restricting the safe development of energy storage power stations. At the gas detection level, mainstream technologies rely on electrochemical sensors or catalytic combustion sensors. The former is less effective at detecting key characteristic gases in the early stages of thermal runaway, such as… Insufficient sensitivity to CO The detection limits are mostly 5-10 ppm, and for CO it is 10-20 ppm, which cannot capture trace amounts of 1 ppm in the initial stage of thermal runaway. The false alarm rate is as high as 15%-20% when the gas is released and affected by temperature, humidity, and cross-gas interference; the latter is only sensitive to the "lower explosive limit concentration" of combustible gases, such as... The required concentration is 40,000 ppm, completely missing the early warning window. Furthermore, traditional sensors have response times of 1-3 seconds, while thermal runaway can progress from gas release to open flame in just tens of seconds, resulting in significant lag. Temperature monitoring technology also has shortcomings. Contact sensors like thermocouples / resistance temperature detectors (RTDs) are limited by installation density, covering only 30% of the batteries, making it difficult to capture localized hotspots in densely stacked battery packs. Additionally, the 0.5-1 second delay in metal conduction fails to reflect internal short-circuit heating within the cells. While non-contact infrared thermal imagers can achieve wide-area monitoring, traditional equipment has an accuracy of only ±0.5℃-±1℃, making it difficult to identify minute temperature differences of 0.5-1℃ in the early stages of thermal runaway. Moreover, the 320×240 pixel resolution leads to a positioning error exceeding 10cm at 5 meters, making it impossible to pinpoint a single faulty battery. Furthermore, signal interference from insulation layers and dust can cause temperature measurement deviations exceeding ±2℃. Deficiencies in data processing and response mechanisms further amplify the risks. In the existing system, gas and temperature data are collected in isolation and lack correlation verification. A sudden rise in ambient temperature may falsely trigger a temperature warning, while early gas release may be missed because the temperature does not change significantly.

[0003] Prior art one, Chinese patent, application number: 202410029397.0 relates to the field of fire protection technology in electrical equipment, specifically relates to a kind of energy storage power station battery thermal runaway protection device. Including shell, still including foam stock solution tank, water tank, gas cylinder, foam generator, heating device and ventilation device, foam stock solution tank is connected with stock solution pump, first electric proportional valve, first electric proportional valve is connected with stock solution delivery pipe, stock solution delivery pipe is connected with stock solution pump;Water tank is connected with second electric proportional valve, second electric proportional valve is connected with water delivery pipe;Gas cylinder is connected with third electric proportional valve, third electric proportional valve is connected with gas delivery pipe,

[0004] Gas delivery pipe is connected with pumping bin, foam generator is connected with pumping bin, foam generator is also connected with gas delivery pipe, foam generator is connected with foam delivery pipe, foam delivery pipe is communicated with liquid outlet, liquid outlet is also communicated with fire water pipeline;Heating device is fixedly arranged on the tank wall of water tank;Ventilation device is fixed to the shell. Although it can be used for energy storage power station battery thermal runaway to carry out fire extinguishing;However, the "latent stage" of thermal runaway cannot be covered, and the temperature and gas sensor needs an abnormal signal to reach the threshold value to alarm.

[0005] Prior art two, Chinese patent, application number: 202510494412.3 discloses a kind of container type energy storage power station thermal runaway precise positioning fire-fighting system and method, which is composed of energy storage power station, array distribution battery pack pack and fire-fighting system. Each battery pack pack is equipped with a gas safety valve and a fire extinguishing device, the energy storage power station is arranged to cover the hydrogen sensor array of the whole area, the gas safety valve guides the hydrogen generated in the initial stage of thermal runaway, and the sensor array monitors the global concentration in real time. Industrial computer receives sensor data, combines pre-stored battery pack space distribution information, constructs gas concentration distribution map, and displays thermal runaway area on display screen in real time. Positioning program is based on the analysis of hydrogen concentration field based on gas diffusion characteristics, and accurately identifies the position of fault battery pack. After that, the industrial computer triggers the fire extinguishing device in the corresponding pack to implement point fire fighting, forming an early fire extinguishing closed loop. Although, through multi-sensor data fusion and gas distribution mapping technology, the source of thermal runaway is quickly located and accurately suppressed, and the safety protection level of energy storage power station is effectively improved;However, the batteries in the power station are dense, the sensors are difficult to cover due to installation cost and space limitation, and the internal module and the gap between the batteries are prone to monitoring dead angles.

[0006] The prior art three, Chinese patent, application number: 202411870709.7 discloses a kind of based on flame-retardant foaming realization thermal runaway grading early warning energy storage power station safety management system, by controlling refrigerant spraying and initiating foaming device opening, high-temperature resistant flame-retardant foaming agent is released quickly in large quantities, battery module is rapidly wrapped and isolated oxygen, by refrigerant spraying and foaming material combination, avoid the further spread risk of battery module thermal runaway, realize the safety management of energy storage power station.Through the control of valve, the switching of different operating modes can be realized, and the safety management of thermal runaway grading early warning of energy storage power station is realized.Although, by setting two-stage grading early warning stage to energy storage power station battery, the battery cooling operating mode of two-stage early warning stage and the blocking mode of battery thermal runaway under one-stage early warning mode are realized, the operation temperature of battery can be reasonably regulated and controlled, the operation efficiency of battery is improved, and the risk propagation of battery thermal runaway is effectively blocked;However, the high-temperature and electromagnetic environment of power station are prone to sensor false alarm, and the cross of different battery characteristic gases also affects the accuracy of gas monitoring, so it is difficult to accurately early warn.

[0007] The prior art one, the prior art two and the prior art three have the problems of monitoring lag, space blind area and poor anti-interference.Therefore, the present application provides a kind of monitoring method and system for thermal runaway protection of energy storage power station. SUMMARY

[0008] To achieve the above purpose, the present application adopts the following technical solutions:

[0009] In one aspect of the present application, a kind of monitoring method for thermal runaway protection of energy storage power station is provided, comprising the following steps:

[0010] The thermal runaway characteristic signal of liquid-cooled battery unit is synchronously collected and processed by photoacoustic module and infrared imager, to obtain And CO concentration time series data and temperature cloud map space data;

[0011] And CO concentration time series data and temperature cloud map space data are filtered, feature extraction and multi-modal fusion processing by AI intelligent controller, to obtain sparse feature vector of fusion gas concentration and temperature feature;

[0012] Sparse feature vector of fusion gas concentration and temperature feature is processed by compressed sensing and risk decision model, to form thermal runaway early warning signal and positioning result of liquid-cooled battery unit.

[0013] In an alternative embodiment, the process of obtaining And CO concentration time series data and temperature cloud map space data comprises the following steps:

[0014] Energy storage power station environment gas is sampled by micro-pump of photoacoustic module and processed by laser specific wavelength irradiation, to obtain and the periodic pressure fluctuation signal generated after the CO molecules absorb energy; the periodic pressure fluctuation signal is captured by a high-sensitivity microphone and converted into an electrical signal for processing, forming an original photoacoustic voltage signal; the original photoacoustic voltage signal is analyzed and calculated by the signal intensity and frequency of the data processing unit to obtain the time-stamped concentration sequence and the CO concentration sequence;

[0015] The liquid-cooled battery cell surface infrared radiation is inducted by an infrared thermal imager containing an infrared lens and an infrared focal plane detector, to obtain an original electrical signal array corresponding to the temperature distribution; the original electrical signal array is pre-processed by non-uniformity correction and noise reduction filtering to form standardized radiation intensity data; the standardized radiation intensity data is converted and processed by a temperature calibration algorithm based on the blackbody radiation law, to finally generate 640x512 pixel temperature cloud map spatial data;

[0016] The photoacoustic gas detection channel of the photoacoustic module and the infrared thermal imaging channel of the infrared thermal imager are coordinated and processed by a hardware-level time synchronization mechanism to obtain aligned time-spatial synchronization acquisition data; the synchronization acquisition data is processed by time sequence marking and spatial registration to form and CO concentration time series data and temperature cloud map spatial data.

[0017] In an optional implementation, the process of obtaining a sparse feature vector that fuses gas concentration and temperature characteristics includes the following steps:

[0018] The CO and CO concentration time series data are processed by a time series noise suppression algorithm based on Kalman filtering to obtain smoothed concentration data that eliminates pulse fluctuations; the smoothed concentration data are processed by multi-dimensional feature extraction to obtain a time series feature set containing concentration peak value, concentration change rate and concentration fluctuation variance within 1 minute; the time series feature set is processed by minimum-maximum standardization to form a gas concentration feature vector standardized to the [0, 1] interval;

[0019] The temperature cloud map spatial data are processed by non-uniformity correction and pixel-level compensation to obtain a corrected temperature field that eliminates lens response differences and edge temperature deviations; the corrected temperature field is processed by image pyramid downsampling and spatial feature extraction to form a spatial feature set containing temperature mean value, hot spot area proportion and highest temperature position; the spatial feature set is processed by regional weight allocation to obtain a temperature spatial feature vector that strengthens the temperature > 40℃ area feature;

[0020] The gas concentration feature vector and the temperature space feature vector are subjected to double-branch feature extraction processing of a bidirectional long short-term memory network and a convolutional neural network, to obtain 256-dimensional time sequence features and 512-dimensional space features; the time sequence features and the space features are subjected to cross-modal attention mechanism fusion processing, to form a 768-dimensional joint feature vector; the 768-dimensional joint feature vector is subjected to feature selection and dimension compression processing, and finally a 512-dimensional sparse feature vector of the fused gas concentration and temperature features is obtained.

