A method and device for identifying and tracing the failure evolution stage of a sodium-ion battery energy storage system

Abstract

The application provides a kind of sodium ion battery energy storage system fault evolution stage identification and tracing method and device.The method comprises: collecting multi-source monitoring data;Carrying out electrical-thermal coupling feature extraction, temperature change rate is calculated by coupling propagation operator containing Joule heat heat generation term and voltage fluctuation heat effect term;Four-stage fault evolution hidden Markov model is constructed, state transition matrix adopts upper triangular constraint structure to reflect irreversible characteristics;Unsupervised parameter estimation is carried out using Baum-Welch algorithm with physical constraint regularization, and the regularization term ensures that the parameters are in the physical feasible region;Adaptive multi-scale backtracking algorithm decoding is carried out using evolution stage confidence driving, suspicious transition points are marked by confidence evaluation and consistency voting is performed on multiple scales;Calculate fault evolution time window output trace report.The application realizes accurate identification and tracing of fault evolution stage through physical constraint regularization and confidence driven backtracking mechanism.

Description

Technical Field

[0001] This invention relates to the field of battery fault assessment technology, and in particular to a method and apparatus for identifying and tracing the fault evolution stages of a sodium-ion battery energy storage system. Background Technology

[0002] Sodium-ion battery energy storage systems are key equipment for new energy distribution and storage, and their safe and stable operation directly affects the reliability of the power grid. Energy storage system faults exhibit significant evolutionary characteristics: faults do not occur instantaneously, but rather evolve gradually from a normal state through a series of transitional stages to a fault state. Taking an overcharge-induced fault as an example, its typical evolution process is: from a normal charging state to a transitional state I with a slow voltage rise, then to a transitional state II with a rapid voltage rise accompanied by a temperature increase, and finally to a fault state where the voltage exceeds the limit, triggering protection shutdown. This evolutionary process typically completes within minutes to hours. If the evolutionary stage of the fault can be accurately identified and its evolutionary path traced back, targeted preventative measures can be taken before the fault fully develops.

[0003] In existing research, Hidden Markov Models (HMMs) have been applied to fault diagnosis in battery and fuel cell systems. For example, existing literature discloses a fault diagnosis strategy based on K-means clustering using discrete HMMs, employing K-means clustering for sample filtering and the Baum-Welch algorithm for model training to detect and isolate six fault states in fuel cells. Other literature discloses a fuel cell fault prediction method based on Hidden Semi-Markov Models (HSMs) and empirical models, using the Welch-Baum algorithm to update model parameters and estimating the health state through a forward algorithm. Furthermore, an improved long short-term memory (LSTM) network combined with an HMM framework has also been used for early warning of battery thermal runaway. Chinese patent CN107169243B discloses a fuel cell fault prediction method based on HSMs and empirical models, which uses the Welch-Baum algorithm to update model parameters and estimates the health state through a forward algorithm. Although this method involves parameter estimation and health status assessment, the parameter estimation process of its Baum-Welch algorithm does not incorporate the physical constraints of the modeled object. The estimation results may fall within the physically infeasible region, leading to a disconnect between the model parameters and the actual battery characteristics. Furthermore, it uses a single decoding to determine the optimal state sequence without assessing the reliability of the decoding results at each time step. In summary, existing research has several significant shortcomings in identifying the fault evolution stages of sodium-ion battery energy storage systems. Specifically: First, existing technologies use independent health level labels for hidden states and employ a fully connected structure in the state transition matrix to allow bidirectional transitions. This fails to incorporate the irreversible nature of fault evolution into the model structure, making it impossible to capture the physical meaning and temporal information of the transition stages from normal to fault states. Second, the parameter estimation process of the existing Baum-Welch algorithm is purely data-driven, without incorporating battery physical constraints, and the estimation results may violate fundamental thermodynamic laws. Third, existing technologies use a single decoding to determine the optimal state sequence, without quantitatively evaluating the reliability of the decoding results at each time point. This easily leads to misjudgments at transition state boundaries and lacks automatic correction capabilities. Fourth, existing technologies do not establish a coupled propagation model between electrical and thermal anomalies, treating voltage, current, and temperature as independent variables, failing to describe the physical laws governing the propagation of electrical anomalies to thermal anomalies via Joule heating and polarization heating paths. Therefore, a method for fault evolution stage identification and tracing is urgently needed. Summary of the Invention

[0004] The purpose of this invention is to provide a method and apparatus for identifying and tracing the fault evolution stage of a sodium-ion battery energy storage system, in order to solve the problems in the prior art where the hidden state of the Hidden Markov Model does not reflect the irreversible evolution characteristics, the Baum-Welch parameter estimation lacks physical constraints, the reliability of the Viterbi single decoding is not evaluated, and the electrical-thermal coupling propagation is not modeled.

