Electrochemical energy storage thermal runaway multi-dimensional monitoring method and device for dynamic feature fusion analysis

Abstract

The invention discloses an electrochemical energy storage thermal runaway multi-dimensional monitoring method and device for dynamic feature fusion analysis, and particularly relates to the technical field of electrochemical energy storage. Thermal runaway early-stage feature parameters from an electrochemical energy storage battery are recognized through a multi-dimensional detector, a voltage monitoring module and an expansion force monitoring module; the monitored thermal runaway early-stage characteristic parameters are transmitted to an edge calculation unit through a data bus for data preprocessing, and a standardized data stream is formed through a multi-source data fusion analysis engine configured by a central processing unit; when the standardized data flow is transmitted to a central processing unit through a data bus, the central processing unit performs dynamic correlation analysis on multi-source data, and analyzes and judges the thermal runaway risk probability in real time based on a composite monitoring model of multi-dimensional data weight fusion; then, a confidence coefficient iteration sub-module carries out iterative calculation on the thermal runaway risk probability.

Description

Technical Field

[0001] The present invention relates to the technical field of electrochemical energy storage, and particularly relates to a multi-dimensional monitoring method and device for thermal runaway of electrochemical energy storage with dynamic feature fusion analysis. Background Art

[0002] Driven by the "dual carbon" goal, electrochemical energy storage technology has become the core infrastructure to support the new power system, and the construction of its thermal safety prevention and control system faces severe challenges. Currently, the thermal runaway of electrochemical energy storage power stations mainly stems from incentives such as thermal abuse, electrical abuse, and mechanical abuse. The traditional detection system relies on discrete temperature sensing and gas detection after the explosion valve, and can only capture obvious features such as sudden temperature rise and voltage collapse during the thermal runaway propagation period, and there are monitoring blind spots for early signs such as BMS voltage monitoring, micro-smoke release, and mechanical deformation. The existing technology is limited by the single-dimensional threshold judgment and static data fusion mechanism, resulting in an insufficient early warning window period and unable to meet the front-end monitoring requirements of extremely early warning and precise positioning. There is an urgent need to build a new monitoring system that integrates multi-physical field dynamic perception of "spatial temperature - expansion force - voltage signal - characteristic gas - smoke" and low-latency feature decoupling, and break through the identification bottleneck of the key stages of "chemical heat generation - gas expansion - diaphragm rupture" in the thermal runaway chain reaction of the existing technology. Summary of the Invention

[0003] Therefore, the present invention provides a multi-dimensional monitoring method and device for thermal runaway of electrochemical energy storage with dynamic feature fusion analysis to solve the problems raised in the background art.

[0004] To achieve the above object, the present invention provides the following technical solution: A multi-dimensional monitoring method for thermal runaway of electrochemical energy storage with dynamic feature fusion analysis, in which the early thermal runaway characteristic parameters from the electrochemical energy storage battery are respectively identified by a multi-dimensional detector, a voltage monitoring module, and an expansion force monitoring module, and the monitored early thermal runaway characteristic parameters are transmitted to the edge computing unit through a data bus for data preprocessing, and a multi-source data fusion analysis engine configured by a central processing unit is used to form a standardized data stream;

[0005] When the standardized data stream is transmitted to the central processor through the data bus, the central processor performs dynamic correlation analysis on the multi-source data, and based on a composite monitoring model of multi-dimensional data weight fusion, analyzes and judges the thermal runaway risk probability in real time; subsequently, the confidence iteration sub-module iteratively calculates the thermal runaway risk probability;

[0006] Preferably, the steps for obtaining the early thermal runaway characteristic parameters of the electrochemical energy storage battery are as follows: The multi-dimensional detector extracts the gas inside the energy storage module housing at a constant flow rate through its own detection holes, the voltage monitoring module captures the voltage fluctuation signal of the energy storage element inside the energy storage module housing through the electrode wires, and the stress gauge array of the expansion force monitoring module measures the strain distribution on the surface of the energy storage module housing.

[0007] Preferably, in view of the problems that the traditional static threshold criterion has response lag and outlier interference in multiple application scenarios and cannot flexibly cope with the dynamic changing environment and data, a dynamic monitoring method of multi-dimensional fusion is proposed. Specifically: The gas inside the energy storage module housing is extracted at a constant flow rate through the detection holes of the multi-dimensional detector, and the following operations are performed synchronously:

[0008] The distributed optical fiber temperature sensor array monitors the temperature distribution of 8×8 grid points inside the energy storage module housing with a sampling period of 0.5 s.

[0009] The multi-spectral scanner enhances the ppm-level monitoring of characteristic gases such as CO and H2 through a Pt-Pd catalytic coating.