[0021] In an alternative embodiment, the process of forming a thermal runaway early warning signal and a positioning result of a liquid-cooled battery cell comprises the following steps:

[0022] The sparse feature vector of the fused gas concentration and temperature features is subjected to linear projection processing of a Gaussian random observation matrix, to obtain a compressed data vector with reduced dimension; the compressed data vector is subjected to a reconstruction algorithm based on L1 norm constraint processing, to restore a 512-dimensional reconstructed feature vector highly similar to the original feature vector;

[0023] The 512-dimensional reconstructed feature vector is subjected to multi-index joint risk judgment processing, to obtain a thermal runaway risk probability score by simultaneously evaluating a weighted combination value of the gas concentration anomaly index and the temperature anomaly space distribution coefficient; the thermal runaway risk probability score is subjected to double-threshold early warning mechanism processing, to generate different levels of thermal runaway early warning signals when the score exceeds a preset threshold;

[0024] The space feature component in the reconstructed feature vector is subjected to high-temperature region coordinate inversion processing, to determine the two-dimensional coordinate position of the heat source in the battery cell by decoding the spatial position information and temperature distribution pattern contained in the feature vector; the two-dimensional coordinate position is subjected to unit mapping conversion processing, to finally form a precise positioning result specific to the serial number of a single liquid-cooled battery cell.

[0025] In an alternative embodiment, the process of generating different levels of thermal runaway early warning signals when the score exceeds a preset threshold comprises the following steps:

[0026] The 512-dimensional reconstructed feature vector is subjected to gas concentration anomaly index processing, to obtain a concentration anomaly index quantifying the gas anomaly level by analyzing the numerical deviation degree of the gas concentration related dimensions in the reconstructed feature vector; and is simultaneously subjected to temperature anomaly space distribution coefficient processing, to obtain a space distribution coefficient representing the temperature anomaly degree by analyzing the distribution rule and anomaly pattern of the space feature dimensions;

[0027] The concentration anomaly index and the spatial distribution coefficient are subjected to dynamic weight fusion processing, the weighted proportion is automatically adjusted according to the real-time reliability evaluation results of the two indexes, and a weighted combination value that comprehensively reflects the abnormal degree of gas and temperature is obtained; the weighted combination value is subjected to nonlinear risk mapping function conversion processing, and through the saturation characteristics of the hyperbolic tangent function, the continuous numerical value is mapped into a thermal runaway risk probability score in the range of 0 to 1;

[0028] The thermal runaway risk probability score is subjected to hierarchical threshold comparison processing, the relative size relationship between the score value and the preset early warning threshold and emergency warning threshold is compared, and the corresponding risk level identifier is obtained; the risk level identifier is subjected to early warning signal generation logic processing, and a thermal runaway early warning signal containing different warning levels is formed; when the risk probability score exceeds the two thresholds at the same time, an emergency warning signal of the highest level is generated.

[0029] In an optional implementation, the process of mapping the continuous numerical value into the thermal runaway risk probability score in the range of 0 to 1 includes the following steps:

[0030] The 512-dimensional reconstructed feature vector is subjected to gas concentration dimension deviation degree processing to obtain a concentration anomaly index reflecting the abnormal degree of each gas concentration dimension value; the 512-dimensional reconstructed feature vector is simultaneously subjected to temperature spatial distribution discrete feature analysis processing to obtain a spatial distribution coefficient representing the abnormal region distribution characteristics of temperature;

[0031] The concentration anomaly index and the spatial distribution coefficient are subjected to dynamic weight factor allocation processing, and the differentiated weight proportion is allocated according to the real-time data quality evaluation result, so as to obtain a comprehensive quantitative value of gas and temperature anomaly;

[0032] The comprehensive quantitative value is subjected to input range normalization processing to obtain a normalized value in a standard defined interval; the normalized value is subjected to saturation nonlinear transformation processing with S-shaped curve characteristics to form a probability output value presenting progressive saturation characteristics, and the probability output value is subjected to output range calibration processing to finally obtain a thermal runaway risk probability score limited in the numerical range of 0 to 1.

[0033] In an optional implementation, the process of forming the probability output value presenting the progressive saturation characteristics includes the following steps:

[0034] The normalized value is subjected to S-shaped curve function conversion processing, the input normalized value is converted into a corresponding function output value through the inherent S-shaped nonlinear response characteristics, and an intermediate conversion value with smooth transition characteristics is obtained;

[0035] The intermediate conversion value is processed by asymptotic boundary convergence, uses the mathematical property of S-shaped curve function that the output value asymptotically converges to two fixed limit values when the input value tends to positive or negative infinity, evaluates the saturation degree by calculating the asymptotic distance between the output value and the limit value, and forms the uncalibrated probability value with the asymptotic property of upper and lower boundaries;

[0036] The uncalibrated probability value is processed by linear affine transformation, the linear mapping relationship between the uncalibrated probability value and the standard probability interval is established, the mathematical transformation method combining translation and scaling is used to adjust the uncalibrated probability value from its original range to the standard probability interval of 0 to 1, and finally the probability output value meeting the probability distribution characteristics is obtained.

[0037] In an optional embodiment, the process of adjusting the uncalibrated probability value from its original range to the standard probability interval of 0 to 1 includes the following steps:

[0038] The uncalibrated probability value is processed by original range extreme value determination to obtain the actual minimum and maximum values of the uncalibrated probability value in the current numerical interval; the actual minimum and maximum values are processed by target interval mapping relationship establishment to form linear transformation parameters from the uncalibrated interval to the standard probability interval;

[0039] The linear transformation parameters are processed by translation operation to obtain the translated values with zero as the reference by subtracting the lower limit value of the original interval from the uncalibrated probability value; the translated values are processed by scaling operation to obtain the preliminary scaling results by dividing the original interval width and multiplying the target interval width;

[0040] The preliminary scaling results are processed by target interval reference alignment to finally obtain the standardized probability output values distributed in the closed interval of 0 to 1 by adding the lower limit value of the standard probability interval.

[0041] In an optional embodiment, the process of processing the translated values by scaling operation includes the following steps:

[0042] The translated values are processed by relative position scaling to convert the absolute values into pure proportional values representing their relative positions in the original interval by division operation between the translated values and the total span distance of the original interval, to obtain the unit proportional values eliminating the influence of the original interval range;

[0043] The unit proportional values are processed by target range mapping by multiplication operation between the unit proportional values and the total span distance of the standard probability interval;

[0044] The pure proportional values are converted into actual values meeting the scale requirements of the target interval to obtain the preliminary scaling results matching the scale of the target interval.

[0045] Another aspect of the present application provides a monitoring system for thermal runaway protection of an energy storage power station, which implements the monitoring method for thermal runaway protection of the energy storage power station, comprising a liquid cooling box, an AI intelligent controller, a cooling liquid reservoir, a condenser, a circulating pump, an electric valve, a pressure relief valve, an infrared imager, an optoacoustic module, and a liquid-cooled battery unit.

[0046] The liquid-cooled battery unit is connected to the inlet end of the cooling liquid reservoir through a pipeline, the outlet end of the cooling liquid reservoir is connected to the inlet end of the condenser, the outlet end of the condenser is connected to the input end of the circulating pump, the output end of the circulating pump is connected to one end of the electric valve, and the other end of the electric valve is communicated with the liquid-cooled battery unit. The infrared imager is embedded in the lower right end of the liquid cooling box, the optoacoustic module is embedded in the upper right end of the liquid cooling box, and the pressure relief valve is embedded in the top end of the liquid cooling box. The AI intelligent controller is connected to the pressure relief valve, the infrared imager, and the optoacoustic module through a data line.

[0047] The present application synchronously collects / CO gas concentration time series signals and temperature cloud map space signals through the optoacoustic module and the infrared imager, realizes multi-physical quantity parallel perception of thermal runaway early features, solves the limitations of traditional single-mode detection methods in sensitivity or spatial resolution, adopts a processing method combining time series filtering and spatial feature extraction, aligns the dynamic features of gas concentration changes with the spatial features of temperature distribution, and performs feature-level fusion, eliminating the uncertainty of single-sensor data and improving the completeness of feature representation. Through the compression sensing algorithm, the multi-modal fusion features are sparsely represented, the data dimension is reduced under the premise of retaining key discriminative information, and the risk decision model is combined to recognize the compressed features, realizing early determination and spatial positioning of the thermal runaway state. This technology breaks through the bottleneck of traditional systems in data transmission and real-time processing. BRIEF DESCRIPTION OF DRAWINGS

[0048] The accompanying drawings are included to provide a further understanding of the present application, and constitute a part of the specification, which together with the embodiments of the present application, serve to explain the present application, and do not constitute a limitation of the present application. In the drawings:

[0049] Figure 1 A flowchart of the monitoring method for thermal runaway protection of an energy storage power station is provided in Example 1 of the present application;

[0050] Figure 2 A schematic diagram of the monitoring method for thermal runaway protection of an energy storage power station is provided in Example 1 of the present application;

[0051] Figure 3 A process diagram for obtaining and CO concentration time series data and temperature cloud map space data is provided in Example 2 of the present application;

[0052] Figure 4 Process diagram for obtaining the sparse feature vector of the fusion gas concentration and temperature characteristics provided in Embodiment 5 of the present application;

[0053] Figure 5 Process diagram for forming the thermal runaway pre-warning signal and positioning result of the liquid-cooled battery cell provided in Embodiment 9 of the present application;

[0054] Figure 6 Block diagram of the monitoring system for energy storage power station thermal runaway protection provided in Embodiment 15 of the present application;

[0055] Figure 7 Principle diagram of the photoacoustic module provided in Embodiment 15 of the present application;

[0056] Figure 8 Principle diagram of the infrared imager provided in Embodiment 15 of the present application;

[0057] Figure 9 Block diagram of the electronic device provided by the present application;

[0058] Figure 10 Block diagram of the computer readable storage medium provided by the present application;

[0059] Reference signs: 1, liquid cooling box; 2, AI intelligent controller; 3, coolant reservoir; 4, condenser; 5, circulating pump; 6, electric valve; 7, pressure relief valve; 8, infrared imager; 9, photoacoustic module; 10, liquid-cooled battery cell; 11, central processing unit / microprocessor / main control chip, etc.; 12, storage medium; 13, data bus; 14, input / output bus / external bus / device bus, etc.; 15, display; 16, input / output device; 17, computer readable instructions; 18, non-transitory computer readable storage medium. DETAILED DESCRIPTION

[0060] The technical solutions in the embodiments of the present application will be described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all.

