Energy storage power station battery module thermal diffusion probability evaluation method based on data driving

Abstract

The invention relates to the technical field of energy storage power station state evaluation, and discloses an energy storage power station battery module thermal diffusion probability evaluation method based on data driving, and the method comprises the following steps: S1, collecting historical thermal runaway test data of a battery module in an energy storage power station; based on historical thermal runaway test data, thermal runaway influence characteristics are extracted; calculating the importance index of each thermal runaway influence characteristic; s2, calculating the weight of each importance index through a fuzzy entropy weight method, and evaluating to obtain a comprehensive importance index of each thermal runaway influence feature; based on the comprehensive importance indexes, screening out thermal runaway key factors; s3, constructing a thermal diffusion probability evaluation model based on fuzzy reasoning; state data of a certain battery monomer in the battery module is used as input, and the probability of thermal runaway of other battery monomers in the battery module is used as output. The thermal diffusion probability of the battery module of the energy storage power station can be accurately and efficiently evaluated, and reliable support is provided for safe operation of the energy storage power station.

Description

Technical Field

[0001] The present invention relates to the technical field of energy storage power station state assessment, and in particular to a data-driven method for assessing the thermal diffusion probability of a battery module in an energy storage power station. Background Art

[0002] Electrochemical energy storage power stations can promote large-scale access to renewable energy and improve the stability of regional energy systems by participating in power generation regulation, peak shaving and valley filling, and frequency regulation services. Therefore, they have become a key investment and construction direction for current power generation companies.

[0003] Among them, lithium-ion batteries occupy a dominant position in the electrochemical energy storage power stations that have been put into operation. The total energy of lithium-ion battery energy storage projects put into operation nationwide has reached 72.86GWh. At present, most electrochemical energy storage power stations use a large number of battery modules connected in series or in parallel to meet the voltage and capacity requirements of the energy storage system. However, with the increase in service life and the influence of the time-varying characteristics of operating conditions, battery modules are prone to serious thermal diffusion safety accidents during their aging process due to thermal runaway of battery cells, which seriously threatens the personal safety of operators and the safe and reliable operation of energy storage power stations. Therefore, it is of great significance to reasonably evaluate the thermal diffusion probability of battery modules in electrochemical energy storage power stations and guide the efficient identification of weak links in electrochemical energy storage power stations to ensure and improve the safety and operational reliability of energy storage power stations. At the same time, the evaluation results can also provide a scientific basis for on-site personnel to assist in formulating strategies for battery module replacement priorities, thereby optimizing the allocation of operation and maintenance resources.

[0004] However, the existing thermal diffusion probability assessment schemes still have the following defects: (1) The current thermal diffusion probability assessment methods mainly rely on the simulation calculation of multi-physics field models or periodic tests based on existing guidelines and standards to evaluate the thermal diffusion probability of battery modules. However, these methods are difficult to adapt to the online monitoring needs of the operating status of battery modules in energy storage power stations, which limits their direct integration and deployment in practical applications. (2) The existing thermal diffusion probability assessment model fails to deeply explore the key characteristic factors that affect the thermal runaway behavior of battery cells in the battery module, and fails to comprehensively consider the coupling effects of differentiating factors such as operating conditions and battery location, and does not use advanced technologies such as intelligent algorithms to improve the accuracy of the model, resulting in the inability to effectively improve the thermal diffusion warning capability of the battery module. (3) The current assessment method has not yet formed a complete functional module design, lacking an offline modeling and online application framework including "feature extraction and analysis module-key factor identification module-thermal diffusion probability modeling and assessment module-assessment result output and risk warning module", so that the existing method is difficult to embed into the operation monitoring platform of the existing electrochemical energy storage power station, limiting its promotion of practical application. Summary of the invention

[0005] The present invention aims to provide a data-driven method for evaluating the thermal diffusion probability of battery modules in energy storage power stations, which can accurately and efficiently evaluate the thermal diffusion probability of battery modules in energy storage power stations and provide reliable support for the safe operation of energy storage power stations.

