Energy storage battery safety risk quantitative evaluation and failure behavior deduction method based on Bayesian network

Abstract

The invention relates to the field of energy storage battery risk assessment, in particular to an energy storage battery safety risk quantitative assessment and failure behavior deduction method based on a Bayesian network. According to the method, the Bayesian network of the energy storage battery failure path is established, the safety state function of the corresponding Bayesian network node commonly used in various failure modes is provided, the parameters of the safety state function are determined, and the probability under the corresponding condition is obtained. And continuously updating and learning dynamic Bayesian network model parameters based on observation data, realizing backward risk prediction and forward reasoning diagnosis of the failure mode, and performing deduction. The method can reasonably reflect the probability determination function of the safety state of the battery under different conditions, and is suitable for quantitative evaluation of the safety state of the battery under the influence of various complex factors. According to the method, reasons and evolution rules of the battery failure modes are analyzed in combination with the Bayesian network, the occurrence probability of each failure mode is dynamically updated based on observation data, and dynamic deduction of the failure path under various influence factors is realized.

Description

Technical Field

[0001] The present invention relates to the field of energy storage battery risk assessment, and in particular to a method for quantitatively assessing safety risks of energy storage batteries and deducing failure behaviors based on a Bayesian network. Background Art

[0002] In recent years, with the rapid increase in global demand for energy storage technology, the investment in energy storage facilities has increased significantly. However, frequent energy storage safety accidents have attracted widespread attention. By real-time monitoring of the working conditions of batteries in energy storage power stations, potential risk factors can be identified in a timely manner to prevent situations that may lead to battery failure. On the other hand, through in-depth analysis of typical battery failure behaviors, studying the causes of battery failure modes and their evolution laws, the key factors that lead to battery failure can be discovered. Based on these analysis results, measures such as adjusting the battery operating environment and optimizing the management system can effectively reduce the risk of thermal runaway and improve the safety and reliability of energy storage systems.

[0003] The quantitative assessment of battery risks mainly includes the safety status assessment of the battery throughout its life cycle, specifically covering battery performance monitoring, fault detection and diagnosis, thermal management analysis, and external environmental impact assessment, etc. Through the quantitative assessment of battery safety under different operating conditions, the internal state of the battery is determined, and the possibility of various typical failure modes is analyzed to achieve a quantitative assessment of battery risks.

[0004] In the previous methods of quantitative battery risk assessment, the internal state of the battery and the probability of occurrence of typical failure modes under various external environments were mainly determined based on limited observation data, experimental data and expert experience. However, this method is limited by the scarcity of failure data in actual engineering and the one-sidedness of expert experience. It cannot fully consider all engineering factors, and therefore it is difficult to adapt to the quantitative assessment of safety risks under complex working conditions of actual power stations. In addition, in terms of the deduction of failure behavior, most of the existing methods are based on conditional probability calculations, which cannot dynamically evaluate the development direction of failure behavior and it is difficult to analyze the evolution of failure modes. Summary of the invention

[0005] Compared with the above method, the present invention proposes a method for quantitative assessment of safety risks and failure behavior deduction of energy storage batteries based on Bayesian networks. This method can reasonably reflect the probability determination function of the safety state of the battery under different conditions, and is suitable for quantitative assessment of the safety state of the battery under the influence of various complex factors. On the other hand, the present invention combines the Bayesian network to analyze the causes and evolution laws of battery failure modes, dynamically updates the probability of occurrence of each failure mode based on observed data, and realizes the dynamic deduction of failure paths under multiple influencing factors.

[0006] In order to achieve the purpose of the present invention, its technical solution is: a method for quantitatively evaluating the safety risk of energy storage batteries and deducing failure behavior based on a Bayesian network. The technical solution provided by the present invention is as follows:

[0007] A method for quantitatively assessing safety risks of energy storage batteries based on a Bayesian network comprises the following steps:

[0008] S1. Collect failure data and establish a Bayesian network of typical failure modes under corresponding conditions;

[0009] S2. Determine the security status between nodes in the Bayesian network based on reliable data and expert experience, where the security status is expressed as a probability density;

[0010] S3. Dynamically update the Bayesian network model parameters based on the observed data;

[0011] S4. Perform backward risk prediction and forward reasoning diagnosis of failure paths based on conditional probability formula.