[0005] The present invention specifically provides the following technical solution: A method for identifying and tracing the fault evolution stages of a sodium-ion battery energy storage system, the method comprising the following steps: S1. Collect multi-source monitoring data of the sodium-ion battery energy storage system, wherein the multi-source monitoring data includes battery cell voltage sequence, charge / discharge current sequence and battery surface temperature sequence; S2. Perform electrical-thermal coupling feature extraction on the multi-source monitoring data to obtain four coupling features: voltage change rate, current integral value, temperature change rate, and temperature gradient. The temperature change rate is calculated using an electrical-thermal bidirectional coupling propagation operator, which includes a Joule heating term and a voltage fluctuation thermal effect term, respectively describing two different physical mechanisms of electrical anomalies propagating to thermal anomalies. Encode the four coupling features into an observation sequence. S3. Construct a four-stage fault evolution Hidden Markov Model, where the hidden state set of the model is {S0, S1, S2, S3}, corresponding to the normal state, transition state I, transition state II and fault state respectively. The state transition probability matrix of the model adopts an upper triangular constraint structure, which only allows transitions from a lower-index state to a higher-index state or staying at the current state. The upper triangular constraint structure reflects the irreversible characteristic of fault evolution; S4. Perform unsupervised estimation on the parameters of the four-stage fault evolution Hidden Markov Model by using the Baum-Welch algorithm with physical constraint regularization. The physical constraint regularization is realized by introducing a regularization term based on battery physical laws in the maximization step of expectation maximization, and the regularization term ensures that both the Joule heat generation coefficient and the voltage fluctuation thermal effect coefficient in the temperature change rate coupling operator are within the physically feasible region; S5. Decode the observed sequence by using an adaptive multi-scale backtracking algorithm driven by evolution stage confidence. The decoding process first backtracks the optimal hidden state sequence through the standard Viterbi algorithm, then calculates the evolution stage confidence at each moment, where the evolution stage confidence is defined as the ratio of the maximum posterior probability to the second maximum posterior probability at that moment, then marks the moments with confidence lower than the preset threshold as suspicious transition points, and finally performs Viterbi backtracking on the suspicious transition points at the original time scale, the 2x downsampling scale and the 5x downsampling scale respectively, and determines the final state through consistency voting to obtain the identification result of the fault evolution stage at each moment; S6. Calculate the fault evolution time window and output a fault evolution traceability report according to the moment when the state first transitions from the normal state to the transition state and the moment when the state first reaches the fault state in the fault evolution stage identification result.

[0006] Optionally, in S2, the expression of the electrical-thermal bidirectional coupling propagation operator is: wherein, is the Joule heat generation term, is the voltage fluctuation thermal effect term, is the Joule heat generation coefficient with the unit of °C / W, is the voltage fluctuation thermal effect coefficient with the unit of °C·s / V, I is the charge-discharge current, R int is the internal resistance of the battery, is the voltage change rate.

[0007] Optionally, in S3, the state transition probability matrix A=[a ij 4×4 satisfies: a ij =0 when j<i, that is, reverse transitions from a state with a larger index to a state with a smaller index are prohibited; the sum of probabilities in each row is 1; a​33 =1, meaning that fault state S3 is an absorption state.

[0008] Optionally, in S4, the regularization term Rphys in the maximization step of the Baum-Welch algorithm with physical constraint regularization is specifically: Where λ1, λ2, λ3, and λ4 are regularization intensity coefficients, R max and R min These represent the upper and lower physical limits of the battery's internal resistance.

[0009] Optionally, in S5, the formula for calculating the confidence level of the evolution stage is C. t =P max / P second , where P max =max i P t post i P represents the maximum posterior probability at that moment. second The second largest posterior probability, the posterior probability P t pos ti =P t fwd i ·P t bwd i / Σ j P t fwd j ·P t bwd j , where P t fwd Let P be the forward probability. t bwd This represents the backward probability.

[0010] Optionally, in S5, the specific process of determining the final state by multi-scale consensus voting on suspicious transition points is as follows: For time periods marked as suspicious, Viterbi backtracking is performed on the original time scale, the second downsampling scale, and the fifth downsampling scale to obtain three state subsequences. At each time point, the frequency of each state in the three subsequences is counted, and the state with the highest frequency is taken as the final identification result for that time point. If the three subsequences are completely consistent, the confidence level of that time point is marked as the highest level.

[0011] Optionally, in S4, before the Baum-Welch algorithm iteration, K-means clustering based on electrical-thermal coupling characteristics is used to initialize the model parameters. The number of clusters K=4 corresponds to the four evolution stages from S0 to S3. The cluster center is used as the initial mean vector of the Gaussian distribution of each state, the sample covariance within the cluster is used as the initial covariance matrix, and the sample proportion of each cluster is used as the initial state distribution. The feature vector used for clustering only includes the temperature change rate, voltage change rate, and temperature gradient calculated by the coupling propagation operator. The current integral value is not included in the clustering to avoid interference from the charging and discharging conditions on the initialization.

[0012] Optionally, in S6, the fault evolution time window Δt is used to determine the emergency level: when Δt is less than 10 minutes, it is emergency level I, and the machine must be shut down immediately; when Δt is greater than or equal to 10 minutes and less than 30 minutes, it is emergency level II, and maintenance must be carried out within 30 minutes; when Δt is greater than or equal to 30 minutes, it is emergency level III, and the machine is included in routine inspection.

[0013] The present invention also provides a device for identifying and tracing the fault evolution stage of a sodium-ion battery energy storage system. The system includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the method described.