[0006] The basic solution provided by the present invention is: a data-driven method for evaluating the thermal diffusion probability of a battery module in an energy storage power station, comprising the following steps:

[0007] S1, collecting historical thermal runaway test data of battery modules in energy storage power stations; extracting thermal runaway influence features based on the historical thermal runaway test data; and calculating the importance index of each thermal runaway influence feature; the importance index includes a first type of importance index obtained based on the LASSO regression algorithm and a second type of importance index obtained based on the random forest algorithm;

[0008] S2, calculate the weight of each importance index by fuzzy entropy weight method, and evaluate the comprehensive importance index of each thermal runaway influencing feature; based on the comprehensive importance index, screen out the key factors of thermal runaway;

[0009] S3, constructing a thermal diffusion probability assessment model based on fuzzy reasoning; taking the status data of a battery cell in the battery module as input, and taking the probability of thermal runaway of the remaining battery cells in the battery module as output; wherein, the status data of the battery cell is determined according to the key factors of thermal runaway.

[0010] The working principle and advantages of the present invention are:

[0011] The data-driven method for evaluating the thermal diffusion probability of battery modules in energy storage power stations is based on the historical thermal runaway test data of the collected battery modules. The method first uses the LASSO regression algorithm and the random forest algorithm to extract and analyze features, and calculates key influencing factors through the fuzzy entropy weight method to remove redundant factors and reduce the complexity of subsequent modeling, which has high engineering applicability. On this basis, a multi-input-single-output fuzzy reasoning system is constructed, with the extracted key factors as input and the thermal diffusion probability of the battery module as output. By optimizing the fuzzy membership function, a refined thermal diffusion probability evaluation model for lithium-ion battery modules is further established, which can accurately and efficiently evaluate the thermal diffusion probability of battery modules in energy storage power stations. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a schematic diagram of a method flow of an embodiment of a method for evaluating the thermal diffusion probability of a battery module in an energy storage power station based on data-driven method according to the present invention;

[0013] Figure 2 A schematic diagram of the input-output relationship between the modules constructed in the method of the embodiment of the data-driven energy storage power station battery module thermal diffusion probability assessment method of the present invention;

[0014] Figure 3 The importance index calculation result obtained based on the LASSO regression algorithm and the random forest algorithm in the embodiment of the data-driven energy storage power station battery module thermal diffusion probability assessment method of the present invention;

[0015] Figure 4 A typical battery module structure schematic diagram of an embodiment of a data-driven energy storage power station battery module thermal diffusion probability assessment method of the present invention;

[0016] Figure 5 T is an embodiment of the method for evaluating the probability of thermal diffusion of battery modules in energy storage power stations based on data-driven methods of the present invention. s = 0.1, the surface showing the thermal diffusion probability of the battery module as a function of the ambient temperature and the distance from the lithium-ion battery cell;

[0017] Figure 6 T is an embodiment of the method for evaluating the probability of thermal diffusion of battery modules in energy storage power stations based on data-driven methods of the present invention. s = 0.7 when the thermal diffusion probability of the battery module changes with the ambient temperature and the distance from the lithium-ion battery cell;

[0018] Figure 7 is a triangular membership function of an input variable T of an embodiment of a method for evaluating the probability of thermal diffusion of a battery module in an energy storage power station based on data-driven operation according to the present invention;

[0019] Figure 8 is the input variable T of the embodiment of the data-driven energy storage power station battery module thermal diffusion probability evaluation method of the present invention. s The triangle membership function of ;

[0020] Fig. 9 is a triangular membership function of an input variable D of an embodiment of a method for evaluating the probability of thermal diffusion of a battery module in an energy storage power station based on data-driven operation according to the present invention;

[0021] Fig.10 It is the triangular membership function of the output variable P of the embodiment of the data-driven energy storage power station battery module thermal diffusion probability assessment method of the present invention. DETAILED DESCRIPTION

[0022] The following is a further detailed description through specific implementation methods:

[0023] The embodiment is basically as shown in the attached Figure 1 As shown: A data-driven probability assessment method for thermal diffusion of battery modules in energy storage power stations includes the following steps:

[0024] S1, collecting historical thermal runaway test data of battery modules (here specifically lithium-ion battery modules) in energy storage power stations; extracting thermal runaway impact features based on the historical thermal runaway test data; and calculating the importance index of each thermal runaway impact feature; the importance index includes a first type of importance index obtained based on the LASSO regression algorithm and a second type of importance index obtained based on the random forest algorithm.