[0010] Preferably, the early thermal runaway characteristic parameters detected are transmitted to the edge computing unit through the data bus for data preprocessing, and the following operations are performed in sequence:

[0011] Wavelet denoising: The signal is decomposed into different frequency sub-bands by wavelet transform, and the threshold processing is performed on the high-frequency sub-band coefficients to remove noise. The soft threshold function used is:

[0012]

[0013] where is the processed wavelet coefficient, ω j,k is the original wavelet coefficient, sgn is the sign function, and λ is the threshold;

[0014] (x) + = max(x, 0);

[0015] Spatio-temporal registration: For time series data, the time of data collected by different sensors needs to be synchronized. The time interpolation method, such as linear interpolation, can be used. For sensor data at different positions in space, a spatial coordinate mapping relationship can be established based on the structure of the energy storage cabin.

[0016] Linear interpolation formula: Given time points t1, t2 and corresponding data y1, y2, for time point t, the interpolated data is:

[0017]

[0018] Outlier rejection: The 3σ criterion based on statistics is adopted.

[0019] Assume that the data follows a normal distribution, and the data point x i , if then it is determined that x i is an outlier and is rejected; where is the data mean, and σ is the standard deviation;

[0020]

[0021] Dimensional normalization is performed using the maximum-minimum normalization method, and the formula is:

[0022]

[0023] where x is the original data, x min and x max are the minimum and maximum values of the data, and x norm is the normalized data.

[0024] Preferably, the multi-source data fusion analysis engine includes:

[0025] A dynamic feature extraction sub-module for extracting and analyzing multi-dimensional relevant data;

[0026] A dynamic data weight fusion model that generates the probability of thermal runaway risk using the weight relationship of the standardized data stream;

[0027] A confidence iteration sub-module that integrates the XGBoost classifier to perform confidence iteration on the data to improve data reliability.

[0028] Preferably, the dynamic feature extraction sub-module extracts and analyzes multi-dimensional relevant data through a composite monitoring model of multi-dimensional data weight fusion, including:

[0029] Feature extraction. For the temperature data T(x, y, z, t) collected by the temperature sensor, the temperature gradient is:

[0030]

[0031] In the formula, x, y, and z represent the three coordinate axes in the space rectangular coordinate system, representing the position (the unit is usually meter, m), and t is the time variable (the unit is usually second, s). The temperature field may change with time, but only the spatial coordinates are considered during the gradient calculation, represents the direction and magnitude of the maximum change rate of the temperature field in space, represents the change rate of temperature in the x direction (unit: °C / m or K / m), represents the change rate of temperature in the y direction, represents the change rate of temperature in the z direction;

[0032] In the discrete case, the difference approximation is adopted, and the formula is as follows:

[0033]

[0034] In the formula, Δx represents the spatial step (unit: m), representing the distance between adjacent grid points;

[0035] Let the sequence of gas concentration varying with time be C(t), and the gas concentration change rate r is as follows:

[0036]

[0037] where Δt is the time interval;

[0038] Perform a fast Fourier transform FFT on the voltage data V(t) obtained by the voltage acquisition terminal:

[0039]

[0040] In the formula, V(t): the time-domain voltage signal, representing the continuous voltage value varying with time t (unit: volt, V); V(f): the frequency-domain voltage spectrum, representing the complex value corresponding to the frequency variable f (unit: V·s or V / Hz); f: the frequency variable, with the unit of hertz (Hz); j: the imaginary unit, satisfying that the square of j is -1 (j is commonly used in engineering, and i is commonly used in mathematics); e -j2πft : the complex exponential function, describing the sine wave basis function with the frequency variable f;

[0041] Thus, the voltage spectrum V(f) is obtained, and the peak value and frequency distribution characteristics in the spectrum are analyzed;

[0042] For the stress data collected by the expansion force detection module, by setting the threshold S th and the window length N, within the data window, if and S(i)>S th , then S(i) is the stress wave peak;

[0043] In the formula, i is the distance from the stress gauge, and S is the normal stress;

[0044] Let the smoke concentration data M(x, y, z, t) collected by the photoelectric smoke detector, similar to the temperature gradient calculation, the smoke concentration gradient:

[0045]

[0046] In the formula represents the direction and magnitude of the maximum change rate of the smoke concentration in space; represents the change rate of the temperature in the x direction (unit: °C / m or K / m); represents the change rate of the temperature in the y direction; represents the change rate of the temperature in the z direction;

[0047] Realize the prediction of thermal runaway risk probability through a dynamic data weight fusion model: Input multi-source feature vectors into the XGBoost classifier, and generate the thermal runaway risk probability by using the weight relationship of each standardized data stream. Failure mode and effect analysis (FMEA): Determine the severity S, occurrence O, and detection D of each parameter according to historical failure data, and calculate the risk priority number RPN = S × O × D as the weight basis. The risk probability calculation model is as follows:

[0048]