[0012] As an improvement, the specific steps of S1 include:

[0013] S1-1. Collect failure data related to energy storage batteries

[0014] The failure data includes but is not limited to battery manufacturing information, usage conditions, internal status, failure mode, etc.; data sources include experimental data, historical operation data and literature data, etc.;

[0015] S1-2. Based on the collected failure data, analyze the battery development rules under each failure mode and establish a Bayesian network of energy storage battery failure modes.

[0016] As an improvement, the specific steps of S2 include:

[0017] S2-1. Determine the probability density function between nodes of a general Bayesian network based on the abuse status;

[0018] S2-2. Combine expert experience to determine the probability density function parameters of each node in the Bayesian network.

[0019] As an improvement, the specific steps of S3 include:

[0020] S3-1. Calculate the expected log-likelihood of the observed data under the current parameters;

[0021] S3-2. Update parameters to maximize expected log-likelihood;

[0022] S3-3. Repeat steps S3-1 and S3-2 until convergence.

[0023] As an improvement, the S4 step specifically includes:

[0024] S4-1. Infer the result event from the cause event;

[0025] S4-2. Infer the cause event through the result event.

[0026] The present invention also discloses a method for deducing safety risks of energy storage batteries based on a Bayesian network, comprising the following steps:

[0027] S1. Collect failure data and establish a Bayesian network of typical failure modes under corresponding conditions;

[0028] S2. Determine the security status between nodes in the Bayesian network based on reliable data and expert experience, where the security status is expressed as a probability density;

[0029] S3. Dynamically update the Bayesian network model parameters based on the observed data;

[0030] S4. Perform backward risk prediction and forward reasoning diagnosis of failure paths based on conditional probability formula;

[0031] S5. Perform battery failure path deduction based on updated observation data:

[0032] S5-1. When there is new observation data, repeat steps S3 and S4 to obtain forward risk prediction and backward reasoning diagnosis results under different observation data.

[0033] S5-2. If the probability of a certain result continues to increase in the updated forward risk prediction and backward reasoning diagnosis results, the cause of the battery failure mode and its evolution law can be determined.

[0034] The advantages of the present invention are:

[0035] 1. The present invention establishes a Bayesian network of the failure path of the energy storage battery, proposes a safety state function for the corresponding Bayesian network nodes under various failure modes, and then determines the parameters of the safety state function of each node based on expert experience and experimental data, and obtains the probability under the corresponding conditions by integrating the probability density function (safety state function). Then, based on the observation data, the dynamic Bayesian network model parameters are continuously updated and learned to achieve backward risk prediction and forward reasoning diagnosis of the failure mode, and determine the battery development path under the failure mode.

[0036] 2. The present invention has multiple data sources and is not affected by the scarcity of failure data in actual engineering and the one-sidedness of expert experience. It comprehensively considers multiple engineering factors and conducts a comprehensive, comprehensive and accurate quantitative assessment of the safety risks of energy storage batteries. It has a wide range of applications and can be used for quantitative assessment of battery safety status under the influence of various complex factors.

[0037] 3. The present invention combines the Bayesian network to analyze the causes of battery failure modes and their evolution laws, dynamically updates the occurrence probability of each failure mode based on observed data, and realizes the dynamic deduction of failure paths under multiple influencing factors. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is an operation flow chart of the present invention.

[0039] Figure 2 Bayesian network diagram of battery failure path. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described in combination with the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. The components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations.

[0041] Example

[0042] This embodiment discloses a method for quantitatively evaluating the safety risk of energy storage batteries and deducing failure behavior based on a Bayesian network. Figure 1 As shown, the method includes:

[0043] S1. Collect failure data and establish a typical failure mode Bayesian network under corresponding conditions. The specific steps include:

[0044] S1-1. Collect failure data related to energy storage batteries

[0045] Failure data includes but is not limited to battery manufacturing information, usage conditions, internal status, failure mode, etc. Data sources can be experimental data, historical operation data, and literature data.