[0025] Through this step, the ranking of the degree of influence of different variables on the dependent variable can be obtained. And due to the small number of experimental data and variables, it is not suitable for overly complex algorithms. The LASSO regression algorithm selected in this scheme shows good performance in modeling linear relationships. Although linear regression may lead to overfitting of small data sets, the LASSO regression algorithm introduces a regularization function that can prune the model weights, which effectively avoids the problem of overfitting. The random forest model is a non-parametric, non-linear model composed of multiple decision trees. Given its decision tree ensemble structure, the random forest model can measure the relative importance between predictor variables by looking at the degree to which the tree nodes of specific predictor variables reduce the loss function.

[0026] The historical thermal runaway test data includes key factors such as temperature data and time data of thermal runaway of battery cells (here specifically lithium-ion battery cells) under different heating modes, ambient temperatures, SOC levels, and heating area combinations. In addition, the historical thermal runaway test data is recorded according to certain rules.

[0027] The thermal runaway test data of group n is recorded as follows:

[0028]

[0029] In the formula, the test condition parameter T en,n ,U ar,n ,Z hot,n ,SOC n are the ambient temperature (°C) of the battery in the nth group of thermal runaway tests, the heating method of the battery (such as constant temperature heating, etc.), the heating area of ​​the battery (%), and the remaining power of the battery, that is, the SOC level of the battery (%). max,n and t 1,n They respectively represent the temperature at which the battery experiences the first thermal runaway and the test time required from the start of the test to the occurrence of thermal runaway under the nth set of thermal runaway test conditions.

[0030] Taking the heating mode, ambient temperature, SOC level and heating area as the four influencing factors based on expert experience, the temperature and time effects of each influencing factor on battery thermal runaway are evaluated through factor analysis method, and the thermal runaway influence characteristics are extracted.

[0031] Specifically, the factorial analysis method includes the following sub-steps:

[0032] The maximum temperature T of the battery cell when it is in thermal runaway max The time t1 when the battery cell reaches the highest temperature of thermal runaway is taken as the response variable of the linear model of factorial analysis, and the influencing factors are taken as input variables. Meanwhile, the interaction between the four influencing factors is considered to form 10 thermal runaway influence characteristics (represented as C1 to C 10 ) and construct the regression equation f1(·) of the maximum temperature of thermal runaway and the regression equation f2(·) corresponding to the time to reach the maximum temperature of thermal runaway:

[0033] Y1=f1(X1,X2,X3,X4)

[0034] =β0+β1X1+β2X2+β3X3+β4X4+β5X1X2+β6X1X3+β7X1X4+β8X2X3+β9X2X4+β 10 X3X4+ò1;

[0035] Y2=f2(X1,X2,X3,X4)

[0036] =γ0+γ1X1+γ2X2+γ3X3+γ4X4+γ5X1X2+γ6X1X3+γ7X1X4+γ8X2X3+γ9X2X4+γ 10 X3X4+ò2;

[0037] Where, Y1 is the maximum temperature of thermal runaway of the battery cell; Y2 is the time required for the battery cell to reach the maximum temperature; X1, X2, X3, and X4 are the four influencing factors of ambient temperature, heating mode, SOC level, and heating area; β0 and γ0 are the intercepts of the regression equation; β1, ..., β4 and γ1, ..., γ4 are the main effect coefficients of each input variable; β ij and γ ij is the coefficient of interaction between the input variables; ò1 and ò1 are the error terms of the two regression equations.

[0038] The step of calculating the importance index of each thermal runaway impact feature includes the following sub-steps:

[0039] S1.1, based on historical thermal runaway test data, the LASSO regression algorithm (i.e., the least squares method with a regularization term) is used to construct a regression coefficient estimation model.