[0061] Hereinafter, the terms "first", "second", etc. are only used for convenience of description, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Therefore, the features defined with "first", "second", etc. can explicitly or implicitly include one or more of the features. In the description of the present application, unless otherwise specified, the meaning of "a plurality of" is two or more.

[0062] In this invention, unless otherwise explicitly specified and limited, the term "connection" should be interpreted broadly. For example, "connection" can be a fixed mechanical connection, a detachable mechanical connection, or an integral part; or, "connection" can be a direct connection or an indirect connection through an intermediate medium. Furthermore, unless otherwise explicitly specified and limited, the term "coupling" should be interpreted broadly. For example, "coupling" can be a direct electrical connection, such as physical contact and electrical conduction between two components; it can also be understood as an electrical connection between different components in a circuit structure through physical lines capable of transmitting electrical signals, such as copper foil or wires on a printed circuit board (PCB), to transmit electrical signals; or, "coupling" can be an indirect electrical connection between two components through an intermediate medium; or, "coupling" can be an electrical connection between two components in a non-contact manner, such as an electrical connection between two components using capacitive coupling to transmit electrical signals.

[0063] In this embodiment of the invention, directional terms such as "up," "down," "left," and "right" may be defined relative to the orientation of the components shown in the accompanying drawings. It should be understood that these directional terms can be relative concepts, used for relative description and clarification, and can change accordingly depending on the orientation of the components in the accompanying drawings.

[0064] Example 1:

[0065] like Figure 1 As shown in the figure, this embodiment of the invention provides a monitoring method for thermal runaway protection in energy storage power stations, comprising the following steps:

[0066] Step S100: The thermal runaway characteristic signal of the liquid-cooled battery cell is simultaneously acquired and processed by the photoacoustic module and the infrared imager to obtain... CO concentration time series data and temperature contour map spatial data;

[0067] Step S200: The CO concentration time series data and temperature cloud map spatial data are filtered, feature extracted and multimodal fused by the AI ​​intelligent controller to obtain a sparse feature vector that integrates gas concentration and temperature features.

[0068] Step S300: The sparse feature vector that integrates gas concentration and temperature characteristics is processed by compressed sensing and risk decision-making model to form a thermal runaway early warning signal and location result for the liquid-cooled battery unit.

[0069] For details on the principles described in the above embodiments, please refer to the appendix. Figure 2 This embodiment uses a photoacoustic module and an infrared imager to simultaneously acquire data. By combining the temporal signal of CO gas concentration with the spatial signal of temperature cloud map, this technology enables parallel sensing of multiple physical quantities for early thermal runaway characteristics, overcoming the limitations of traditional single-modal detection methods in terms of sensitivity or spatial resolution. A processing approach combining temporal filtering and spatial feature extraction is employed to align and fuse the dynamic characteristics of gas concentration changes with the spatial characteristics of temperature distribution at the modal level, eliminating the uncertainty of single-sensor data and improving the completeness of feature representation. A compressed sensing algorithm is used to sparsely represent the multimodal fused features, reducing data dimensionality while retaining key discriminative information. Finally, a risk decision model is used to perform pattern recognition on the compressed features, enabling early determination and spatial localization of thermal runaway states. This technology overcomes the bottlenecks in data transmission and real-time processing of traditional systems.

[0070] In summary, this embodiment constructs a comprehensive monitoring system from microscopic gas characteristics to macroscopic temperature distribution, forming a closed-loop processing architecture of signal acquisition, feature fusion, and compression decision-making; and realizes rapid extraction and reliable identification of early thermal runaway characteristics.

[0071] Example 2:

[0072] like Figure 3 As shown, based on Example 1, the step S100 provided in this embodiment of the invention yields... The process of combining CO concentration time-series data with temperature contour map spatial data includes the following steps:

[0073] Step S101: The ambient gas in the energy storage power station is sampled by a micro-pump of the photoacoustic module and irradiated with a laser at a specific wavelength to obtain a solution containing... The periodic pressure fluctuation signal generated after CO molecules absorb energy is captured by a high-sensitivity microphone and processed by electrical signal conversion to form the original photoacoustic voltage signal. This original photoacoustic voltage signal is then analyzed and calculated by the data processing unit based on signal strength and frequency to obtain a timestamped signal. Concentration sequence (accuracy 1 ppm) and CO concentration sequence (accuracy 5 ppm).

[0074] Step S102: The infrared radiation on the surface of the liquid-cooled battery cell is processed by an infrared thermal imager containing an infrared lens focusing and an infrared focal plane detector to obtain a raw electrical signal array corresponding to the temperature distribution; the raw electrical signal array is preprocessed by non-uniformity correction and noise reduction filtering to form standardized radiation intensity data; the standardized radiation intensity data is converted and processed by a temperature calibration algorithm based on the blackbody radiation law to finally generate 640×512 pixel temperature cloud map spatial data.

[0075] Step S103: The photoacoustic gas detection channel of the photoacoustic module and the infrared thermal imaging channel of the infrared thermal imager are coordinated and processed through a hardware-level time synchronization mechanism to obtain aligned time-space synchronous acquisition data; the synchronous acquisition data is processed through time sequence marking and space registration to form a time reference and CO concentration time sequence data and temperature cloud space data.

[0076] In the above embodiment, the present embodiment realizes the synchronous capture of gas release and temperature field change in the thermal runaway process through the complementary physical principles of gas molecule vibration-rotation energy level detection of the photoacoustic module and thermal radiation detection of the infrared thermal imager; the time synchronization mechanism ensures that the time alignment accuracy of the two kinds of data reaches the millisecond level. The laser wavelength selectively provides molecular fingerprint identification capability of / CO, and the infrared focal plane array realizes two-dimensional space distribution measurement at a temperature resolution of 0.5 ℃; the hardware-level synchronization of the two forms a coupled correlation data set of gas concentration-temperature gradient. The pressure fluctuation signal of the photoacoustic channel is analyzed in time domain and frequency domain, and the infrared channel overcomes the response difference of the detector pixels through non-uniformity correction. The synchronous data interface uses a hardware trigger signal to eliminate the time sequence jitter of software synchronization. The time stamp marking mechanism unifies the sampling interval of the gas concentration time sequence and the thermal image frame rate to a common time reference, and the space registration algorithm establishes the coordinate mapping relationship between the gas sampling points and the temperature field pixels. The sealed gas chamber design of the photoacoustic module avoids environmental airflow interference, the optical filtering of the infrared thermal imager suppresses non-target band radiation, and the synchronous trigger of the dual-channel ensures the causal correlation traceability of abnormal events.

[0077] Embodiment 3

[0078] On the basis of embodiment 1, the process of obtaining the time-stamped concentration sequence and the CO concentration sequence in step S101 provided by the present embodiment of the application comprises the following steps:

[0079] Step S1011: The original photoacoustic voltage signal is subjected to noise suppression processing based on an adaptive threshold to obtain a denoised photoacoustic time domain signal; the denoised photoacoustic time domain signal is subjected to fast Fourier transform and characteristic frequency band extraction processing to form a dual-channel spectral energy distribution containing a characteristic absorption frequency band and a CO characteristic absorption frequency band;

[0080] Step S1012: The dual-channel spectral energy distribution is subjected to a pre-established laser wavelength-molecular absorption characteristic matching program to obtain a separation and quantization result of a molecular characteristic absorption energy value and a CO molecular characteristic absorption energy value; The separation and quantization result of the molecular characteristic absorption energy value and the CO molecular characteristic absorption energy value is subjected to a multi-concentration calibration curve nonlinear mapping processing to form concentration instantaneous measurement value and the CO concentration instantaneous measurement value;

[0081] Step S1013: The concentration instantaneous measurement value and the CO concentration instantaneous measurement value are subjected to time sequence alignment and sliding window filtering processing, and finally, a concentration sequence with a time resolution of 100 ms and a precision of 1 ppm and a CO concentration sequence with a precision of 5 ppm are obtained, each data point being provided with a millisecond-level time stamp. The concentration sequence and the CO concentration sequence each have a millisecond-level time stamp.

[0082] In the above embodiment, the present embodiment realizes accurate separation of the H2S signal and the CO signal in the frequency domain space, solves the industry problem of cross interference of mixed gases by using a laser wavelength-molecular absorption characteristic matching model, ensures the accuracy of conversion from photoacoustic energy to concentration value by using a multi-concentration calibration curve nonlinear mapping method, and finally guarantees the synchronization of the gas concentration data and the temperature cloud map data through a time sequence alignment mechanism.

[0083] Embodiment 4:

[0084] On the basis of embodiment 3, the present embodiment provides a process for obtaining the separation and quantization results of the H2S molecular characteristic absorption energy value and the CO molecular characteristic absorption energy value in step S1012, and the process comprises the following steps:

[0085] Step S10121: The dual-channel spectral energy distribution is subjected to spectral fingerprint region energy decoupling processing, and H2S characteristic band energy integral values and CO characteristic band energy integral values are obtained. The H2S characteristic band energy integral values and the CO characteristic band energy integral values; The H2S characteristic band energy integral values and the CO characteristic band energy integral values are subjected to dynamic trim processing of an absorption cross section weighting factor, and purified absorption signals and purified CO absorption signals are formed, which eliminate cross interference.

[0086] Step S10122: The purified absorption signals and the purified CO absorption signals are subjected to molecular energy level transition probability correction processing, and H2S molecular characteristic absorption energy estimates and CO molecular characteristic absorption energy estimates are obtained. The H2S molecular characteristic absorption energy estimates and the CO molecular characteristic absorption energy estimates;

[0087] Step S10123: The H2S molecular characteristic absorption energy estimates and the CO molecular characteristic absorption energy estimates are subjected to photon-sound energy conversion efficiency compensation processing, and separation and quantization results of H2S molecular characteristic absorption energy accurate values and CO molecular characteristic absorption energy accurate values are obtained. The H2S molecular characteristic absorption energy accurate values and the CO molecular characteristic absorption energy accurate values.