[0049] In the formula, wi is the dynamic weight of the i-th parameter, and f(xi) is the risk contribution function after parameter normalization (such as Sigmoid, piecewise linear function);

[0050] Preferably, perform confidence iteration calculation through the XGBoost classifier in the central processing unit, and the iterative optimization method includes the following steps:

[0051] Multi-source evidence fusion, adopt the improved D-S evidence theory to fuse the confidence of multi-sensor data to overcome the conflict of multi-source data:

[0052]

[0053] In the formula, K represents the conflict coefficient, defined as K = ∑ B∩C＝A m1(B)m2(C), w B , w C represents the weights of focal elements B and C, which are determined by the reliability of the evidence source; m1(B)m2(C) represents the basic probability assignment of the two evidences to focal elements B and C; m(A) represents the probability assignment of focal element A after fusion.

[0054]

[0055] In the formula, Belt(A) represents the belief degree (lower probability) of proposition A, Pl(A) represents the plausibility degree (upper probability) of proposition A, and m 融合 (B) represents the probability of focal element B after fusion. The final decision selects the focal element with the largest Belt(A) or the smallest Pl(A) - Belt(A); when the conflict coefficient K ≥ 0.5, trigger the sensor credibility re-evaluation mechanism;

[0056] Confidence dynamic correction, introduce a time decay factor to iteratively update the confidence:

[0057] m t (A) = α · w i · m current (A) + (1 - α) · λ τ · m t-τ (A)

[0058] where m t (A) represents the corrected confidence of proposition A at time t, α ∈ (0, 1) is the evidence weight coefficient, and w i is the static weight of the i-th sensor, λ is the historical data decay rate, and τ is the sliding time window length; m current (A) represents the original confidence of proposition A directly output by the sensor at the current moment; m t-τ (A) represents the historical corrected confidence at time t - τ, that is, the confidence at the earliest moment within the sliding window.

[0059] As Figures 2 - 5 shown, the present invention also discloses an electrochemical energy storage thermal runaway multi-dimensional monitoring device for dynamic feature fusion analysis, which is used for the above-mentioned electrochemical energy storage thermal runaway multi-dimensional monitoring method of dynamic feature fusion analysis. The device includes an energy storage module housing, an electrochemical energy storage element located inside the energy storage module housing, and a multi-dimensional detector installed on the surface of the energy storage module housing through threaded fasteners. The detection holes of the multi-dimensional detector penetrate the side wall of the energy storage module housing and extend to the internal cavity of the energy storage module housing to monitor internal parameters; the device also includes a voltage monitoring module and an expansion force monitoring module;

[0060] The multi-dimensional detector includes a multi-spectral scanner, a photoelectric smoke detector, and a distributed optical fiber temperature sensor;

[0061] The voltage monitoring module includes an electrode wire electrically connected to the electrode post of the energy storage cell. The electrode wire adopts a double-layer shielding structure, with a multi-strand silver-plated copper core conductor on the inner layer and a high-temperature resistant fluoroplastic insulation layer on the outer layer. The end of the electrode wire is connected to a voltage acquisition terminal to collect the voltage data of the electrochemical energy storage element in real time, capture voltage changes at a high sampling frequency, and quickly send a warning signal to the central processing unit;

[0062] The expansion force monitoring module includes a stress gauge array mounted on the inner surface of the energy storage module housing, which is connected to an expansion force sensor through an anti-electromagnetic interference signal cable. The expansion force sensor is embedded in the spacer film between adjacent electrochemical energy storage element monomers inside the energy storage module housing; the strain gauge array adopts a Wheatstone bridge structure and is covered with a polyimide protective film resistant to electrolyte corrosion;

[0063] The early thermal runaway characteristic parameters from the electrochemical energy storage battery are respectively identified through the multi-dimensional detector, the voltage monitoring module, and the expansion force monitoring module. The monitored early thermal runaway characteristic parameters are transmitted to the edge computing unit through the data bus for data preprocessing, and a standardized data stream is formed through the multi-source data fusion analysis engine configured by the central processing unit;

[0064] When the standardized data stream is transmitted to the central processing unit through the data bus, the central processing unit performs dynamic correlation analysis on the multi-source data, and based on the composite monitoring model that fuses multi-dimensional data weights, it analyzes and judges the probability of thermal runaway risk in real time; subsequently, the confidence iteration sub-module iteratively calculates the probability of thermal runaway risk.