[0040] Specifically, the core of the LASSO regression algorithm is to introduce a penalty term consisting of the sum of the absolute values ​​of all regression coefficients on the basis of the traditional least squares loss function, thereby exchanging a moderate estimation bias for higher prediction accuracy and stronger generalization ability of the model.

[0041] Taking the regression equation f1(·) as an example, its corresponding loss function can be expressed as:

[0042]

[0043] Where MRSS is the regression loss function after considering the regularization term; the minimum residual sum of squares RSS is the loss function corresponding to the traditional least squares method; λ k represents the regularization term corresponding to the kth regression equation; p is the number of variable coefficients in the regression equation.

[0044] S1.2, the regularization parameter λ was optimized using 10-fold cross validation.

[0045] Specifically, the historical thermal runaway test data is randomly divided into 10 equal parts, the i-th part is used as the validation set, and the remaining 9 parts are used as the training set. This process is repeated 10 times, and the residual sum of squares RSS of each validation set is calculated. test,k,i Finally, the mean residual sum of squares MRSS is obtained by calculating the residual sum of squares of all validation set errors. test,k , to minimize MRSS test,k Find a suitable λ value for the target.

[0046]

[0047] S1.3, repeat S1.2, and preliminarily select B appropriate λ values.

[0048] S1.4, further screen the optimal λ value based on the Bootstrap method.

[0049] Specifically, the Bootstrap method is used to randomly sample a training sample of the same size as the test data set (i.e., a data set composed of historical thermal runaway test data). Then, B different λ values ​​are used in turn, and 10-fold cross validation is used to evaluate the performance of the regression coefficient estimation model. That is, the MRSStest value of the regression coefficient estimation model is calculated, and the λ value that minimizes MRSStest is further selected as the optimal parameter in the regularization term.

[0050] The repeated Bootstrap method was used for 1000 repeated samplings.

[0051] For each sampling, the optimal λ value is used to calculate the coefficient estimation results in the regression coefficient estimation model, and the number of times each coefficient is zero is recorded. The probability of each coefficient being zero is counted; if the probability of a coefficient being zero is greater, it means that the factor corresponding to the coefficient is less important to the response variable, and then the first type of importance index is screened out.

[0052]

[0053] In the formula, BP(C i,k ) represents the regression coefficient β in 1000 regression analyses i Or γ i The probability of zero is used as the first-class importance indicator of the thermal runaway impact characteristic corresponding to this coefficient;

[0054] If the probability is high (such as close to 1), it means that the thermal runaway impact feature corresponding to the coefficient is less important to the response variable; conversely, if the probability is low (such as close to 0), it means that the thermal runaway impact feature corresponding to the coefficient is more important to the response variable.

[0055] Through the above steps, two first-class importance indexes BP (C i,1 ) and BP(C i,2 ).

[0056] S1.5, using a random forest algorithm consisting of multiple decision trees, calculate the second-category importance index of each thermal runaway impact feature.

[0057] Here, the random search method is used to select the hyperparameters in the random forest algorithm, and the exponential distribution is used for sampling to select the best hyperparameter configuration, and the number of decision trees is set to 100 to prevent overfitting.

[0058] When constructing each decision tree, a certain number of features are randomly selected from the 10 thermal runaway impact features composed of 4 influencing factors for node splitting, so that each decision tree is trained on a different feature subset.

[0059] S1.6, calculate the minimum residual sum of squares RSS of the root node before splitting by least squares regression k,init ; Then select a feature and the corresponding split point in sequence to split the data set.

[0060] Assume that feature X is selected i To split, select a split point x i (usually the middle position of the selected characteristic value) and divide the historical thermal runaway test data into two subsets S1 and S2;

[0061]

[0062] For each subset S1 and S2, calculate its minimum residual sum of squares based on the formula in S1.1 and Then the total minimum residual sum of squares after splitting is obtained by weighted averaging:

[0063]

[0064] Where n1 and n2 are the number of samples in subsets S1 and S2 respectively.