[0088] In the above embodiment, the present embodiment realizes accurate separation of mixed spectrum through a spectral fingerprint area energy decoupling algorithm, solves the problem of differences in absorption cross sections of different gas molecules by using an absorption cross section weighting factor dynamic trim technology, uses a molecular energy level transition probability model to correct the physical accuracy of energy calculation, and finally improves the energy quantization accuracy through photon-sound energy conversion efficiency compensation, thereby providing reliable input data for subsequent concentration inversion.

[0089] Embodiment 5:

[0090] As shown in Figure 4 On the basis of embodiment 1, the process of obtaining the sparsified feature vector of the fusion gas concentration and temperature characteristics in step S200 provided by the present embodiment comprises the following steps:

[0091] Step S201: The time series data of the concentration of CO is subjected to time series noise suppression processing based on the Kalman filtering algorithm to obtain smoothed concentration data eliminating ±0.3 ppm pulse fluctuation; the smoothed concentration data is subjected to multi-dimensional feature extraction processing to obtain a time series feature set containing concentration peak value, concentration change rate and concentration fluctuation variance within 1 minute; and the time series feature set is subjected to minimum-maximum standardization processing to form a gas concentration feature vector standardized to the [0, 1] interval.

[0092] Step S202: The temperature cloud map spatial data is subjected to non-uniformity correction and pixel-level compensation processing to obtain a corrected temperature field eliminating lens response difference and edge temperature deviation; the corrected temperature field is subjected to image pyramid down-sampling and spatial feature extraction processing to form a spatial feature set containing temperature mean value, hot spot area proportion and highest temperature position; and the spatial feature set is subjected to regional weight distribution processing to obtain a temperature spatial feature vector strengthening temperature > 40℃ regional features.

[0093] Step S203: The gas concentration feature vector and the temperature spatial feature vector are subjected to double-branch feature extraction processing of bidirectional long short-term memory network and convolutional neural network to obtain 256-dimensional time series features and 512-dimensional spatial features; the time series features and the spatial features are subjected to cross-modal attention mechanism fusion processing to form a 768-dimensional joint feature vector; and the 768-dimensional joint feature vector is subjected to feature selection and dimension compression processing to finally obtain a 512-dimensional sparsified feature vector of fusion gas concentration and temperature characteristics.

[0094] Wherein, the down-sampling temperature field is processed by a spatial domain integral operation to obtain a temperature average value reflecting the overall thermal state; at the same time, the proportion of the high-temperature region is obtained by counting the proportion of the number of pixels exceeding the set threshold value to the total number of pixels through the high-temperature pixel marking technology; and the position coordinates of the highest temperature point in the temperature field are determined by using the extreme value coordinate positioning method, so as to form a spatial feature set containing the above three types of parameters. The spatial feature set is then subjected to weight distribution processing based on the temperature threshold to obtain a preliminary weight distribution; the weight distribution is subjected to high-temperature region feature enhancement operation, that is, a weighting factor is applied to the region parameter whose temperature exceeds 40℃, so as to improve its contribution in the feature vector, and finally a temperature spatial feature vector with enhanced high-temperature region features is formed.

[0095] In the above embodiment, the embodiment eliminates sensor pulse fluctuations and optical system response differences through Kalman filtering time series noise suppression and non-uniformity correction spatial compensation processing, and improves the measurement reliability of gas concentration data and temperature field data. The time series dimension extracts the peak value / variation rate / variance concentration dynamic characteristics, and the spatial dimension extracts the mean value / hot spot proportion / position temperature distribution characteristics, covering the key parameters of gas diffusion and thermodynamic coupling effect. The bidirectional LSTM captures the concentration time series dependence, and the CNN extracts the temperature spatial topology features, and through the cross-modal attention mechanism, the adaptive weighted fusion of time series dynamic characteristics and spatial distribution characteristics is realized. The minimum-maximum standardization and regional weight distribution eliminate the dimension difference and highlight the high-risk temperature area, respectively. Feature selection and dimension compression eliminate redundant information, and finally the generated 512-dimensional sparse feature vector has high information density and low computational complexity characteristics.

[0096] In summary, the embodiment realizes efficient representation of gas concentration and temperature multi-source heterogeneous data in an industrial scene through the cascade processing of noise suppression, feature extraction, modal fusion and sparse processing.

[0097] Embodiment 6:

[0098] On the basis of embodiment 5, the process of forming a gas concentration feature vector standardized to the [0, 1] interval in step S201 provided by the embodiment of the application includes the following steps:

[0099] Step S2011: The smoothed concentration data is subjected to fixed time length data interception processing to obtain a concentration trajectory segment containing 600 consecutive sampling points;

[0100] Step S2012: The concentration trajectory segment is subjected to global scanning comparison processing, and the concentration values of each sampling point in the concentration trajectory segment are compared in turn. A maximum value is updated and recorded in real time to determine and output an exact peak concentration value. The concentration trajectory segment is also subjected to continuous difference average processing to calculate the difference absolute values of the concentration values of adjacent sampling points in all concentration trajectory segments. Then, the difference absolute values are summed and divided by the total number of sampling points minus one to output a scalar value reflecting the average change intensity, thereby obtaining the concentration change per unit time representing the change trend. The concentration trajectory segment is also subjected to standard deviation calculation processing to obtain the dispersion degree of the concentration value distribution. The standard deviation of all sampling points is calculated to quantify the fluctuation range of the data around the average value, and finally a dispersion index representing the concentration stability is output. A feature set including the highest concentration reading, the arithmetic mean of the concentration difference of adjacent sampling points, and the dispersion degree of the concentration value distribution is formed.

[0101] Step S2013: The feature set is subjected to dynamic range determination processing to obtain the historical minimum and maximum values corresponding to each feature parameter. The actual value of each feature parameter is subjected to linear scaling calculation processing to obtain a new normalized value. The new value = (actual value-historical minimum value) / (historical maximum value-historical minimum value). All feature parameters are mapped to the value interval of 0 to 1 to finally form a standardized gas concentration feature vector.

[0102] In the above embodiment, the present embodiment establishes a gas concentration representation system with time sequence integrity, feature orthogonality and scale uniformity; provides an input vector meeting the scale invariance requirement for subsequent pattern recognition algorithms; realizes information compression expression of high-dimensional sensing data through feature space dimension reduction; and forms a unified feature extraction framework suitable for different gas media.

[0103] Embodiment 7:

[0104] On the basis of embodiment 5, the process for obtaining the temperature space feature vector of the region with a strengthened temperature > 40℃ in step S202 provided by the present embodiment comprises the following steps:

[0105] Step S2021: The temperature cloud space data is subjected to pixel response consistency calibration processing to obtain preliminary corrected data eliminating the response differences of each pixel. The preliminary corrected data is subjected to edge gradient adaptive compensation processing to form a corrected temperature field eliminating the edge temperature deviation.

[0106] Step S2022: The corrected temperature field is subjected to multi-scale space feature preservation processing to obtain a down-sampled temperature field preserving the original heat distribution features.

[0107] The down-sampled temperature field is subjected to regional statistical feature calculation processing to form a space feature set including the global temperature mean value, the high-temperature region proportion, and the extreme temperature point coordinate information.

[0108] Step S2023: The spatial feature set is processed by a temperature threshold activated weight distribution mechanism to obtain a basic feature weight distribution; and the basic feature weight distribution is processed by a high-temperature region feature enhancement operation to finally form a temperature spatial feature vector of a region with a reinforced temperature higher than 40℃.

[0109] In the above embodiment, the embodiment establishes a temperature field representation method with spatial consistency, scale adaptability and threshold orientation, realizes multi-dimensional analysis of heat distribution patterns through coupled expression of statistical features and spatial features, and converts a physical threshold into a nonlinear mapping rule of a feature space through a weight distribution mechanism, so that the output feature vector meets the sensitivity requirement of a heat anomaly detection task for a local high-temperature region.

[0110] Embodiment 8

[0111] On the basis of Embodiment 5, the process of obtaining a 512-dimensional sparse feature vector of fused gas concentration and temperature features in step S203 provided by the embodiment of the application includes the following steps:

[0112] Step S2031: The gas concentration feature vector is processed by a time-dependent relationship modeling architecture, a three-layer hidden unit is used to deeply mine the concentration change law, and a 256-dimensional time sequence feature expression is obtained; and the temperature spatial feature vector is simultaneously processed by a spatial feature abstraction architecture, a multi-layer convolution kernel and a spatial attention filtering mechanism are used to extract reinforced high-temperature region features, and a 512-dimensional spatial feature expression is obtained.

[0113] Step S2032: The 256-dimensional time sequence feature and the 512-dimensional spatial feature are processed by a multi-source feature interaction weight distribution, a correlation coefficient between the two feature modalities is calculated, the feature fusion weight is dynamically adjusted, and a 768-dimensional joint feature expression is formed.

[0114] Step S2033: The 768-dimensional joint feature expression is processed by a feature saliency filtering, the contribution of each dimension to the heat runaway identification is evaluated, and the first 512 most discriminative feature dimensions are retained; and then, a non-key dimension is set to zero, and finally, a 512-dimensional sparse feature vector of fused gas concentration and temperature features is obtained; only the feature dimensions that are critical to the heat runaway early warning are retained in the sparse feature vector, the remaining dimensions are assigned a value of zero, and sparse representation of the feature space is realized.

[0115] In the above embodiments, this embodiment deeply mines the temporal evolution law of gas concentration through a time-dependent relationship modeling architecture, strengthens the feature expression of high-temperature regions by adopting a spatial feature abstraction architecture, realizes intelligent fusion of cross-modal features by using multi-source feature interaction weight allocation, and finally achieves sparsity compression of feature space by feature saliency screening and setting non-critical dimensions to zero; the preceding output is directly used as the subsequent input to ensure the complete transmission and efficient utilization of feature information.