[0065] The present invention has the following advantages:

[0066] Based on the multi-parameter data measured by the three monitoring modules of the multi-dimensional detector, voltage monitoring module, and expansion force monitoring module, data pre-analysis and processing are carried out. The wavelet noise reduction of the data is carried out using an algorithm, and at the same time, outliers are removed, and then the timing deviation caused by transmission delay is eliminated; after the central processing unit receives the pre-processed standardized data stream, it extracts the dynamic characteristics of the standardized data stream, obtains the temperature gradient characteristics, gas concentration change rate characteristics, voltage spectrum characteristics, and stress wave peak characteristics, and imports them into the dynamic data weight fusion model for multi-dimensional fusion analysis to predict the probability of thermal runaway risk; to ensure the credibility of the thermal runaway prediction result, the central processing unit uses the confidence model to perform iterative calculations to obtain the convergence value, and finally archives and saves the monitoring results. Compared with the prior art:

[0067] (1) A collaborative monitoring mechanism with multi-dimensional perception is constructed, which integrates parameters such as temperature field, mechanical deformation stress, voltage signal, characteristic gas, and smoke, breaks the limitation of the fragmentation of monitoring data by traditional sensors, accurately captures the early development trend of thermal runaway through spatio-temporal coupling modeling, and realizes the composite monitoring of expansion deformation and chemical reaction;

[0068] (2) Relying on the thermal runaway risk assessment model with confidence weighting, a full-link intelligent judgment system from "thermal runaway perception - correlation analysis - probability prediction - iterative verification" is constructed, which significantly improves the sensitivity and reliability of the recognition of thermal runaway in the very early stage, and effectively delays the problems of high false positives and serious missed reports;

[0069] (3) Modular functional design, using plug-and-play interfaces and redundant fault-tolerant structures, supports rapid deployment and flexible expansion, while ensuring the real-time monitoring, effectively reducing the full-life cycle operation and maintenance costs, and providing a reliable solution for the thermal safety monitoring of electrochemical energy storage systems;

[0070] Other features and advantages of the present invention will be described in the subsequent specification, and, in part, will be obvious from the specification, or will be understood by implementing the present invention. The objectives and other advantages of the present invention can be achieved and obtained through the structures specifically pointed out in the specification, claims, and drawings. Description of the Drawings

[0071] Figure 1 It is a flow chart of the electrochemical multi-dimensional monitoring method for dynamic feature fusion analysis provided by the present invention;

[0072] Figure 2 Schematic diagram of the position and structure of the multi-dimensional detector provided by the present invention;

[0073] Figure 3 Schematic diagram of the structure of the voltage monitoring module provided by the present invention;

[0074] Figure 4 Schematic diagram of the position and structure of the strain gauge array, signal cable and expansion force sensor provided by the present invention;

[0075] Figure 5 Schematic diagram of the structure of the electrochemical multi-dimensional monitoring device for dynamic feature fusion analysis provided by the present invention;

[0076] In the figure: 1. Electrochemical energy storage element; 2. Multi-dimensional detector; 3. Threaded fastener; 4. Detection hole; 5. Electrode wire; 6. Voltage detection module; 7. Stress gauge array; 8. Signal cable; 9. Expansion force sensor; 10. Data bus; 11. Edge computing unit; 12. Central processing unit. Detailed implementation manners

[0077] To make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be noted that, without conflict, the embodiments in this application and the features in the embodiments can be combined with each other arbitrarily.

[0078] The steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. And, although the logical order is shown in the flowchart ( Figure 1 ), in some cases, the steps shown or described can be executed in a different order than here.

[0079] The embodiment of the present invention provides an electrochemical energy storage thermal runaway multi-dimensional monitoring method for dynamic feature fusion analysis. The early thermal runaway characteristic parameters from the electrochemical energy storage battery are respectively identified by a multi-dimensional detector, a voltage monitoring module, and an expansion force monitoring module. The monitored early thermal runaway characteristic parameters are transmitted to the edge computing unit through the data bus for data preprocessing, and a multi-source data fusion analysis engine configured by the central processing unit is used to form a standardized data stream;

[0080] When the standardized data stream is transmitted to the central processor through the data bus, the central processor performs dynamic correlation analysis on the multi-source data, and based on the composite monitoring model of multi-dimensional data weight fusion, analyzes and judges the thermal runaway risk probability in real time; subsequently, the confidence iteration sub-module iteratively calculates the thermal runaway risk probability;

[0081] In an exemplary instance, the steps for obtaining the early thermal runaway characteristic parameters of the electrochemical energy storage battery are as follows: The multi-dimensional detector extracts the gas inside the energy storage module housing at a constant flow rate through its own detection holes. The voltage monitoring module captures the voltage fluctuation signal of the energy storage elements inside the energy storage module housing through the electrode wires. The stress gauge array of the expansion force monitoring module measures the surface strain distribution of the energy storage module housing.

[0082] In an exemplary instance, aiming at the problems that the traditional static threshold criterion has a response lag and outlier interference in multiple application scenarios and cannot flexibly cope with the dynamically changing environment and data, a dynamic monitoring method of multi-dimensional fusion is proposed; specifically: The gas inside the energy storage module housing is extracted at a constant flow rate through the detection holes of the multi-dimensional detector, and the following are executed synchronously:

[0083] The distributed optical fiber temperature sensor array monitors the temperature distribution of 8×8 grid points inside the energy storage module housing with a sampling period of 0.5 s;

[0084] The multi-spectral scanner enhances the ppm-level monitoring of characteristic gases such as CO and H2 through the Pt-Pd catalytic coating.