[0014] This invention offers the following beneficial technical effects: It provides a method for identifying and tracing the fault evolution stages of a sodium-ion battery energy storage system. By designing a state transition matrix with upper triangular constraints, this invention enables the hidden states of the Hidden Markov Model to reflect the irreversible physical characteristics of fault evolution, unlike the health level label design with bidirectional state transitions in existing technologies. Simultaneously, through the Baum-Welch algorithm with physical constraint regularization, it incorporates battery physical constraints into the parameter estimation process, ensuring that the estimation results always remain within the physically feasible region. Furthermore, this invention employs an adaptive multi-scale backtracking algorithm driven by the confidence level of the evolution stages to quantify the reliability of the decoding results at each moment and automatically perform multi-scale consistency correction for low-confidence periods, solving the problem of easy misjudgment at transition boundaries in single Viterbi decoding in existing technologies. A mathematical model describing the coupling propagation law between electrical and thermal anomalies is established using an electrical-thermal bidirectional coupling propagation operator, fully considering both Joule heating and polarization heating physical mechanisms. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 A flowchart of a method for identifying and tracing the fault evolution stages of a sodium-ion battery energy storage system, provided in an embodiment of the present invention.

[0017] Figure 2 A schematic diagram of the state transition and observation distribution of the four-stage fault evolution HMM provided in this embodiment of the invention.

[0018] Figure 3 This is a schematic diagram of the response characteristics of the electrical-thermal bidirectional coupling propagation operator provided in an embodiment of the present invention.

[0019] Figure 4 This is a schematic diagram of Baum-Welch parameter estimation convergence and Viterbi path backtracking provided in an embodiment of the present invention.

[0020] Figure 5 The bar chart and box plot provided for performance comparison in the embodiments of the present invention.

[0021] Figure 6 The fault evolution stage identification distribution map and confusion matrix are provided for embodiments of the present invention. Detailed Implementation

[0022] The embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0023] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. This application can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0024] It should be noted that various aspects of embodiments within the scope of the appended claims are described below. It will be apparent that the aspects described herein can be embodied in a wide variety of forms, and the structure and / or function of any markings described herein are merely illustrative. Based on this application, those skilled in the art will understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number and aspects set forth herein can be used to implement the apparatus and / or practical methods.

[0025] It should also be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this application.

[0026] Additionally, specific details are provided in the following description to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that practice can be carried out without these marked details.

[0027] The purpose of this invention is to provide a method for identifying and tracing the fault evolution stages of a sodium-ion battery energy storage system, so as to achieve accurate identification and tracing of the fault evolution stages.

[0028] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0029] Example 1 Using a 50MW / 200MWh sodium-ion battery energy storage power station as the experimental scenario, square sodium-ion battery cells with layered oxide cathodes and hard carbon anodes were employed, with a rated capacity of 200Ah and a nominal voltage of 3.0V. The multi-source monitoring data was collected through a BMS system and a temperature sensor network, including: 11,520 channels covering 480 battery clusters for battery cell voltage sequence sampling at a frequency of 1Hz with an accuracy of ±5mV; 480 cluster-level channels for charge / discharge current sequence sampling at a frequency of 1Hz with an accuracy of ±0.5A; and 120 temperature sensors evenly distributed per cluster for battery surface temperature sequence sampling at a frequency of 0.1Hz with an accuracy of ±0.5°C. This embodiment collected data for 72 consecutive hours of operation, obtaining a total of 259,200 voltage and current time points and 25,920 temperature time points.

[0030] like Figure 1 This invention illustrates a method for identifying and tracing the fault evolution stages of a sodium-ion battery energy storage system. The method includes the following steps: S1. Collect multi-source monitoring data of the sodium-ion battery energy storage system, including battery cell voltage sequence, charge / discharge current sequence and battery surface temperature sequence.

[0031] S2. Perform electrical-thermal coupling feature extraction on the multi-source monitoring data to obtain four coupling features: voltage change rate, current integral value, temperature change rate, and temperature gradient. The temperature change rate is calculated using an electrical-thermal bidirectional coupling propagation operator, which includes a Joule heating term and a voltage fluctuation thermal effect term, respectively describing two different physical mechanisms of electrical anomalies propagating to thermal anomalies. Encode the four coupling features into an observation sequence.

[0032] In S2, the expression for the electrical-thermal bidirectional coupling propagation operator is: in, For Joule heating, For the voltage fluctuation heat effect term, The Joule heating coefficient is expressed in °C / W. The voltage fluctuation thermal effect coefficient is expressed in °C·s / V, where I is the charging and discharging current, and R is the voltage fluctuation thermal effect coefficient. int This refers to the battery's internal resistance. This represents the rate of change of voltage.

[0033] For the collected multi-source monitoring data, the four coupling characteristics are calculated according to the following formula: Voltage change rate dV / dt| t =V t -V t-1 / Δt reflects the voltage fluctuation rate. In this embodiment, Δt=1s is the voltage sampling interval. For battery cluster No. 42 at t=1800s, V 1800 =3.152V, V 1799 =3.148V, then dV / dt| 1800 =3.152-3.148 / 1=0.004V / s.

[0034] Current integral value Q int | t =Σ τ=t-W+1 t I τ Δt reflects the recent cumulative charge and discharge amount. In this embodiment, the window length W = 60 sampling points corresponding to 60 seconds, and under constant current charging conditions I = 50A, then Q int | t =50×60 / 3600=0.833 Ah.