[0116] Example 9:

[0117] like Figure 5 As shown, based on Example 1, the process of forming the thermal runaway early warning signal and location result of the liquid-cooled battery cell in step S300 of this embodiment of the invention includes the following steps:

[0118] Step S301: The sparse feature vector fused with gas concentration and temperature features is processed by linear projection of the Gaussian random observation matrix to obtain a compressed data vector with a 90% reduction in dimensionality; the compressed data vector is processed by a reconstruction algorithm based on L1 norm constraints to recover a 512-dimensional reconstructed feature vector that is highly similar to the original feature vector; the compressed data vector is processed by optimization iterative calculation based on the sparsity characteristics of the signal to recover a 512-dimensional reconstructed feature vector that is highly similar to the original feature vector.

[0119] Step S302: The 512-dimensional reconstructed feature vector undergoes multi-index joint risk assessment processing. By simultaneously evaluating the weighted combination value of the gas concentration anomaly index and the temperature anomaly spatial distribution coefficient, a thermal runaway risk probability score is obtained. The thermal runaway risk probability score is processed by a dual-threshold early warning mechanism. When the score exceeds the preset threshold, different levels of thermal runaway early warning signals are generated.

[0120] Step S303: The spatial feature components in the reconstructed feature vector are processed by high-temperature region coordinate inversion. By decoding the spatial location information and temperature distribution pattern contained in the feature vector, the two-dimensional coordinate position of the heat source in the battery cell is determined. The two-dimensional coordinate position is processed by cell mapping transformation to finally form a precise positioning result specific to the number of a single liquid-cooled battery cell.

[0121] In the above embodiments, the present embodiment forms a thermal runaway early warning and positioning system based on compressed sensing and multi-modal feature fusion; through the compression and reconstruction technology of the sparse feature vector, the data dimension is significantly reduced while the key features of gas concentration and temperature are retained; through the multi-index joint risk assessment model, the synergistic analysis of gas concentration anomaly and temperature spatial distribution characteristics is realized, and a thermal runaway probability quantitative evaluation mechanism is established; finally, combined with the spatial feature decoding technology, the abstract abnormal signal is converted into the physical positioning of the specific coordinates inside the battery cell. The technical coupling of each step realizes the whole-chain processing from multi-dimensional sensor data to accurate early warning signal and spatial positioning, while ensuring the integrity of the feature information, improving the real-time performance of the system, and providing accurate spatial reference for subsequent thermal management intervention.

[0122] Embodiment 10:

[0123] On the basis of embodiment 9, the process of generating different levels of thermal runaway early warning signals when the score exceeds the preset threshold in step S302 provided by the present embodiment includes the following steps:

[0124] Step S3021: The 512-dimensional reconstructed feature vector is processed by the gas concentration anomaly index, and the concentration anomaly index quantifying the gas anomaly level is obtained by analyzing the numerical deviation degree of the gas concentration related dimensions in the reconstructed feature vector; at the same time, the spatial distribution coefficient of temperature anomaly is processed, and the spatial distribution coefficient representing the temperature anomaly degree is obtained by analyzing the distribution rule and abnormal mode of the spatial feature dimensions;

[0125] Step S3022: The concentration anomaly index and the spatial distribution coefficient are processed by dynamic weight fusion, and the weighted combination value comprehensively reflecting the abnormal degree of gas and temperature is obtained by automatically adjusting the weighting proportion according to the real-time reliability evaluation results of the two indexes;

[0126] The weighted combination value is converted and processed by a nonlinear risk mapping function, and the continuous numerical value is mapped to a thermal runaway risk probability score in the range of 0 to 1 through the saturation characteristics of the hyperbolic tangent function;

[0127] Step S3023: The thermal runaway risk probability score is processed by a hierarchical threshold comparison, and the corresponding risk level identifier is obtained by comparing the relative size relationship of the score value and the preset early warning threshold and emergency warning threshold; the risk level identifier is processed by the early warning signal generation logic, and finally the thermal runaway early warning signal containing different warning levels is formed; when the risk probability score exceeds the two thresholds at the same time, the system generates the highest level of emergency warning signal.

[0128] In the above embodiment, the present embodiment constructs a multi-level thermal runaway risk assessment and early warning signal generation mechanism; by extracting the gas concentration anomaly index and the temperature anomaly spatial distribution coefficient, the two-dimensional abnormal characteristics of gas and temperature are quantified; by dynamically weighting the fusion of the adaptive adjustment of the contribution weight of the gas and temperature index, the reliability of the comprehensive evaluation result is ensured; the nonlinear risk mapping function is used to convert the multi-dimensional features into normalized probability scores, eliminating the influence of different physical dimensions; finally, based on the hierarchical threshold comparison, the accurate division of the risk level is realized, and the step-by-step early warning signal output is formed. The synergistic effect of each step realizes the complete conversion from the original feature to the hierarchical early warning, and provides a quantitative decision basis for the early identification and hierarchical response of the thermal runaway state.

[0129] Embodiment 11:

[0130] On the basis of embodiment 10, the process of mapping the continuous numerical value to the thermal runaway risk probability score in the range of 0 to 1 in step S3022 provided by the present embodiment comprises the following steps:

[0131] Step S30221: The 512-dimensional reconstructed feature vector is subjected to gas concentration dimension deviation degree processing to obtain a concentration anomaly index reflecting the abnormal degree of each gas concentration dimension value; the 512-dimensional reconstructed feature vector is simultaneously subjected to temperature spatial distribution discrete feature analysis processing to obtain a spatial distribution coefficient representing the abnormal region distribution characteristics of the temperature;

[0132] Step S30222: The concentration anomaly index and the spatial distribution coefficient are subjected to dynamic weighting factor allocation processing, and different weight proportions are allocated according to the real-time data quality evaluation result to obtain a comprehensive quantitative value of the gas and temperature anomaly;

[0133] Step S30223: The comprehensive quantitative value is subjected to input range normalization processing to obtain a normalized value in the standard definition interval; the normalized value is subjected to saturation nonlinear transformation processing with S-shaped curve characteristics to form a probability output value showing gradual saturation characteristics; the probability output value is subjected to output range calibration processing to finally obtain a thermal runaway risk probability score limited in the range of 0 to 1.

[0134] In the above embodiment, the present embodiment constructs a nonlinear risk assessment system based on dynamic fusion of multi-dimensional features. The system extracts gas abnormal features through gas concentration dimension deviation degree processing, and captures temperature abnormal patterns through temperature spatial distribution discrete feature analysis processing; on this basis, dynamic weighting factor allocation processing realizes reliable adaptive fusion of the two types of features; the synergistic effect of each step realizes accurate mapping from the multi-dimensional feature space to the standardized risk score, ensuring that the evaluation result not only retains the key abnormal features of the original data, but also has cross-condition comparability, providing a unified quantitative benchmark for subsequent hierarchical early warning.

[0135] Embodiment 12

[0136] On the basis of Embodiment 11, the process of forming the probability output value with progressive saturation characteristics in step S30223 provided by the embodiment of the application comprises the following steps:

[0137] Step S302231: The normalized value is subjected to S-shaped curve function conversion processing, and the input normalized value is converted into a corresponding function output value through the inherent S-shaped nonlinear response characteristics of the S-shaped curve function, so that an intermediate conversion value with smooth transition characteristics is obtained; the S-shaped curve function can be a Logistic function or a hyperbolic tangent function;

[0138] Step S302232: The intermediate conversion value is subjected to asymptotic boundary convergence processing, and the output value is evaluated for saturation degree by calculating the asymptotic distance of the output value from the limit value, by using the mathematical characteristics of the S-shaped curve function that the output value asymptotically converges to two fixed limit values when the input value tends to positive or negative infinity, so as to form an uncalibrated probability value with upper and lower boundary asymptotic characteristics;

[0139] Step S302233: The uncalibrated probability value is subjected to linear affine transformation processing, and the uncalibrated probability value is adjusted from its original range to the standard probability interval of 0 to 1 by using a mathematical transformation method combining translation and scaling, so as to finally obtain a probability output value meeting the probability distribution characteristics, by establishing a linear mapping relationship between the uncalibrated probability value and the standard probability interval.

[0140] In the above embodiments, the S-shaped curve function conversion processing of the embodiment realizes smooth mapping from input to output by using a specific nonlinear function, avoiding the sudden change problem that may be caused by linear conversion; the asymptotic boundary convergence processing fully utilizes the inherent asymptotic characteristics of the function, ensuring that the output value is always within a controllable range; and the linear affine transformation processing realizes accurate calibration of the range by simple mathematical transformation. Both the characteristic information of the original data and the mathematical requirements of the final output value meeting the probability value are ensured.

[0141] Embodiment 13

[0142] On the basis of Embodiment 12, the process of adjusting the uncalibrated probability value from its original range to the standard probability interval of 0 to 1 in step S302233 provided by the embodiment of the application comprises the following steps:

[0143] Step S3022331: The uncalibrated probability value is subjected to original range extreme value determination processing, so as to obtain the actual minimum value and the maximum value of the uncalibrated probability value in the current value interval; the actual minimum value and the maximum value are subjected to target interval mapping relationship establishment processing, so as to form linear transformation parameters from the uncalibrated interval to the standard probability interval;

[0144] Step S3022332: The linear transformation parameter is processed by a translation operation, and a post-translation value with zero as a reference is obtained by subtracting the lower limit value of the original interval from the uncalibrated probability value; the post-translation value is processed by a scaling operation, and a preliminary scaling result is obtained by dividing the original interval width and multiplying the target interval width.

[0145] Step S3022333: The preliminary scaling result is processed by a target interval reference alignment operation, and a standardized probability output value distributed in the closed interval [0, 1] is finally obtained by adding the lower limit value of the standard probability interval.