[0085] In an exemplary instance, the early thermal runaway characteristic parameters detected are transmitted to the edge computing unit through the data bus for data preprocessing, and the following are executed in sequence:

[0086] Wavelet denoising: The signal is decomposed into different frequency sub-bands by wavelet transform, and the coefficients of the high-frequency sub-bands are threshold processed to remove noise; the soft threshold function used is:

[0087]

[0088] Among them, is the processed wavelet coefficient, ω j,k is the original wavelet coefficient, sgn is the sign function, and λ is the threshold;

[0089] (x) + = max(x, 0);

[0090] Space-time registration: For time series data, the time of the data collected by different sensors needs to be synchronized; the time interpolation method, such as linear interpolation, can be used; for the data of sensors at different positions in space, a spatial coordinate mapping relationship can be established based on the structure of the energy storage cabin.

[0091] Linear interpolation formula: Given time points t1, t2 and corresponding data y1, y2, for time point t, the interpolated data is:

[0092]

[0093] Outlier rejection: The 3σ criterion based on statistics is adopted;

[0094] Assume the data follows a normal distribution, and the data point x i , if then determine that x i is an outlier and eliminate it; where is the data mean, and σ is the standard deviation;

[0095]

[0096] For dimension normalization, use the maximum - minimum normalization method, and the formula is:

[0097]

[0098] where x is the original data, x min and x max are the minimum and maximum values of the data, and x norm is the normalized data.

[0099] In an exemplary instance, the multi - source data fusion analysis engine includes:

[0100] A dynamic feature extraction sub - module, which is used to extract and analyze multi - dimensional relevant data;

[0101] A dynamic data weight fusion model, which generates a thermal runaway risk probability using the weight relationship of the standardized data stream;

[0102] A confidence iteration sub - module, which integrates the XGBoost classifier to perform confidence iteration on the data to improve data reliability.

[0103] In an exemplary instance, the dynamic feature extraction sub - module extracts and analyzes multi - dimensional relevant data through a composite monitoring model of multi - dimensional data weight fusion, including:

[0104] Feature extraction. For the temperature data T(x, y, z, t) collected by the distributed fiber optic temperature sensor, the temperature gradient is:

[0105]

[0106] where x, y, z represent the three coordinate axes in the space rectangular coordinate system, representing the position (the unit is usually meters, m), and t is the time variable (the unit is usually seconds, s). The temperature field may change with time, but when calculating the gradient, it is only for the spatial coordinates, represents the direction and magnitude of the maximum change rate of the temperature field in space, represents the change rate of temperature in the x - direction (unit: °C / m or K / m), represents the change rate of temperature in the y - direction, represents the change rate of temperature in the z - direction;

[0107] In the discrete case, a difference approximation is adopted, and the formula is as follows:

[0108]

[0109] where Δx represents the spatial step (unit: m), which is the distance between adjacent grid points;

[0110] Let the sequence of gas concentration varying with time be C(t), and the gas concentration change rate r is as follows:

[0111]

[0112] where Δt is the time interval;

[0113] Perform a fast Fourier transform FFT on the voltage data V(t) obtained by the voltage acquisition terminal:

[0114]

[0115] where, V(t): the voltage signal in the time domain, representing the continuous voltage value varying with time t (unit: volt, V); V(f): the voltage spectrum in the frequency domain, representing the complex value corresponding to the frequency variable f (unit: V·s or V / Hz); f: the frequency variable, with the unit of hertz (Hz); j: the imaginary unit, satisfying that the square of j is -1 (j is commonly used in engineering, and i is commonly used in mathematics); e -j2πft : the complex exponential function, describing the sine wave basis function with the frequency variable f;

[0116] Thus, the voltage spectrum V(f) is obtained, and the peak value and frequency distribution characteristics in the spectrum are analyzed;

[0117] For the stress data collected by the expansion force detection module, by setting the threshold S th and the window length N, within the data window, if and S(i)>S th , then S(i) is the stress wave peak;

[0118] where, i is the distance from the stress gauge, and S is the normal stress;

[0119] Let the smoke concentration data M(x, y, z, t) collected by the photoelectric smoke detector, similar to the temperature gradient calculation, the smoke concentration gradient:

[0120]

[0121] where represents the direction and magnitude of the maximum change rate of the smoke concentration in space; represents the change rate of the temperature in the x direction (unit: °C / m or K / m); Represents the rate of change of temperature in the y direction; Represents the rate of change of temperature in the z direction;