[0035] Rate of temperature change dT / dt| t =α·I t 2 ·R int +β·dV / dt| t Calculated through an electro-thermal bidirectional coupling propagation operator. For example... Figure 3 The response characteristics of the electrical-thermal bidirectional coupling propagation operator are shown, where Figure 3 In the diagram, (a) represents the time-domain response of the coupling operator under a short-circuit scenario. Figure 3 In (b), the electro-thermal coupling response surface dT / dt=f(I, dV / dt) is defined. In this embodiment, the battery internal resistance R... int =0.008Ω, measured by HPPC, α=0.008°C / W, β=3.5°C·s / V. When I=50A, the Joule heating term α·I... 2 ·R int=0.008×2500×0.008=0.16°C / s; Voltage fluctuation heat effect term β·dV / dt=3.5×0.004=0.014°C / s; Total temperature change rate dT / dt=0.174°C / s. If the temperature difference method is used directly without the coupling operator, when T 1800 =28.5°C, T 1799 At 28.3°C, dT / dt = 0.2°C / s, which deviates from the result of the coupling operator by 15%, proving that the coupling operator can effectively smooth measurement noise.

[0036] Temperature gradient G T | t =max i,j∈模组 |T t i -T t j This reflects the degree of temperature unevenness within the module. In this embodiment, the readings of the eight temperature measuring points within the module at t=1800s are 28.5°C, 28.7°C, 29.2°C, 31.5°C, 28.4°C, 28.6°C, 29.0°C, and 28.8°C, respectively. Therefore, G... T =31.5-28.4=3.1°C, exceeding the 3°C threshold, triggering a local thermal anomaly warning.

[0037] The above four features are Z-score standardized and then encoded into a four-dimensional observation vector. t =[dV / dt-μ1 / σ1,Q int -μ2 / σ2,dT / dt-μ3 / σ3, G T -μ4 / σ4] T The observation sequence is composed of O={o1,o2,...,o...} T In this embodiment, T=3600 time steps correspond to 1 hour of monitoring data, with mean vector μ=[0.002, 0.833, 0.05, 1.2] and standard deviation vector σ=[0.008, 0.278, 0.12, 0.8].

[0038] S3. Construct a four-stage fault evolution hidden Markov model. The hidden state set of the model is {S0, S1, S2, S3}, which correspond to the normal state, transition state I, transition state II and fault state, respectively. The state transition probability matrix of the model adopts an upper triangular constraint structure, which only allows transitions from lower-order states to higher-order states or maintaining the current state. The upper triangular constraint structure reflects the irreversible nature of fault evolution.

[0039] In S3, the state transition probability matrix A = [a ij ] 4×4 Satisfy: aij =0 When j<i, reverse transition from a state with a larger number to a state with a smaller number is prohibited; the sum of probabilities in each row is 1; a 33 =1, that is, the fault state S3 is an absorbing state.

[0040] A four-stage fault evolution Hidden Markov Model λ=A,B,π is constructed, with the hidden state set being {S0, S1, S2, S3}. As shown Figure 2 in the four-stage fault evolution HMM state transition and observation distribution, Figure 2 (a) is the four-stage fault evolution HMM state transition diagram, Figure 2 and (b) is the observation probability distribution under each evolution state. The state transition probability matrix A adopts an upper triangular constraint structure: A = [[0.92,0.05,0.02,0.01], [0,0.85,0.10,0.05], [0,0,0.78,0.22], [0,0,0,1.00]] The physical meaning of the above matrix is: from the normal state S0, there is a 92% probability of maintaining normality, a 5% probability of entering transition state I, a 2% probability of entering transition state II corresponding to a sudden fault of jump transition, and a 1% probability of directly entering the fault state corresponding to extreme working conditions. From transition state I, there is an 85% probability of maintaining the state, a 10% probability of deteriorating to transition state II, and a 5% probability of directly entering the fault state. From transition state II, there is a 78% probability of maintaining the state and a 22% probability of deteriorating to the fault state. The fault state S3 is an absorbing state with a self-transition probability of 1.00. All elements in the lower triangle of the matrix are 0, which reflects the irreversible characteristic of fault evolution.

[0041] The initial state distribution π=[1.0,0,0,0], that is, the probability that the system is in the normal state at startup is 1.0.

[0042] The observation probability distribution B is modeled by a four-dimensional Gaussian distribution N μi , Σ i . The mean vectors of the four states are: μ0=[0,0,0,0.5] means that each feature of the normal state fluctuates slightly around the reference value; μ1=[-0.8,0.3,0.4,2.0] means that in transition state I, the voltage change rate is negative, reflecting that the polarization loss and temperature start to rise; μ2=[0.5,-0.2,1.5,5.0] means that in transition state II, the voltage rises back, the temperature rises rapidly, and the temperature difference expands significantly; μ3=[2.0,1.5,4.0,10.0] means that all features of the fault state deviate severely from the normal range.

[0043] S4. The Baum-Welch algorithm with physical constraint regularization is used to perform unsupervised estimation of the parameters of the four-stage fault evolution hidden Markov model. The physical constraint regularization is achieved by introducing a regularization term based on battery physics in the maximization step of expectation maximization. The regularization term ensures that the Joule heating coefficient and voltage fluctuation thermal effect coefficient in the temperature change rate coupling operator are both in the physical feasible region.

[0044] In S4, before the Baum-Welch algorithm iteration, K-means clustering based on electrical-thermal coupling characteristics is used to initialize the model parameters. The number of clusters K=4 corresponds to the four evolution stages from S0 to S3. The cluster center is used as the initial mean vector of the Gaussian distribution of each state, the sample covariance within the cluster is used as the initial covariance matrix, and the sample proportion of each cluster is used as the initial state distribution. The feature vector used for clustering only includes the temperature change rate, voltage change rate, and temperature gradient calculated by the coupling propagation operator. The current integral value is not included in the clustering to avoid interference from the charging and discharging conditions on the initialization.