[0146] In the above embodiment, the embodiment realizes the deterministic linear mapping from an arbitrary input range to the closed interval [0, 1] through three-stage operations of extreme value dynamic detection, translation-scaling operation and reference calibration. The input range is not required to be preset, and the optimality of the transformation parameter is ensured through real-time extreme value calculation; the precision loss in combined calculation is avoided through step-by-step operation, and the strict constraint of the boundary value is ensured; the linear transformation characteristic allows the original data distribution to be restored through inverse operation.

[0147] Embodiment 14:

[0148] On the basis of embodiment 13, the process of processing the post-translation value by a scaling operation in step S3022332 provided by the embodiment of the application comprises the following steps:

[0149] S30223321: The post-translation value is processed by a relative position scaling operation, and a unit proportion value eliminating the influence of the original interval range is obtained by performing a division operation on the post-translation value and the total span distance of the original interval, so as to convert the absolute value into a pure proportion value representing the relative position of the post-translation value in the original interval.

[0150] S30223322: The unit proportion value is processed by a target range mapping operation, and a multiplication operation is performed on the unit proportion value and the total span distance of the standard probability interval.

[0151] S30223323: The pure proportion value is converted into an actual value meeting the scale requirement of the target interval, and a preliminary scaling result matching the scale of the target interval is obtained.

[0152] In the above embodiments, the relative position ratio calculation process in this embodiment normalizes the values ​​through division, uniformly converting input values ​​of any range to the [0,1] ratio range; the target range mapping process restores the scale of the ratio through multiplication, mapping the ratio value to the specified target range; the two processing steps are executed sequentially based on the translated values ​​obtained in the previous steps and the established mapping relationship parameters, ensuring that the conversion process is mathematically rigorous: the relative position ratio calculation process depends on the translated values ​​and the original range width parameters of the previous step, while the target range mapping process is based on the unit ratio value output in the previous step and the target range width parameters, forming a complete data processing chain.

[0153] Example 15:

[0154] like Figure 6-8 As shown, based on Embodiments 1-14, the monitoring system for thermal runaway protection of energy storage power stations provided in this embodiment of the invention includes: a liquid-cooled box 1, an AI intelligent controller 2, a coolant storage tank 3, a condenser 4, a circulating pump 5, an electric valve 6, a pressure relief valve 7, an infrared imager 8, a photoacoustic module 9, and a liquid-cooled battery unit 10.

[0155] The liquid-cooled housing 1 is equipped with multiple liquid-cooled battery units 10. The liquid-cooled battery units 10 are connected to the inlet end of the coolant reservoir 3 through pipes. The outlet end of the coolant reservoir 3 is connected to the inlet end of the condenser 4. The outlet end of the condenser 4 is connected to the input end of the circulation pump 5. The output end of the circulation pump 5 is connected to one end of the electric valve 6. The other end of the electric valve 6 is connected to the liquid-cooled battery unit 10. The infrared imager 8 is embedded in the lower right end of the liquid-cooled housing 1. The photoacoustic module 9 is embedded in the upper right end of the liquid-cooled housing 1. The pressure relief valve 7 is embedded in the top of the liquid-cooled housing 1. The AI ​​intelligent controller 2 is connected to the pressure relief valve 7, the infrared imager 8 and the photoacoustic module 9 through a data cable.

[0156] In the above embodiment, the heat generated by the liquid-cooled battery unit 10 is carried away by the coolant, which flows into the coolant reservoir 3 through a pipeline. The coolant then enters the condenser 4 from the coolant reservoir 3 for cooling. The circulating pump 5 drives the coolant flow, and after the flow rate is regulated by the electric valve 6, it flows back to the liquid-cooled battery unit 10, forming a closed-loop cooling circuit. The pressure relief valve 7 automatically opens when the system pressure is too high to ensure safe operation. The infrared imager 8 monitors the temperature distribution of the liquid-cooled battery unit 10 in real time and detects abnormal temperature rises. The photoacoustic module 9 detects trace amounts of gas released in the early stages of battery thermal runaway. The system provides early warning of gas leaks, including CO; the AI ​​intelligent controller 2 receives monitoring data from the infrared imager 8 and the photoacoustic module 9, comprehensively analyzes the risk of thermal runaway, and controls the electric valve 6 to adjust the coolant flow rate, and triggers the pressure relief valve 7 to safely relieve pressure when necessary.

[0157] The infrared imager 8 of the embodiment provides a temperature cloud map, the photoacoustic module 9 detects trace gases, and early warning of thermal runaway is realized; the battery temperature is adjusted by circulating the cooling liquid, combined with the pressure relief valve 7 to prevent excessive pressure, and the risk of thermal runaway is reduced; the AI intelligent controller 2 comprehensively analyzes the temperature and gas data, realizes adaptive regulation and control, and improves the system safety redundancy; the closed-loop cooling system maintains the stability of the battery temperature, and reduces the risk of performance degradation or failure caused by overheating.

[0158] The embodiment combines the gas detection accuracy of photoacoustic spectroscopy and the spatial resolution of infrared imaging to realize a thermal runaway protection system for energy storage power stations that can accurately locate thermal runaway in the early stage. In view of the technical difficulties of early detection of thermal runaway in energy storage power stations, a collaborative monitoring scheme is proposed that combines photoacoustic spectroscopy gas detection and infrared imaging. The core is to realize accurate identification and rapid response of thermal runaway hazards through multi-physical field information fusion and edge computing optimization. The embodiment takes the high-sensitivity photoacoustic module 9 and the high-precision infrared thermal imager 8 as the sensing core: the photoacoustic module 9 can capture trace leaks of 1 ppm level and 5 ppm level CO in real time, with a response time controlled within 500 ms, accurately capturing the gas characteristics in the early stage of thermal runaway; the infrared thermal imager 8 generates a temperature cloud map with a temperature measurement accuracy of ±0.1°C, and uses AI algorithms to locate hot spots within 2 cm, synchronously obtaining spatial temperature anomaly information. The data of gas characteristics and spatial temperature anomaly information are deeply fused by the edge computing unit, the correlation analysis of gas concentration and temperature gradient is used to eliminate false positives, and the data volume is compressed by 90% with the help of the compression sensing algorithm, ensuring that the transmission delay is less than 100 ms, solving the data congestion problem of traditional systems. The overall process forms a full-dimensional monitoring closed loop from micro gas characteristics to macro temperature anomalies through the collaborative mechanism of gas early warning-temperature verification-spatial positioning-edge compression-quick transmission, which not only makes up for the defects of insufficient sensitivity of traditional gas detection and fuzzy infrared positioning, but also breaks through the bottleneck of cloud response lag through real-time processing at the edge, ultimately realizing accurate positioning and efficient early warning of thermal runaway in the early stage of liquid-cooled battery units 10 in energy storage power stations, and providing core technical support for safe operation and maintenance.

[0159] As shown in Figure 7 , the process of photoacoustic spectroscopy gas detection, the photoacoustic module 9 uses photoacoustic spectroscopy technology to realize high-precision detection of and CO. The core principle is based on the absorption characteristics of substances to specific wavelengths of light. When specific wavelength laser is irradiated to the substance containing When the gas molecules absorb the light energy of corresponding wavelength and jump to high-energy state, the molecules in high-energy state will return to low-energy state through non-radiative transition, and the released energy is converted into heat energy, which makes the gas temperature rise and further leads to periodic fluctuations of gas pressure, forming the photoacoustic signal. At the same time, a microphone or pressure sensor with high sensitivity is equipped as a detector to accurately capture the weak photoacoustic signal. By analyzing and processing the intensity and frequency of the photoacoustic signal, the concentration of H2 and CO can be accurately determined The whole detection process responds quickly, meeting the requirement of <500 ms. In this embodiment, the gas sample in the energy storage power station is introduced into the photoacoustic cell by a miniature air pump or natural diffusion; then, the laser emits laser of specific wavelength into the photoacoustic cell, which interacts with H2 and CO molecules in the gas sample to produce photoacoustic signals; finally, the microphone or pressure sensor in the photoacoustic cell captures the photoacoustic signals and converts them into electrical signals transmitted to the data processing unit, and the gas concentration is obtained by analysis and calculation.

[0160] As shown in Figure 8 , the working principle of the infrared thermal imager 8 is as follows: the infrared thermal imager 8 generates a temperature cloud image by detecting the infrared radiation emitted by the liquid-cooled battery unit 10 equipment surface in the energy storage power station. As long as the temperature of an object is higher than absolute zero, it will radiate infrared rays, and the size of the radiation energy is related to the temperature of the object. The infrared detector in the infrared thermal imager 8 can receive these infrared radiations and convert them into electrical signals; to achieve the performance of ±0.1℃ accuracy, the infrared thermal imager 8 uses high-performance infrared focal plane detectors, which have high resolution and high sensitivity and can accurately perceive small temperature differences on the surface of an object. In this embodiment, the radiation intensity corresponding to different temperatures is converted into different gray scales or colors by processing and converting the electrical signals, thereby generating a temperature cloud image reflecting the temperature distribution of the equipment surface; the infrared thermal imager 8 realizes real-time scanning of the target area by collecting infrared radiation emitted by the equipment surface in the energy storage power station through the infrared detector. First, the infrared lens receives the infrared radiation emitted by the equipment surface in the energy storage power station and focuses it on the infrared focal plane detector; the infrared focal plane detector converts the infrared radiation signal into an electrical signal, generates raw thermal image data after preprocessing such as non-uniformity correction and noise reduction filtering; and finally, the electrical signal is converted into corresponding temperature values based on the blackbody radiation law through a temperature calibration algorithm, and is output in the form of a temperature cloud image. The complete acquisition process responds in <33 ms, ensuring time synchronization with the detection of the photoacoustic module 9.