[0065] S1.7, according to the rules of S1.6, traverse all selected features and possible split points to find the RSS after splitting split The minimum feature and split point are used as the optimal splitting rule for the current node.

[0066] Then the RSS difference before and after the split is calculated and defined as a measure of the contribution of the feature split to the decision number:

[0067] ΔRSS dis,k,i =RSS root,k,i -RSS split,k,i k=1,2.

[0068] S1.8, repeat the above S1.6 and S1.7 to generate G decision trees.

[0069] Then calculate each feature in each decision tree T g The feature contribution on is normalized and calculated to obtain the importance index of each feature calculated by the random forest method. The larger the importance index value (i.e., the numerical value of the importance), the greater the contribution of the feature (thermal runaway influence feature) to the prediction of the target variable, and vice versa.

[0070]

[0071] In the formula, is the sum of the importance of the i-th feature C1 in the G decision trees corresponding to the regression equation f1. i,k ) is the normalized index of the sum of the importance of the i-th feature C1 regression equation f1 in the G decision trees.

[0072] Through the above steps, the other two importance indexes Im(C i,1 ) and Im(C i,2 ), which is the second type of importance indicator.

[0073] In addition, the models and methods proposed in the above steps are integrated, space is reserved in the existing data workstation of the energy storage power station, the above historical thermal runaway test data is imported into the data workstation through the data interface, and an offline feature extraction and analysis module is constructed; the feature extraction and analysis module is used to execute S1.

[0074] S2, calculate the weight of each importance index by fuzzy entropy weight method, and evaluate the comprehensive importance index of each thermal runaway influencing feature; based on the comprehensive importance index, screen out the key factors of thermal runaway.

[0075] The method of calculating the weight of each importance index by the fuzzy entropy weight method and evaluating the comprehensive importance index of each thermal runaway impact feature includes the following sub-steps:

[0076] S2.1, create a membership matrix, in which the values ​​of each element are the importance indexes corresponding to each thermal runaway impact feature.

[0077] Specifically, create a membership matrix V with 10 rows and 4 columns.

[0078] Each element in the matrix V iq The values ​​of (i=1,2,3…,10; q=1,…,4) are respectively the four importance indices corresponding to the 10 thermal runaway influence features of each thermal runaway test data after being processed by S1.

[0079] For the convenience of modeling and explanation, the four importance indicators of the 10 thermal runaway impact characteristics are recorded as follows:

[0080] Thermal safety importance evaluation index of energy storage battery based on fuzzy entropy weight method

[0081] S2.2, normalize the elements in the membership matrix, and calculate the information entropy of each importance index of each thermal runaway impact feature.

[0082]

[0083] Where, VI iq It is element C iq Normalized value.

[0084]

[0085] In the formula, IF q Represents the information entropy corresponding to the qth importance index.

[0086] S2.3, based on information entropy and the preference weights of energy storage power station operators, calculate the proportional weight of each importance indicator in the thermal runaway impact characteristics.

[0087]

[0088] In the formula, ξ q Represents the fuzzy entropy weight of the qth importance index. g Represents the preference weight of the energy storage power station operator for the qth importance indicator.

[0089] S2.4, based on the proportional weights, evaluate and obtain the comprehensive importance index of each thermal runaway impact feature.

[0090]

[0091] Where WI iq Represents the comprehensive importance index of the i-th thermal runaway impact feature.

[0092] When screening the key factors of thermal runaway, based on the comprehensive importance index, two relatively important influencing factors are extracted from the influencing factors as the key factors of thermal runaway, which helps to reduce the input variables of the fuzzy system in the subsequent S3, thereby greatly reducing the difficulty of establishing the fuzzy system subsequently.

[0093] In this embodiment, the key factors for thermal runaway are the heating mode and the ambient temperature. Figure 3 As shown, the calculation results of the importance index based on the LASSO regression algorithm and the random forest algorithm are displayed. It can be seen that the heating mode and the ambient temperature are the two most important factors affecting the severity of battery thermal runaway.