[0045] Before the Baum-Welch iteration, K-means clustering based on electrical-thermal coupling characteristics was used to initialize the model parameters. K=4 clustering was performed on 3600 four-dimensional observation vectors. The feature vectors used for clustering only included the rate of temperature change, rate of voltage change, and temperature gradient; the current integral value was not included in the clustering to avoid interference from differences in charging and discharging conditions on the initialization. The clustering results are as follows: C0 contains 1985 samples corresponding to normal operating conditions, and the cluster center is μ0. 0 =[0.02,0.01,0.01,0.5]. C1 contains 875 samples corresponding to early anomalies, with cluster center μ1. 0 =[-0.78,0.32,0.42,2.1]. C2 contains 580 samples corresponding to mid-term deterioration, with cluster centers μ2. 0 =[0.48,-0.18,1.48,4.9]. C3 contains 160 samples corresponding to severe faults, with cluster centers μ3. 0 =[1.95,1.42,3.85,9.8]. The proportion of samples in each cluster is used as the initial state distribution π. 0 =[0.551,0.243,0.161,0.044].

[0046] The Baum-Welch iterative process with physical constraint regularization is as follows. Based on the standard expectation-maximization algorithm, a regularization term R is added to the objective function of the maximization step. phys : Where λ1, λ2, λ3, and λ4 are regularization intensity coefficients, R maxand R min These represent the upper and lower physical boundaries of the battery's internal resistance. In this embodiment, the regularization intensity coefficients are λ1=λ2=100.0, λ3=λ4=50.0, and the physical boundary of the internal resistance R... min =0.001Ω, R max =0.100Ω. When the Baum-Welch standard update step produces an estimate of α < 0, the regularization term R... phys The first term in the equation imposes a penalty of strength 100, and the driving parameter is shifted back to the positive domain.

[0047] The convergence of the iteration is as follows: the log-likelihood value L in the first iteration λ =-125.6. 5th iteration L=-62.3. 10th iteration L=-38.7. 20th iteration L=-25.4. 35th iteration L=-22.1. 42nd iteration L=-22.05, the absolute value of the difference between L and the 41st iteration is 0.00003, which is less than 10. -4 The convergence condition has been met. For example... Figure 4 The diagram shows the convergence of Baum-Welch parameter estimation and the backtracking of the Viterbi path, where... Figure 4 (a) in the figure is the convergence curve of the Baum-Welch parameter estimation. Figure 4 (b) in the table represents the Viterbi backtracking optimal state sequence.

[0048] The log-likelihood value increases rapidly in the first 20 iterations with a slope of approximately -5.0 each time, and then enters a period of gradual convergence with a slope of approximately -0.05 each time.

[0049] The converged state transition matrix A * =[[0.935,0.042,0.018,0.005], [0,0.862,0.095,0.043], [0,0,0.795,0.205], [0,0,0,1.000]], the self-transition probability is slightly increased, reflecting the strong persistence of each state in the data. The coupling coefficient α after convergence. * =0.0082°C / W, β * =3.48°C·s / V, R int * =0.0079Ω are all within the physically feasible region.

[0050] S5. The observed sequence is decoded using an adaptive multi-scale backtracking algorithm driven by evolutionary stage confidence. The decoding process first backtracks the optimal hidden state sequence using the standard Viterbi algorithm, then calculates the evolutionary stage confidence at each time step. The evolutionary stage confidence is defined as the ratio of the maximum posterior probability to the second-largest posterior probability at that time step. Then, time steps with confidence below a preset threshold are marked as suspicious transition points. Finally, Viterbi backtracking is performed on the suspicious transition points at the original time scale, the double downsampling scale, and the five-fold downsampling scale, respectively, and the final state is determined by consensus voting to obtain the fault evolutionary stage identification results at each time step.

[0051] In S5, the formula for calculating the confidence level of the evolution stage is C. t =P max / P second , where P max =max i P t post i P represents the maximum posterior probability at that moment. second The second largest posterior probability, the posterior probability P t pos ti =P t fwd i ·P t bwd i / Σ j P t fwd j ·P t bwd j , where P t fwd Let P be the forward probability. t bwd This represents the backward probability.

[0052] In S5, the specific process of determining the final state by multi-scale consensus voting on suspicious transition points is as follows: For time periods marked as suspicious, Viterbi backtracking is performed on the original time scale, the second downsampling scale, and the fifth downsampling scale to obtain three state subsequences. At each time point, the frequency of each state in the three subsequences is counted, and the state with the highest frequency is taken as the final identification result for that time point. If the three subsequences are completely consistent, the confidence level of that time point is marked as the highest level.

[0053] First, the observation sequence is decoded using the standard Viterbi algorithm to obtain the initial optimal state sequence Q. vitThen, for each time t, the posterior probability distribution P is calculated. t post i =P t fwd i ·P t bwd i / Σ j P t fwd j ·P t bwd j , where P fwd Let P be the forward probability. bwd This is the backward probability. The confidence level for each evolutionary stage is defined as C. t =P max / P second That is, the ratio of the maximum a posteriori probability to the second-largest a posteriori probability.