[0161] The AI data fusion principle of this embodiment is: directly performing data fusion at the sensor end, which is to fuse the photoacoustic signals detected by the photoacoustic module 9 and the temperature cloud image generated by the infrared thermal imager 8. The AI model is trained to learn the relationship between the photoacoustic signals and the temperature cloud image, and the AI model is used to analyze and process the fused data to obtain the concentration of H2 and CO in the energy storage power station. The AI model is trained by a large amount of data, and the training process is as follows: first, the photoacoustic signals and the temperature cloud image are collected; then, the collected data is labeled; finally, the labeled data is used to train the AI model. The AI model is trained to learn the relationship between the photoacoustic signals and the temperature cloud image, and the AI model is used to analyze and process the fused data to obtain the concentration of H2 and CO in the energy storage power station. ​The temperature cloud map generated by the infrared thermal imager 8 and the hot spot area data identified by the AI algorithm are integrated and processed; these data reflect the state information of the energy storage power station from different dimensions, and through data fusion, data redundancy and contradictions can be eliminated, the reliability and accuracy of the data can be improved, and more comprehensive and effective information can be provided for subsequent analysis and decision-making; the compressive sensing algorithm is based on the sparsity principle of signals, that is, many natural signals are sparse in a certain transform domain. The data obtained by the photoacoustic module 9 and the infrared thermal imager 8 has the characteristic of sparsity in a certain transform domain; the compressive sensing algorithm obtains a small amount of measurement values by designing a suitable measurement matrix to make a small amount of random measurement on the original data. Then, using the sparse reconstruction algorithm, the original data is reconstructed from these small amount of measurement values, thereby achieving a substantial reduction in data volume; through the compressive sensing algorithm, the data volume can be reduced by 90%, greatly reducing the burden of data transmission. At the same time, since the data compression processing is completed at the sensor end, the amount of data transmitted is reduced, so that the data transmission delay is <100 ms, effectively solving the data congestion problem caused by large data volume in the traditional system.

[0162] Gas concentration data processing of the embodiment: the 100ms / second output by the photoacoustic module 9 and CO concentration data 100ms / time first pass time filtering and feature extraction: using Kalman filtering algorithm to eliminate transient noise, for example, when the concentration appears ±0.3ppm pulse fluctuation, through the prediction-update mechanism to correct to smooth curve; extract key timing features: including concentration peak value, average change rate such as the rate of concentration from 1ppm to 3ppm, concentration fluctuation variance to judge whether it is a persistent leakage, and standardize these features to the interval [0, 1] as the time sequence input of the AI model.

[0163] Temperature cloud data processing of the embodiment: the 640x512 temperature cloud 33ms / frame generated by the infrared thermal imager 8 needs to complete spatial feature enhancement: using non-uniformity correction algorithm to eliminate lens response difference, pixel-level compensation is performed on the temperature deviation of the edge area such as-0.5℃ error caused by dust shielding; using image pyramid technology to reduce the temperature cloud, while reducing the calculation amount, extracting temperature mean, hot spot area ratio, highest temperature position and other spatial features.

[0164] AI model architecture and training process of the embodiment: gas feature branch: using bidirectional LSTM long short-term memory network, the input is 100 time steps of gas concentration features including and CO change rate, peak value, etc., through 3-layer LSTM unit to capture the time sequence dependence of concentration change such as The hysteresis response of CO concentration after concentration increases is analyzed, outputting a 256-dimensional temporal feature vector. Temperature feature branch: Based on an improved ResNet-18 convolutional neural network, the classification header before the fully connected layer is removed, retaining the 512-dimensional spatial features output by the convolutional layer. A spatial attention module is embedded in the network, assigning higher weights to regions with temperatures >40℃ to enhance hotspot feature extraction.

[0165] This embodiment combines AI algorithms with compressed sensing algorithms, essentially based on a collaborative optimization mechanism of multimodal feature sparsity: the AI ​​module controller 2 extracts spatial features of temperature cloud maps and temporal features of gas concentration through an improved ResNet-18 and LSTM network, and forms a highly discriminative sparse feature vector after being filtered by an attention mechanism, providing a natural sparsity prior for compressed sensing; the compressed sensing algorithm designs a Gaussian measurement matrix for this fused feature, and achieves a 90% compression rate through L1 norm constraints, completing lightweight compression processing at the edge and significantly reducing transmission bandwidth requirements; the cloud and local terminals reconstruct the feature vector based on the gradient descent L1 norm minimization algorithm, and combine it with the prior knowledge of the AI ​​model to optimize the reconstruction accuracy; the decision layer ensures that the accuracy of thermal runaway risk judgment based on reconstructed features remains above 95% by training a robust model adapted to compression errors.

[0166] This embodiment overcomes the limitations of traditional single detection methods by synergistically integrating photoacoustic spectroscopy and infrared imaging technologies: the photoacoustic module... With ultra-high sensitivity of 1ppm and CO5ppm, it can capture trace gas releases in the initial stage of thermal runaway, detecting anomalies 3-5 minutes earlier than traditional electrochemical sensors. The infrared thermal imager's temperature measurement accuracy of ±0.1℃ and the AI ​​algorithm's positioning error of <2cm can accurately identify minute temperature differences of 0.5-1℃ caused by local micro-short circuits in the battery pack, avoiding missed detections due to temperature monitoring blind spots. The correlation verification mechanism between the two data sets shows that the overlap between gas anomalies and hotspot areas is >80%, reducing the false alarm rate from 15%-20% in traditional systems to below 3%, significantly improving the reliability of early warnings.

[0167] Figure 9 A block diagram of an exemplary electronic device suitable for implementing embodiments of the present invention is shown.

[0168] The electronic device may include a central processing unit / microprocessor / main control chip, etc. 11; and a storage medium 12 coupled to the central processing unit / microprocessor / main control chip, etc. 11, and storing computer-executable instructions therein for performing the steps of various methods of embodiments of the present invention when executed by the processor.

[0169] The central processing unit / microprocessor / main control chip, etc., 12 may include, but are not limited to, one or more processors or microprocessors.

[0170] The storage medium 12 can include, but is not limited to, for example, a random access memory (RAM), a read only memory (ROM), a flash memory, an EPROM memory, an EEPROM memory, a register, a computer storage medium (such as a hard disk, a floppy disk, a solid state disk, a removable disk, a CD-ROM, a DVD-ROM, a Blu-ray disk, etc.).

[0171] In addition, the electronic device can further include, but is not limited to, a data bus 13, an input / output bus / external bus / device bus, etc. 14, a display 15, and an input / output device 16 (for example, a keyboard, a mouse, a speaker, etc.), etc.

[0172] The central processing unit / microprocessor / master control chip, etc. 11 can communicate with external devices (15, 16, etc.) via a wired or wireless network (not shown) through the input / output bus / external bus / device bus, etc. 14.

[0173] The storage medium 12 can further store at least one computer executable instruction for performing the steps of the various functions and / or methods in the embodiments described in the present technology when executed by the central processing unit / microprocessor / master control chip, etc. 11.

[0174] In one embodiment, the at least one computer executable instruction can also be compiled or composed as a software product in which one or more computer executable instructions are executed by a processor to perform the steps of the various functions and / or methods in the embodiments described in the present technology.

[0175] Figure 10 A schematic diagram of a computer readable storage medium according to an embodiment of the present application is shown.

[0176] As Figure 10 shown, the non-transitory computer readable storage medium 18 stores instructions, for example, computer readable instructions 17. When the computer readable instructions 17 are executed by a processor, the various methods described above can be performed. The non-transitory computer readable storage medium includes, but is not limited to, for example, a volatile memory and / or a non-volatile memory. The volatile memory can include, for example, a random access memory (RAM) and / or a cache memory, etc. The non-transitory non-volatile memory can include, for example, a read only memory (ROM), a hard disk, a flash memory, etc. For example, the non-transitory computer readable storage medium 18 can be connected to a computing device such as a computer, and then when the computing device executes the computer readable instructions 17 stored on the non-transitory computer readable storage medium 18, the various methods described above can be performed.

[0177] In several embodiments provided by the present application, it should be understood that the disclosed apparatus and method can be implemented in other manners. For example, the described apparatus embodiments are merely schematic. The unit division is merely logical function division. There can be other division manners in actual implementation. For example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not implemented. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections between different units, or the among different units, can be indirect couplings or communication connections through some interfaces, devices, or unit intermediaries, and can be in electrical, mechanical, or other forms.

[0178] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they can be located in one place, or distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.

[0179] In addition, the functional units in each embodiment of the present application can be integrated into a processing unit, or each unit can be a physically independent unit, or two or more units can be integrated into a unit. The integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

[0180] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a plurality of instructions for executing all or part of the steps of the embodiments of the present application by a computer device, which can be a personal computer, a server, or a network device, etc. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0181] The above embodiments are merely used to describe the technical solutions of the present application, rather than limit them. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or make equivalent replacements to some technical features. Such modifications or replacements do not cause the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A monitoring method for thermal runaway protection in an energy storage power station, characterized in that, Includes the following steps: The thermal runaway characteristic signals of the liquid-cooled battery cell are simultaneously acquired and processed by a photoacoustic module and an infrared imager to obtain... CO concentration time series data and temperature contour map spatial data; The CO concentration time series data and temperature cloud map spatial data are filtered, feature extracted and multimodal fused by the AI ​​intelligent controller to obtain a sparse feature vector that integrates gas concentration and temperature features. The CO concentration time series data were processed by time series noise suppression based on the Kalman filter algorithm to obtain smoothed concentration data with pulse fluctuations eliminated; the smoothed concentration data were then processed by multi-dimensional feature extraction to obtain a time series feature set containing the concentration peak, concentration change rate and concentration fluctuation variance within 1 minute. The temporal feature set is processed by min-max normalization to form a gas concentration feature vector normalized to the [0,1] interval; the spatial data of the temperature cloud map is processed by non-uniformity correction and pixel-level compensation to obtain a corrected temperature field that eliminates lens response differences and edge temperature deviations; the corrected temperature field is processed by image pyramid downsampling and spatial feature extraction to form a spatial feature set containing the mean temperature, hot spot area ratio, and the location of the highest temperature; the spatial feature set is processed by regional weight allocation to obtain a temperature spatial feature vector that enhances the features of areas with temperatures >40℃; the gas concentration feature vector and the temperature spatial feature vector are processed by a bidirectional long short-term memory network and a convolutional neural network for dual-branch feature extraction to obtain 256-dimensional temporal features and 512-dimensional spatial features; Temporal and spatial features are fused through a cross-modal attention mechanism to form a 768-dimensional joint feature vector. The 768-dimensional joint feature vector is then processed by feature selection and dimensionality compression to finally obtain a 512-dimensional sparse feature vector that integrates gas concentration and temperature features. The sparse feature vector, which integrates gas concentration and temperature characteristics, is processed by compressed sensing and risk decision-making models to form a thermal runaway early warning signal and location result for the liquid-cooled battery cell.