[0046] S1-2. Based on the collected failure data, analyze the battery development rules under each failure mode and establish a Bayesian network of energy storage battery failure modes.

[0047] like Figure 2 As shown in the figure, energy storage batteries are affected by external conditions such as manufacturing process, ambient temperature and charging status, which cause changes in their internal state, such as irreversible lithium deposition and internal short circuit. If not controlled, they will eventually develop into thermal runaway. In addition, energy storage batteries will also have various abnormal failure problems. Based on the above data, a Bayesian network of energy storage battery failure modes is established.

[0048] S2. Determine the probability between nodes in the Bayesian network based on reliable data and expert experience. The specific steps include:

[0049] S2-1. Determine the security status between nodes in the general Bayesian network based on the abuse status, that is, its probability density. The expression of this function is:

[0050]

[0051] In the formula, g(x) is an abuse function and can take any value, including 0 and negative values. k , abuse the function h(x k ) can be expressed as h(x k )=x k In this case, the function is abused to directly obtain the variable value; for multiple variables x1, x2, x3, ..., x n ,The abuse function can combine these variables. For example, multiple variables can be integrated into one abuse function through linear ,combination, nonlinear combination, etc.

[0052] a and b are constants used to control the shape and rate of change of the abuse function. b represents the baseline, i.e. the function value when there is no abuse. a controls the sensitivity of the abuse function. k represents the number of faults that the battery has experienced. c reflects the degree of influence of the number of faults on the safety status. Its value is greater than 0 and can be adjusted according to the specific application scenario. If you want the number of faults to have a more significant impact on the safety status, you can choose a larger c; otherwise, choose a smaller c.

[0053] like Figure 2 As shown in the figure, according to the influencing factors and failure results, multiple functions are involved, among which the functions related to the manufacturing process are: f(normal│manu), f(in sc │manu) and f(age│manu), respectively, represent the probability density of normal condition, internal short circuit and abnormal failure caused by the manufacturing process.

[0054] The functions related to the ambient temperature are: f(ab_age|env_temp) and f(normal|env_temp), which represent the probability density of abnormal failure and normal conditions caused by the ambient temperature, respectively.

[0055] The functions related to the charging state are: f(in_sc|char_state), f(ab_age|char_state), and f(normal|

[0056] char_state), which represent the density of the probability of the charging state causing internal short circuit, abnormal failure and normal state respectively.

[0057] The function of thermal runaway caused by internal short circuit is: f(thermal_runaway|in_sc).

[0058] In this step, the security status between the nodes of the Bayesian network, that is, the probability density data, is first collected to determine the function value of the above function, and then the process proceeds to step S2-2 to reversely determine the parameters in each function.

[0059] S2-2. Combine expert experience to determine the probability density function parameters of each node in the Bayesian network.

[0060] (1) Functions related to manufacturing process

[0061] Since the manufacturing process of energy storage batteries is complex and has many influencing factors, this method simplifies the impact of various factors on the battery manufacturing process and determines the corresponding probability based on expert experience:

[0062]

[0063] Therefore, the functions related to the manufacturing process: f(normal│manu), f(in sc The values ​​of the functions f(age│manu) and f(age│manu) are:

[0064] P(normal│manu)=0.95

[0065] P(in sc │manu)=0.03

[0066] P(age│manu)=0.02

[0067] (2) Function related to ambient temperature

[0068] The ambient temperature is only temperature dependent, so its abuse function can be defined as:

[0069] g(env_temp) = T

[0070] Where T is the corresponding ambient temperature. This abuse function is applicable to f(ab_age|env_temp) and f(normal|

[0071] env_temp) security state function.