[0054] In this embodiment, the preset threshold τ=2.0. Confidence levels are analyzed for 3600 time points: C from t=0 to 840 time points. t All values ​​with a mean of 4.2 are greater than the threshold, indicating reliable state identification. The period from t=841 to 1520 is the transition zone from S0 to S1, C t It fluctuates between 1.5 and 3.0, with C occurring at eight time points: t=845, 878, 920, 965, 1010, 1120, 1250, and 1380. t Points with a Ct value below 2.0 are marked as suspicious transition points. The period from t=1521 to 2330 is the transition zone from S1 to S2, where Ct is below 2.0 at t=1525, 1600, 1720, 1850, 1980, 2100, and 2220, and is marked as suspicious points. The period from t=2331 to 3600... t The mean of 5.8 is entirely reliable.

[0055] Multi-scale consensus voting correction was performed on the 15 suspicious times mentioned above. Taking t=845 as an example: the Viterbi decoding result at the original time scale is S1. At a double downsampling scale, the result at o... 843 to o 847 The result of re-decoding from 5 points is S0. The result of this is achieved by downsampling o at a scale of five. 833 to o 857 The result of re-decoding 25 points is S0. The voting results across the three scales show S0 appearing twice and S1 appearing once; ultimately, S0 is taken as the corrected result for t=845. For example, at t=1525: the original scale decoding is S2, the 2x downsampling decoding is S1, and the 5x downsampling decoding is also S1. The voting result shows S1 accounting for 2 / 3, so it is corrected to S1.

[0056] S6. Based on the moment when the state first transitions from the normal state to the transition state and the moment when it first reaches the fault state in the fault evolution stage identification results, calculate the fault evolution time window and output the fault evolution tracing report.

[0057] In S6, the output fault evolution tracing report includes four items: evolution stage sequence, propagation path analysis, response time window recommendation, and emergency level determination. The propagation path analysis determines the dominant propagation mechanism based on the relative contribution ratio of Joule heating and voltage fluctuation heat effect in each evolution stage.

[0058] In S6, the fault evolution time window Δt is used to determine the emergency level: when Δt is less than 10 minutes, it is emergency level I, and the machine must be shut down immediately; when Δt is greater than or equal to 10 minutes and less than 30 minutes, it is emergency level II, and maintenance must be carried out within 30 minutes; when Δt is greater than or equal to 30 minutes, it is emergency level III, and the machine is included in routine inspection.

[0059] The final state sequence after multi-scale backtracking correction is as follows: From t=0 to 845, corresponding to 0 to 14.1 minutes, the system is in the S0 normal state. From t=846 to 1520, corresponding to 14.1 to 25.3 minutes, the system is in the S1 transition state I, where the average voltage change rate slowly increases from 0.002V / s to 0.008V / s, and the temperature gradient expands from 1.2°C to 2.8°C. From t=1521 to 2340, corresponding to 25.3 to 39.0 minutes, the system is in the S2 transition state II, where the temperature change rate rises sharply to 1.8°C / s, and the maximum temperature difference of the module reaches 7.5°C. From t=2341 to 3600, corresponding to 39.0 to 60.0 minutes, the system is in the S3 fault state, where the voltage exceeds the 3.65V upper limit, triggering BMS protection.

[0060] Calculate the fault evolution time window Δt=t fault -t onset =2341-846=1495 seconds, approximately 24.9 minutes. According to the emergency level determination rules, Δt=24.9 minutes is greater than or equal to 10 minutes and less than 30 minutes, which is determined to be an emergency level II, and maintenance must be arranged within 30 minutes.

[0061] The generated fault evolution tracing report includes the following four items. The first item is the evolution phase sequence: S0, 0 to 14.1 min to S1, 14.1 to 25.3 min to S2, 25.3 to 39.0 min to S3, and 39.0 to 60.0 min. The second item is the propagation path analysis: the initial electrical anomaly, followed by Joule heating accumulation, then polarization heating superposition, ultimately leading to temperature over-limit voltage protection triggering. Joule heating contributes 87.3% as the dominant propagation mechanism, while polarization heating contributes 12.7% as a secondary mechanism. The third item is the response time window: 24.9 minutes from the initial entry into S1 to reaching S3. It is recommended to complete load reduction or switch to a standby unit before 20 minutes. The fourth item is the emergency level: Level II, requiring a response within 30 minutes.

[0062] To verify the effectiveness of the method of this invention, the performance of the proposed solution was compared with two comparative solutions. Comparative solution A is a traditional BMS fault diagnosis method based on threshold judgment, which only classifies the fault as normal or faulty. Comparative solution B is a fault diagnosis method based on supervised random forest, which requires 450 sets of labeled samples for training.

[0063] The test dataset contains 750 samples, including 150 real-world fault cases (50 cases of overcharging, 50 cases of internal short circuits, 50 cases of thermal runaway) and 600 cases of normal operating conditions. Each sample contains a multi-source monitoring sequence with 3600 time steps.