2. The monitoring method for thermal runaway protection of an energy storage power station as described in claim 1, characterized in that, get The process of combining CO concentration time-series data with temperature contour map spatial data includes the following steps: The ambient gas in the energy storage power station is sampled by a miniature gas pump of a photoacoustic module and processed by laser wavelength irradiation to obtain a solution containing... The periodic pressure fluctuation signal generated after CO molecules absorb energy is captured by a high-sensitivity microphone and processed by electrical signal conversion to form the original photoacoustic voltage signal. This original photoacoustic voltage signal is then analyzed and calculated by the data processing unit based on signal strength and frequency to obtain a timestamped signal. Concentration sequence and CO concentration sequence; The infrared radiation from the surface of the liquid-cooled battery cell is processed by an infrared thermal imager, which includes an infrared lens focusing and an infrared focal plane detector, to obtain a raw electrical signal array corresponding to the temperature distribution. The original electrical signal array undergoes non-uniformity correction and noise reduction filtering preprocessing to form standardized radiation intensity data; The standardized radiation intensity data is converted and processed based on the temperature calibration algorithm of the blackbody radiation law, and finally 640×512 pixel temperature cloud map spatial data is generated. The photoacoustic gas detection channel of the photoacoustic module and the infrared thermal imaging channel of the infrared thermal imager are coordinated through a hardware-level time synchronization mechanism to obtain aligned time-space synchronized acquisition data. Synchronously acquired data undergoes time-series tagging and spatial registration to form a unified time reference. CO concentration time series data and temperature contour map spatial data.

3. The monitoring method for thermal runaway protection of an energy storage power station as described in claim 1, characterized in that, The process of generating a thermal runaway early warning signal and location result for a liquid-cooled battery cell includes the following steps: The sparsed feature vector, which integrates gas concentration and temperature characteristics, is processed by linear projection onto a Gaussian random observation matrix to obtain a dimensionality-reduced compressed data vector. This compressed data vector is then processed by a reconstruction algorithm based on L1 norm constraints to recover a 512-dimensional reconstructed feature vector that highly approximates the original feature vector. quantity; The 512-dimensional reconstructed feature vector undergoes multi-index joint risk assessment processing. By simultaneously evaluating the weighted combination value of the gas concentration anomaly index and the temperature anomaly spatial distribution coefficient, a thermal runaway risk probability score is obtained. The thermal runaway risk probability score is then processed by a dual-threshold early warning mechanism. When the score exceeds a preset threshold, different levels of thermal runaway early warning signals are generated. The spatial feature components in the reconstructed feature vector are processed by high-temperature region coordinate inversion. By decoding the spatial location information and temperature distribution pattern contained in the feature vector, the two-dimensional coordinate position of the heat source in the battery cell is determined. The two-dimensional coordinate position is then processed by cell mapping transformation to finally form a precise positioning result specific to the number of a single liquid-cooled battery cell.

4. The monitoring method for thermal runaway protection of energy storage power stations as described in claim 3, characterized in that, The process of generating different levels of thermal runaway early warning signals when the score exceeds a preset threshold includes the following steps: The 512-dimensional reconstructed feature vector is processed by the gas concentration anomaly index. Analysis of the reconstructed feature vector... The degree of deviation of the values ​​of gas concentration-related dimensions is used to obtain the concentration anomaly index, which quantifies the gas anomaly level; at the same time, after processing the spatial distribution coefficient of temperature anomaly, the spatial distribution coefficient characterizing the degree of temperature anomaly is obtained by analyzing the distribution law and anomaly pattern of spatial feature dimensions. The concentration anomaly index and spatial distribution coefficient are dynamically weighted and fused. The weighting ratio is automatically adjusted according to the real-time reliability assessment results of the two indicators to obtain a weighted combination value that comprehensively reflects the degree of gas and temperature anomalies. The weighted combination value is transformed by a nonlinear risk mapping function. Through the saturation characteristics of the hyperbolic tangent function, the continuous value is mapped to a thermal runaway risk probability score in the range of 0 to 1. The thermal runaway risk probability score is processed by a tiered threshold comparison. By comparing the relative magnitude of the score with the preset early warning threshold and emergency warning threshold, the corresponding risk level identifier is obtained. The risk level identifier is then processed by the warning signal generation logic to form thermal runaway warning signals containing different warning levels. When the risk probability score exceeds both thresholds at the same time, the highest level emergency warning signal is generated.

5. The monitoring method for thermal runaway protection of an energy storage power station as described in claim 4, characterized in that, The process of mapping continuous numerical values ​​to a thermal runaway risk probability score in the range of 0 to 1 includes the following steps: The 512-dimensional reconstructed feature vector, after undergoing gas concentration dimension deviation processing, yields a result reflecting the concentrations of each gas. The concentration anomaly index represents the degree of numerical anomaly in dimensionality; the 512-dimensional reconstructed feature vector is simultaneously processed by the discrete feature analysis of temperature spatial distribution to obtain the spatial distribution coefficient characterizing the distribution characteristics of the temperature anomaly region. The concentration anomaly index and spatial distribution coefficient are processed by dynamic weighting factor allocation, and differentiated weight ratios are assigned according to the real-time data quality assessment results to obtain a comprehensive quantitative value of gas and temperature anomalies. The comprehensive quantized value is normalized within the input range to obtain a normalized value that falls within the standard defined range; The normalized values ​​are processed by a saturated nonlinear transformation with S-shaped curve characteristics to form a probabilistic output value with asymptotic saturation characteristics. The probabilistic output value is then calibrated for the output range to finally obtain a thermal runaway risk probability score limited to the range of 0 to 1.

6. The monitoring method for thermal runaway protection of an energy storage power station as described in claim 5, characterized in that, The process of generating probabilistic output values ​​that exhibit asymptotic saturation characteristics includes the following steps: The normalized values ​​are processed by the S-curve function transformation. Through its inherent S-shaped nonlinear response characteristics, the input normalized values ​​are converted into the corresponding function output values, resulting in intermediate transformed values ​​with smooth transition characteristics. The intermediate transformation values ​​undergo asymptotic boundary convergence processing, utilizing an S-curve function as the input value approaches positive. The mathematical property that the output value asymptotically converges to two fixed limit values ​​at negative infinity is used to evaluate the degree of saturation by calculating the asymptotic distance between the output value and the limit values, thus forming an uncalibrated probability value with asymptotic characteristics at the upper and lower boundaries. The uncalibrated probability value is processed by linear affine transformation. By establishing a linear mapping relationship between the uncalibrated probability value and the standard probability interval, a mathematical transformation method combining translation and scaling is used to adjust the uncalibrated probability value from its original range to the standard probability interval of 0 to 1, and finally obtain a probabilistic output value that conforms to the probability distribution characteristics.

7. The monitoring method for thermal runaway protection of an energy storage power station as described in claim 6, characterized in that, The process of adjusting an uncalibrated probability value from its original range to a standard probability interval of 0 to 1 includes the following steps: The uncalibrated probability value is processed by determining the extreme values ​​of the original range to obtain the actual minimum and maximum values ​​of the uncalibrated probability value in the current numerical range; the actual minimum and maximum values ​​are processed by establishing the target range mapping relationship to form the linear transformation parameters from the uncalibrated range to the standard probability range; The linear transformation parameters are processed by translation, which involves subtracting the lower limit of the original interval from the uncalibrated probability value to obtain the translated value with zero as the reference; the translated value is then scaled. The preliminary scaling result is obtained by dividing by the original interval width and then multiplying by the target interval width. The initial scaling results are aligned to the target interval baseline and then subjected to the addition of a lower bound of the standard probability interval. The calculation of the values ​​ultimately yields standardized probability output values ​​distributed within the closed interval between 0 and 1.

8. The monitoring method for thermal runaway protection of an energy storage power station as described in claim 7, characterized in that, The process of scaling the translated values ​​involves the following steps: After translation, the values ​​are processed by relative position scaling. By dividing the translated values ​​by the total span of the original interval, the absolute values ​​are converted into pure proportional values ​​that represent their relative positions within the original interval, thus obtaining unit proportional values ​​that eliminate the influence of the original interval range. The unit proportion value is processed by target range mapping by multiplying the unit proportion value by the total span distance of the standard probability interval; The pure proportional values ​​are converted into actual values ​​that meet the target interval scale requirements to obtain the preliminary scaling results that match the target interval scale.

9. A monitoring system for thermal runaway protection in an energy storage power station, characterized in that, The monitoring method for thermal runaway protection of an energy storage power station as described in any one of claims 1 to 8 is characterized by comprising: liquid... Cold box, AI intelligent controller, coolant reservoir, condenser, circulating pump, electric valve, pressure relief valve, infrared imager, photoacoustic module, liquid-cooled battery unit; The liquid-cooled housing contains multiple liquid-cooled battery units. These units are connected to the inlet of a coolant reservoir via pipes. The outlet of the coolant reservoir is connected to the inlet of a condenser. The outlet of the condenser is connected to the input of a circulating pump. The output of the circulating pump is connected to one end of an electric valve, and the other end of the electric valve is connected to the liquid-cooled battery units. An infrared imager is embedded in the lower right of the liquid-cooled housing, a photoacoustic module is embedded in the upper right, and a pressure relief valve is embedded in the top. An AI intelligent controller is connected to the pressure relief valve, the infrared imager, and the photoacoustic module via data cables.

Images