[0072] (3) Functions related to charging status

[0073] The charging state is affected by factors such as battery voltage and charging time, and its abuse function can be defined as:

[0074] g(char_state)=xV+yt

[0075] Where V represents the battery voltage; x and y represent the weights of the corresponding influencing factors. The abuse function is applicable to the safety state functions f(in_sc|char_state), f(ab_age|char_state) and f(normal|char_state).

[0076] (4) Correlation functions of thermal runaway caused by internal short circuit

[0077] Under the condition of internal short circuit, the probability of abnormal failure or thermal runaway is mainly affected by the internal short circuit current, and its abuse function can be defined as:

[0078] g(x)=I msc

[0079] In the formula, I msc is the internal short-circuit current. This abuse function is applicable to the f(thermal_runaway|in_sc) safe state function.

[0080] By determining the abuse function between corresponding nodes, and then based on expert experience and experimental data support, the corresponding parameters a, b and c of the security state (probability density) function between corresponding nodes can be determined by determining the security state under three conditions.

[0081] Since the manufacturing process of the battery has been determined when it is prepared, the functions related to the manufacturing process are: f(normal│manu), f(in sc │manu) and f(age│manu) have fixed values, no variables, and no need to further determine the parameters in the formula.

[0082] The functions related to ambient temperature, the functions related to the charging state, and the functions caused by internal short circuits leading to thermal runaway all need to determine the corresponding parameters a, b, and c. There are 6 functions in total, and a total of 6 groups of functions and their corresponding parameters need to be determined.

[0083] S3. Dynamically update the Bayesian network model parameters based on the observed data. The specific steps include:

[0084] S3-1. Calculate the expected log-likelihood of the observed data under the current parameters

[0085] For each unobserved latent variable X j , calculate its posterior probability P(X) under given observation data and current parameters θ j |X obs ,θ).

[0086] S3-2. Update parameters to maximize expected log-likelihood

[0087] Use the posterior probability calculated in step S3-1 to update the conditional probability table, for each node X i and its parent node set Pa(X i ), update the probability condition table according to the expected sufficient statistics.

[0088] Let N(x i ,pa i ) indicates that in the data X i =x i And Pa(X i )=pa i The expected count of , then the update formula is:

[0089]

[0090] Among them, the expected count N(x i ,pa i ) is calculated as:

[0091]

[0092] Here, K is the total number of observations, is the kth observation data.

[0093] S3-3. Repeat steps S3-1 and S3-2 until convergence.

[0094] The convergence criterion is that the change in parameter update is less than a certain threshold, or the change in log-likelihood value is less than a certain threshold.

[0095] S4. Perform backward risk prediction and forward reasoning diagnosis of failure paths based on conditional probability formula. The specific steps include:

[0096] S4-1. Inferring the result event from the cause event

[0097] The inferred formula is:

[0098]

[0099] Among them, P(B i ) represents the prior probability of the original event, and n represents the number of original events that cause event Q to occur.

[0100] S4-2. Inferring the cause event from the result event

[0101] The inferred formula is:

[0102]

[0103] Among them, P(B i |Q) indicates that in B iThe posterior probability of the battery state Q under the state information of Q. In this way, the possible cause of the current failure can be diagnosed.

[0104] S5. Perform battery failure path deduction based on updated observation data. The specific steps include:

[0105] S5-1. When there is new observation data, repeat steps S3 and S4 to obtain forward risk prediction and backward reasoning diagnosis results under different observation data.

[0106] S5-2. If the probability of a certain result continues to increase in the updated forward risk prediction and backward reasoning diagnosis results, the cause of the battery failure mode and its evolution law can be determined.

[0107] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for quantitatively assessing safety risks of energy storage batteries based on Bayesian networks, characterized in that: The steps include: S1. Collect failure data and establish a Bayesian network of typical failure modes under corresponding conditions; S2. Determine the security status between nodes in the Bayesian network based on reliable data and expert experience, where the security status is expressed as a probability density; S3. Dynamically update the Bayesian network model parameters based on the observed data; S4. Perform backward risk prediction and forward reasoning diagnosis of failure paths based on conditional probability formula.