[0064] Table 1 Performance Comparison Results of Different Solutions As shown in Table 1, the accuracy of the proposed solution in identifying evolutionary stages reaches 91.2%, an improvement of 19.4 percentage points compared to the comparative solution B and 28.7 percentage points compared to the comparative solution A. Key advantages are reflected in: Regarding transition state detection capability, this solution can identify 87.3% of transition state I and 89.6% of transition state II, while the comparative solution B, lacking a stage evolution model, can only incorrectly classify transition states as normal or faulty, resulting in a transition state detection rate of less than 50%. In terms of early warning lead time, this solution issues fault warnings an average of 18.5 minutes in advance, 9.8 minutes earlier than the comparative solution B and 13.3 minutes earlier than the comparative solution A. Regarding label-free operation, this solution employs Baum-Welch unsupervised learning with physical constraint regularization, achieving maximum performance without any labeled samples. In terms of new fault mode discovery, this solution achieves a 72.3% recognition rate for novel fault modes not found in the training set, while the comparative solution B only achieves 15.2%.

[0065] like Figure 5 The performance comparison is shown in bar charts and box plots. Figure 5 (a) in the chart is a bar chart comparing key indicators. Figure 5(b) in the figure is a box plot of the fault identification score distribution. The proposed solution significantly outperforms the two comparative solutions in all key indicators, and the median of the overall fault identification score distribution is 0.89 with an interquartile range of 0.08, indicating a concentrated and relatively high distribution. Figure 6 The diagram shows the fault evolution stage identification distribution and confusion matrix, where... Figure 6 (a) in the figure is a scatter plot of the distribution of fault evolution stages. Figure 6 (b) in the figure represents the confusion matrix for the evolution stage identification. The fault evolution stage exhibits a clear clustering structure on the dV / dt-dT / dt feature plane, and the confusion matrix shows that the identification accuracy of S0 is 93%, S1 is 82%, S2 is 85%, and S3 is 97%.

[0066] The system architecture of this invention consists of three layers: an input layer, a core algorithm layer, and an output layer. The input layer is responsible for acquiring multi-source monitoring data. The core algorithm layer comprises four core modules: four-stage HMM evolution modeling, an electrical-thermal bidirectional coupling propagation operator, Baum-Welch parameter estimation with physical constraint regularization, and adaptive multi-scale backtracking driven by evolution stage confidence. The output layer generates a fault evolution tracing report that includes the evolution stage sequence, propagation path analysis, suggested response time windows, and emergency level determination.

[0067] Based on the same inventive concept, according to another aspect of the present invention, embodiments of the present invention also provide a computer device, comprising: At least one processor; and The memory stores a computer program that can run on the processor. When the processor executes the program, it performs the steps of the above-mentioned method for identifying and tracing the fault evolution stages of a sodium-ion battery energy storage system.

[0068] Based on the same inventive concept, according to another aspect of the present invention, embodiments of the present invention also provide a computer-readable storage medium storing computer program instructions, which, when executed by a processor, perform the steps of the above-described method for identifying and tracing the fault evolution stages of a sodium-ion battery energy storage system. Finally, it should be noted that those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc. The above computer program embodiments can achieve the same or similar effects as any of the corresponding foregoing method embodiments.

[0069] Furthermore, typically, the devices and equipment disclosed in the embodiments of this invention can be various electronic terminal devices, such as mobile phones, personal digital assistants (PDAs), tablet computers (PADs), smart TVs, etc., or they can be large terminal devices, such as servers. Therefore, the scope of protection disclosed in the embodiments of this invention should not be limited to a specific type of device or equipment. The client disclosed in the embodiments of this invention can be applied to any of the above-mentioned electronic terminal devices in the form of electronic hardware, computer software, or a combination of both. Furthermore, the method disclosed in the embodiments of the present invention can also be implemented as a computer program executed by a CPU, which may be stored in a computer-readable storage medium. When the computer program is executed by the CPU, it performs the functions defined in the method disclosed in the embodiments of the present invention. Furthermore, the above-described method steps and system units can also be implemented using a controller and a computer-readable storage medium for storing a computer program that enables the controller to perform the functions of the above-described steps or units. Furthermore, it should be understood that the computer-readable storage medium (e.g., memory) described herein can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. By way of example, and not limitation, non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM), which can act as external cache memory. By way of example, and not limitation, RAM may be available in various forms, such as synchronous RAM (DRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), and direct Rambus RAM (DRRAM). The storage devices disclosed herein are intended to include, but are not limited to, these and other suitable types of memory. The above are exemplary embodiments disclosed in this invention. However, it should be noted that various changes and modifications can be made without departing from the scope of the embodiments of this invention as defined by the claims. The functions, steps, and / or actions of the methods according to the disclosed embodiments described herein do not need to be performed in any marked order. Furthermore, although the elements disclosed in the embodiments of this invention may be described or claimed individually, they may be understood as multiple unless explicitly limited to a singular. In this specification, the same or similar parts between the various embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the descriptions of the embodiments described later are relatively simple, and relevant parts can be referred to the descriptions of the foregoing embodiments.