[0122] Realize the prediction of the probability of thermal runaway risk through the dynamic data weight fusion model: input the multi-source feature vectors into the XGBoost classifier, generate the probability of thermal runaway risk by using the weight relationship of each standardized data stream, Failure Mode and Effects Analysis FMEA: Determine the severity S, occurrence O, and detection D of each parameter according to the historical failure data, calculate the Risk Priority Number RPN = S × O × D as the weight basis, and the risk probability calculation model is as follows:

[0123]

[0124] Where wi is the dynamic weight of the i-th parameter, and f(xi) is the risk contribution function after parameter normalization (such as Sigmoid, piecewise linear function);

[0125] In an exemplary example, the confidence is calculated iteratively through the XGBoost classifier in the central processing unit, and the iterative optimization method includes the following steps:

[0126] Multi-source evidence fusion, adopt the improved D-S evidence theory to fuse the confidence of multi-sensor data to overcome the conflict of multi-source data:

[0127]

[0128] Where K represents the conflict coefficient, defined as K = Σ B∩C＝A m1(B)m2(C), w B , w C Represents the weights of the focal elements B and C, which are determined by the reliability of the evidence sources; m1(B)m2(C) represents the basic probability assignments of the two evidences to the focal elements B and C; m(A) represents the probability assignment of the focal element A after fusion.

[0129]

[0130] Where Belt(A) represents the degree of belief (lower probability) in proposition A, Pl(A) represents the plausibility (upper probability) in proposition A, and m 融合 (B) represents the probability of the focal element B after fusion, and the final decision selects the focal element with the largest Belt(A) or the smallest Pl(A)-Belt(A); when the conflict coefficient K ≥ 0.5, trigger the sensor credibility re-evaluation mechanism;

[0131] Confidence dynamic correction, introduce the time decay factor to iteratively update the confidence:

[0132] m t (A) = α·w i ·mcurrent (A) + (1 - α)·λ τ ·m t-τ (A)

[0133] In the formula, m t (A) represents the corrected confidence in proposition A at time t, α ∈ (0, 1) is the evidence weight coefficient, w i is the static weight of the i-th sensor, λ is the historical data decay rate, τ is the sliding time window length; m current (A) represents the original confidence in proposition A directly output by the sensor at the current moment; m t-τ (A) represents the historical corrected confidence at time t - τ, that is, the confidence at the earliest moment within the sliding window.

[0134] An embodiment of the present invention also discloses an electrochemical energy storage thermal runaway multi-dimensional monitoring device for dynamic feature fusion analysis, which is used for the above-mentioned electrochemical energy storage thermal runaway multi-dimensional monitoring method for dynamic feature fusion analysis. The device includes an energy storage module housing, an electrochemical energy storage element 1 located inside the energy storage module housing, a multi-dimensional detector 2 installed on the surface of the energy storage module housing through a threaded fastener 3, and the detection holes 4 of the multi-dimensional detector 2 penetrate through the side wall of the energy storage module housing and extend to the internal cavity of the energy storage module housing to monitor internal parameters; the device also includes a voltage monitoring module and an expansion force monitoring module;

[0135] The multi-dimensional detector 2 includes a multi-spectral scanner, a photoelectric smoke detector, and a distributed optical fiber temperature sensor;

[0136] The voltage monitoring module includes an electrode wire 5 electrically connected to the electrode post of the energy storage cell. The electrode wire 5 adopts a double-layer shielding structure, with a multi-strand silver-plated copper core conductor on the inner layer and a high-temperature resistant fluoroplastic insulating layer on the outer layer. The end of the electrode wire 5 is connected to a voltage sampling terminal to collect the voltage data of the electrochemical energy storage element 1 in real time and quickly send a warning signal to the central processor 12;

[0137] The expansion force monitoring module includes a stress gauge array 7 mounted on the inner surface of the energy storage module housing, which is connected to an expansion force sensor 9 through an anti-electromagnetic interference signal cable 8. The expansion force sensor 9 is embedded in the spacer film between adjacent electrochemical energy storage element 1 monomers inside the energy storage module housing; the strain gauge array 7 adopts a Wheatstone bridge structure and is covered with a polyimide protective film resistant to electrolyte corrosion;

[0138] The early thermal runaway characteristic parameters from the electrochemical energy storage battery are respectively identified by the multi-dimensional detector 2, the voltage monitoring module, and the expansion force monitoring module, and the monitored early thermal runaway characteristic parameters are transmitted through the data bus 10 to the edge computing unit 11 for data preprocessing, and a standardized data stream is formed by the multi-source data fusion analysis engine configured by the central processing unit 12;

[0139] When the standardized data stream is transmitted to the central processor 12 through the data bus 11, the central processor 12 performs dynamic correlation analysis on the multi-source data, and based on the composite monitoring model of multi-dimensional data weight fusion, analyzes and judges the thermal runaway risk probability in real time; subsequently, the confidence iteration sub-module iteratively calculates the thermal runaway risk probability.