2. According to claim 1, a method for quantitatively assessing safety risks of energy storage batteries based on Bayesian networks is characterized in that: The specific steps of S1 include: S1-1. Collect failure data related to energy storage batteries The failure data includes but is not limited to battery manufacturing information, usage conditions, internal status, failure mode, etc.; data sources include experimental data, historical operation data and literature data, etc.; S1-2. Based on the collected failure data, analyze the battery development rules under each failure mode and establish a Bayesian network of energy storage battery failure modes.

3. According to claim 1, a method for quantitatively assessing safety risks of energy storage batteries based on Bayesian networks is characterized in that: The specific steps of S2 include: S2-1. Determine the probability density function between nodes of a general Bayesian network based on the abuse status; The expression of the probability density function is: Where g(x) is the abuse function; For a single variable x k , abuse the function h(x k ) is h(x k )=x k In this case, the function value of the abused function is the direct variable value; for multiple variables x1, x2, x3, …, x n ,The abuse function integrates multiple variables into one abuse function through linear combination or nonlinear combination; a and b are constants used to control the shape and rate of change of the abuse function. b represents the baseline, that is, the function value when there is no abuse. a controls the sensitivity of the abuse function. k represents the number of failures that the battery has experienced. c reflects the degree of influence of the number of failures on the safety status. Its value is greater than 0 and is adjusted according to the specific application scenario. S2-2. Combine expert experience to determine the probability density function parameters of each node in the Bayesian network.

4. According to claim 1, a method for quantitatively assessing safety risks of energy storage batteries based on Bayesian networks is characterized in that: The specific steps of S3 include: S3-1. Calculate the expected log-likelihood of the observed data under the current parameters For each unobserved latent variable X j , calculate its posterior probability P(X) under given observation data and current parameters θ j |X obs ,θ). S3-2. Update parameters to maximize expected log-likelihood Use the posterior probability calculated in step S3-1 to update the conditional probability table, for each node X i and its parent node set Pa(X i ), update the probability condition table according to the expected sufficient statistics; Let N(x i ,pa i ) indicates that in the data X i =x i And Pa(X i )=pa i The expected count of , then the update formula is: Among them, the expected count N(x i ,pa i ) is calculated as: Where K is the total number of observations, is the kth observation data; S3-3. Repeat steps S3-1 and S3-2 until convergence. The convergence criterion is that the change in parameter update is less than a certain threshold, or the change in log-likelihood value is less than a certain threshold.

5. According to claim 1, a method for quantitatively assessing safety risks of energy storage batteries based on Bayesian networks is characterized in that: The S4 step specifically includes: S4-1. Inferring the result event from the cause event The inferred formula is: Among them, P(B i ) represents the prior probability of the original event, and n represents the number of original events that cause event Q to occur; S4-2. Inferring the cause event from the result event The inferred formula is: Among them, P(B i |Q) indicates that in B i The posterior probability of the battery state Q under the state information.

6. According to claim 2, a method for quantitatively assessing safety risks of energy storage batteries based on Bayesian networks is characterized in that: In the Bayesian network of the energy storage battery failure described in S1-2, the influencing factors include: manufacturing process, ambient temperature, and charging status; the resulting states include normal states and failure states, and the failure states include: internal short circuit, abnormal exhaustion, and thermal runaway.

7. A deduction method based on the evaluation method of claim 1, characterized in that: The steps include: S1. Collect failure data and establish a Bayesian network of typical failure modes under corresponding conditions; S2. Determine the security status between nodes in the Bayesian network based on reliable data and expert experience, where the security status is expressed as a probability density; S3. Dynamically update the Bayesian network model parameters based on the observed data; S4. Perform backward risk prediction and forward reasoning diagnosis of failure paths based on conditional probability formula; S5. Perform battery failure path deduction based on updated observation data: S5-1. When there is new observation data, repeat steps S3 and S4 to obtain forward risk prediction and backward reasoning diagnosis results under different observation data. S5-2. If the probability of a certain result continues to increase in the updated forward risk prediction and backward reasoning diagnosis results, the cause of the battery failure mode and its evolution law can be determined.

Images