[0070] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for identifying and tracing the fault evolution stages of a sodium-ion battery energy storage system, characterized in that, The method includes the following steps: S1. Collect multi-source monitoring data of the sodium-ion battery energy storage system, wherein the multi-source monitoring data includes battery cell voltage sequence, charge / discharge current sequence and battery surface temperature sequence; S2. Perform electrical-thermal coupling feature extraction on the multi-source monitoring data to obtain four coupling features: voltage change rate, current integral value, temperature change rate, and temperature gradient. The temperature change rate is calculated using an electrical-thermal bidirectional coupling propagation operator, which includes a Joule heating term and a voltage fluctuation thermal effect term, respectively describing two different physical mechanisms of electrical anomalies propagating to thermal anomalies. Encode the four coupling features into an observation sequence. S3. Construct a four-stage fault evolution hidden Markov model. The hidden state set of the model is {S0, S1, S2, S3}, which correspond to the normal state, transition state I, transition state II and fault state, respectively. The state transition probability matrix of the model adopts an upper triangular constraint structure, which only allows transitions from lower-order states to higher-order states or maintaining the current state. The upper triangular constraint structure reflects the irreversible characteristics of fault evolution. S4. The Baum-Welch algorithm with physical constraint regularization is used to perform unsupervised estimation of the parameters of the four-stage fault evolution hidden Markov model. The physical constraint regularization is achieved by introducing a regularization term based on battery physics in the maximization step of expectation maximization. The regularization term ensures that the Joule heating coefficient and voltage fluctuation thermal effect coefficient in the temperature change rate coupling operator are both in the physical feasible region. S5. The observed sequence is decoded using an adaptive multi-scale backtracking algorithm driven by evolutionary stage confidence. The decoding process first backtracks the optimal hidden state sequence using the standard Viterbi algorithm, then calculates the evolutionary stage confidence at each time step. The evolutionary stage confidence is defined as the ratio of the maximum posterior probability to the second-largest posterior probability at that time step. Then, time steps with confidence below a preset threshold are marked as suspicious transition points. Finally, Viterbi backtracking is performed on the suspicious transition points at the original time scale, the scale of 2x downsampling, and the scale of 5x downsampling, respectively, and the final state is determined by consensus voting to obtain the fault evolutionary stage identification results at each time step. S6. Based on the moment when the state first transitions from the normal state to the transition state and the moment when it first reaches the fault state in the fault evolution stage identification results, calculate the fault evolution time window and output the fault evolution tracing report.

2. The method according to claim 1, characterized in that, In S2, the expression for the electrical-thermal bidirectional coupling propagation operator is: in, For Joule heating, For the voltage fluctuation heat effect term, The Joule heating coefficient is expressed in °C / W. The voltage fluctuation thermal effect coefficient is expressed in °C·s / V, where I is the charging and discharging current, and R is the voltage fluctuation thermal effect coefficient. int This refers to the battery's internal resistance. This represents the rate of change of voltage.

3. The method according to claim 1, characterized in that, In the described S3, the state transition probability matrix A=a ij 4×4 satisfies: a ij =0 when j<i, that is, reverse transition from a state with a larger number to a state with a smaller number is prohibited;​ The sum of the probabilities in each row is 1; a 33 =1, meaning that fault state S3 is an absorption state.

4. The method according to claim 2, characterized in that, In S4, the regularization term Rphys in the maximization step of the Baum-Welch algorithm with physical constraint regularization is specifically: Where λ1, λ2, λ3, and λ4 are regularization intensity coefficients, R max and R min These represent the upper and lower physical limits of the battery's internal resistance.

5. The method according to claim 1, characterized in that, In S5, the formula for calculating the confidence level of the evolution stage is C. t =P max / P second , where P max =max i P t post i P represents the maximum posterior probability at that moment. second The second largest posterior probability, the posterior probability P t pos ti =P t fwd i ·P t bwd i / Σ j P t fwd j ·P t bwd j , where P t fwd Let P be the forward probability. t bwd This represents the backward probability.

6. The method according to claim 1, characterized in that, In S5, the specific process of determining the final state by multi-scale consensus voting on suspicious transition points is as follows: For time periods marked as suspicious, Viterbi backtracking is performed on the original time scale, the second downsampling scale, and the fifth downsampling scale to obtain three state subsequences. At each time point, the frequency of each state in the three subsequences is counted, and the state with the highest frequency is taken as the final identification result for that time point. If the three subsequences are completely consistent, the confidence level of that time point is marked as the highest level.

7. The method according to claim 2, characterized in that, In S4, before the Baum-Welch algorithm iteration, K-means clustering based on electrical-thermal coupling characteristics is used to initialize the model parameters. The number of clusters K=4 corresponds to the four evolution stages from S0 to S3. The cluster center is used as the initial mean vector of the Gaussian distribution of each state, the sample covariance within the cluster is used as the initial covariance matrix, and the sample proportion of each cluster is used as the initial state distribution. The feature vector used for clustering only includes the temperature change rate, voltage change rate, and temperature gradient calculated by the coupling propagation operator. The current integral value is not included in the clustering to avoid interference from the charging and discharging conditions on the initialization.

8. The method according to claim 1, characterized in that, In S6, the output fault evolution tracing report includes four items: evolution stage sequence, propagation path analysis, response time window recommendation, and emergency level determination. The propagation path analysis determines the dominant propagation mechanism based on the relative contribution ratio of Joule heating and voltage fluctuation heat effect in each evolution stage.

9. The method according to claim 8, characterized in that, In S6, the fault evolution time window Δt is used to determine the emergency level: when Δt is less than 10 minutes, it is emergency level I, and the machine must be shut down immediately; when Δt is greater than or equal to 10 minutes and less than 30 minutes, it is emergency level II, and maintenance must be carried out within 30 minutes; when Δt is greater than or equal to 30 minutes, it is emergency level III, and the machine is included in routine inspection.

10. A device for identifying and tracing the fault evolution stages of a sodium-ion battery energy storage system, characterized in that, The device is deployed on a fault evolution stage identification and tracing platform. The device includes a processor and a memory. The memory stores a computer program. When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 9.

Images