[0140] Although the present invention has been described in detail with general descriptions and specific embodiments above, modifications or improvements can be made to it on the basis of the present invention, which are obvious to those skilled in the art. Therefore, these modifications or improvements made without departing from the spirit of the present invention all fall within the scope of protection required by the present invention.

Claims

1. An electrochemical energy storage thermal runaway multi-dimensional monitoring method based on dynamic feature fusion analysis, characterized in that: The early thermal runaway characteristic parameters of the electrochemical energy storage battery are respectively identified by a multi-dimensional detector, a voltage monitoring module, and an expansion force monitoring module. The monitored early thermal runaway characteristic parameters are transmitted to the edge computing unit through a data bus for data preprocessing, and a multi-source data fusion analysis engine configured by the central processing unit is used to form a standardized data stream; When the standardized data stream is transmitted to the central processor through the data bus, the central processor performs dynamic correlation analysis on the multi-source data, and based on the composite monitoring model of multi-dimensional data weight fusion, analyzes and judges the thermal runaway risk probability in real time; Subsequently, the confidence iteration sub-module iteratively calculates the thermal runaway risk probability.

2. The multi-dimensional monitoring method for thermal runaway of electrochemical energy storage based on dynamic feature fusion analysis according to claim 1, wherein: The steps for obtaining the early thermal runaway characteristic parameters of the electrochemical energy storage battery are as follows: The multi-dimensional detector extracts the gas inside the energy storage module housing at a constant flow rate through its own detection holes, the voltage monitoring module captures the voltage fluctuation signal of the energy storage element inside the energy storage module housing through the electrode wire, and the stress gauge array of the expansion force monitoring module measures the surface strain distribution of the energy storage module housing.

3. The multi-dimensional monitoring method for thermal runaway of electrochemical energy storage based on dynamic feature fusion analysis according to claim 2, characterized in that: Extract the gas inside the energy storage module housing at a constant flow rate through the detection holes of the multi-dimensional detector, and synchronously execute: (3.1) The temperature sensor array monitors the temperature distribution of 8×8 grid points inside the energy storage module housing with a sampling period of 0.5 s; (3.2) The multi-spectral scanner enhances the ppm-level monitoring of CO and H2 characteristic gases through a Pt-Pd catalytic coating.

4. The multi-dimensional monitoring method for thermal runaway of electrochemical energy storage based on dynamic feature fusion analysis according to claim 1, characterized in that: The multi-source data fusion analysis engine includes: (4.1) A dynamic feature extraction sub-module for extracting and analyzing multi-dimensional relevant data; (4.2) A dynamic data weight fusion model that generates a thermal runaway risk probability using the weight relationship of the standardized data stream; (4.3) A confidence iteration sub-module that integrates an XGBoost classifier to perform confidence iteration on the data to improve data reliability.

5. The multi-dimensional monitoring method for thermal runaway of electrochemical energy storage based on dynamic feature fusion analysis according to claim 4, characterized in that: The dynamic feature extraction sub-module described in step (4.1) extracts and analyzes multi-dimensional relevant data through a composite monitoring model of multi-dimensional data weight fusion, including: Feature extraction. For the temperature data T(x, y, z, t) collected by the temperature sensor, the temperature gradient is: where x, y, and z represent the three coordinate axes in a rectangular spatial coordinate system, representing position, and t is the time variable; represents the direction and magnitude of the maximum rate of change of the temperature field in space, represents the rate of change of temperature in the x direction, represents the rate of change of temperature in the y direction, represents the rate of change of temperature in the z direction; In the discrete case, a differential approximation is adopted, and the formula is as follows: In the formula, Δx represents the spatial step size, which is the distance between adjacent grid points; Let the sequence of gas concentration changing with time be C(t), and the gas concentration change rate r is as follows: where Δt is the time interval; Perform a fast Fourier transform FFT on the voltage data V(t) obtained by the voltage acquisition terminal: Where, V(t): the voltage signal in the time domain, representing the continuous voltage value that varies with time t; V(f): the voltage spectrum in the frequency domain, representing the complex value corresponding to the frequency variable f; f: the frequency variable; j: the imaginary unit, satisfying that the square of j is -1; e -j2πft : the complex exponential function, describing the sine wave basis function with the frequency variable f; Thus, the voltage spectrum V(f) is obtained, and the peak value and frequency distribution characteristics in the spectrum are analyzed; For the stress data collected by the expansion force detection module, by setting the threshold S th and the window length N, within the data window, if and S(i) > S th , then S(i) is the stress wave peak; In the formula, i is the distance from the stress gauge, and S is the normal stress; Let the smoke concentration data M(x, y, z, t) collected by the photoelectric smoke detector, similar to the temperature gradient calculation, the smoke concentration gradient: wherein represents the direction and magnitude of the maximum rate of change of the smoke concentration in space; represents the rate of change of the temperature in the x direction; represents the rate of change of the temperature in the y direction; represents the rate of change of the temperature in the z direction; Step (4.2) realizes the prediction of the thermal runaway risk probability through a dynamic data weight fusion model: Input the multi-source feature vectors into the XGBoost classifier, and generate the thermal runaway risk probability by using the weight relationship of each standardized data stream. Failure Mode and Effects Analysis (FMEA): Determine the severity S, occurrence O, and detection D of each parameter according to historical failure data, and calculate the Risk Priority Number RPN = S × O × D as the weight basis. The risk probability calculation model is as follows: Where wi is the dynamic weight of the i-th parameter, and f(xi) is the risk contribution function after parameter normalization.

6. The multi-dimensional monitoring method for thermal runaway of electrochemical energy storage based on dynamic feature fusion analysis according to claim 4, characterized in that: Step (4.3) performs confidence iteration calculation through the XGBoost classifier in the central processing unit to calculate the confidence. The iterative optimization method includes the following steps: Multi-source evidence fusion, to overcome the conflict of multi-source data, an improved D-S evidence theory is used to fuse the confidence of multi-sensor data: where K represents the conflict coefficient, defined as K = ∑ B∩C＝A m1(B)m2(C), w B , w C represents the weights of focal elements B and C, determined by the reliability of the evidence sources; m1(B)m2(C) represents the basic probability assignments of the two pieces of evidence to focal elements B and C; m(A) represents the probability assignment of focal element A after fusion. Wherein, Belt(A) represents the degree of belief in proposition A, Pl(A) represents the likelihood of proposition A, and m 融合 (B) represents the probability of the focal element B after fusion, and the final decision selects the focal element with the largest Belt(A) or the smallest Pl(A)-Belt(A); when the conflict coefficient K≥0.5, the sensor credibility re-evaluation mechanism is triggered; Confidence dynamic correction, introducing a time decay factor to iteratively update the confidence: m t (A) = α·w i ·m current (A) + (1 - α)·λ τ ·m t-τ (A) where m t (A) represents the revised confidence of proposition A at time t, α ∈ (0, 1) is the evidence weight coefficient, w i is the static weight of the i-th sensor, λ is the historical data decay rate, and τ is the sliding time window length; m current (A) represents the original confidence of proposition A directly output by the sensor at the current moment; m t-τ (A) represents the historical revised confidence at time t - τ, that is, the confidence at the earliest moment within the sliding window.

7. An electrochemical energy storage thermal runaway multi-dimensional monitoring device for dynamic feature fusion analysis, which is used for the dynamic feature fusion analysis electrochemical energy storage thermal runaway multi-dimensional monitoring method according to any one of claims 1-6, and is characterized in that: The device includes a energy storage module housing, an electrochemical energy storage element (1) located inside the energy storage module housing, and a multi-dimensional detector (2) installed on the surface of the energy storage module housing through threaded fasteners (3). The detection holes (4) of the multi-dimensional detector (2) penetrate the side wall of the energy storage module housing and extend to the internal cavity of the energy storage module housing to monitor internal parameters; the device also includes a voltage monitoring module and an expansion force monitoring module; The multi-dimensional detector (2) includes a multi-spectral scanner, a photoelectric smoke detector, and a temperature sensor; The voltage monitoring module includes an electrode wire (5) electrically connected to the electrode post of the energy storage cell. The end of the electrode wire (5) is connected to a voltage acquisition terminal to collect the voltage data of the electrochemical energy storage element (1) in real time and send a warning signal to the central processing unit (12); The expansion force monitoring module includes a stress gauge array (7) mounted on the inner surface of the energy storage module housing, which is connected to an expansion force sensor (9) through an anti-electromagnetic interference signal cable (8). The expansion force sensor (9) is embedded in the spacer film between adjacent monomers of the electrochemical energy storage element (1) inside the energy storage module housing; The early thermal runaway characteristic parameters from the electrochemical energy storage battery are respectively identified by the multi-dimensional detector (2), the voltage monitoring module, and the expansion force monitoring module. The monitored early thermal runaway characteristic parameters are transmitted to the edge computing unit (11) through the data bus (10) for data preprocessing, and a standardized data stream is formed by the multi-source data fusion analysis engine configured by the central processing unit (12); When the standardized data stream is transmitted to the central processing unit (12) through the data bus (11), the central processing unit (12) performs dynamic correlation analysis on the multi-source data, and based on the composite monitoring model of multi-dimensional data weight fusion, analyzes and judges the thermal runaway risk probability in real time; subsequently, the confidence iteration sub-module iteratively calculates the thermal runaway risk probability.

Images