[0094] In addition, the models and methods proposed in the above steps are integrated to construct an offline influencing factor optimization module in the existing data workstation of the energy storage power station; the influencing factor optimization module is used to execute S2.

[0095] S3, constructing a thermal diffusion probability assessment model based on fuzzy reasoning; taking the status data of a battery cell in the battery module as input, and taking the probability of thermal runaway of the remaining battery cells in the battery module (i.e., thermal diffusion probability) as output; wherein, the status data of the battery cell is determined according to the key factors of thermal runaway.

[0096] When determining the status data of the battery cell, the parameters corresponding to the key factors of thermal runaway are mined as status data through cause-effect analysis.

[0097] When the key factors of thermal runaway are the heating mode and the ambient temperature, the corresponding parameters are the battery temperature T, the heating mode T s and the distance D between the battery cell and the battery cell that has not experienced thermal runaway. The logic of the causal analysis mining is: the higher the ambient temperature of the battery (mainly determined by the distance between the battery itself and the battery cell that has experienced thermal runaway), the more direct the way the battery is heated (that is, the faster the ambient temperature rises), the higher the probability of thermal runaway. In addition, according to an objective law of heat diffusion in lithium-ion battery modules, the higher the battery's own temperature T, the more likely the battery will experience thermal runaway.

[0098] When the key factors of thermal runaway are SOC level and heated area, the corresponding parameters are SOC (i.e. SOC level value, in %) and heated area. The logic of causal analysis mining is: the larger the heated area of ​​the battery, the higher the heat exchange rate with the external high temperature environment, and the greater the possibility of thermal runaway; the greater the remaining battery power, the more active its internal chemical reaction, the more heat released when heated, and the greater the risk of thermal runaway.

[0099] In this embodiment, the key factors of thermal runaway are the heating mode and the ambient temperature, and the T and T of the single battery in the battery module are selected. s , D as input (the following steps are based on the input T, T s , D).

[0100] When constructing a heat diffusion probability assessment model based on fuzzy reasoning, the following sub-steps are included:

[0101] S3.1, constructing a fuzzy inference system based on the historical test data of the energy storage power station; the fuzzy inference system is used to convert the input variables (i.e., T of the single battery, T s , D).

[0102] Specifically, the energy storage power station historical test data includes historical thermal runaway test data and COMSOL thermal runaway simulation data.

[0103] By normalizing the battery temperature T and the heating mode T s Normalized to [0,1] respectively.

[0104] According to expert experience, as long as the temperature of the lithium-ion battery itself exceeds 260°C, the thermal runaway of the lithium-ion battery will be irreversible. Therefore, in this embodiment, the fuzzy generator uses 260°C as the threshold to classify the temperature state of the lithium-ion battery into thermal runaway and thermal runaway, so as to reduce the interference of temperature fluctuations on the system.

[0105] When the D value is 0, it means that the current maximum temperature of the lithium-ion battery in the module does not exceed 260°C.

[0106] Take a typical battery module containing 4 battery cells as an example (see the attached structure diagram Figure 4 ), the value range of D is [0,3]; if battery #1 experiences thermal runaway, and the impact of thermal runaway of battery #1 on thermal runaway of battery #4 is to be evaluated, then the value of D is 3.

[0107] S3.2, establish fuzzy language values ​​and input variables (T, T s , D), the fuzzy domain of the output variable (P, referring to the probability of thermal diffusion), as shown in Table 1 below:

[0108] Table 1 Fuzzy language value and fuzzy domain setting table

[0109] variable Fuzzy language value Fuzzy Domain T ST,MT,BT [0,1] <![CDATA[T s ]]> <![CDATA[ST s ,MT s ,BT s ]]> [0,1] D VD,SD,MD,BD [0,3] P VP,SP,MP,RP,BP [0,1]

[0110] S3.3, establish a fuzzy control rule with multiple inputs and a single output; and construct a fuzzy membership function for each input variable; the fuzzy membership function is a triangular membership function.

[0111] Specifically, in this embodiment, the basic rules followed by the fuzzy reasoning are set as follows: when the battery temperature T is high, the probability of thermal runaway of the battery is high; when the temperature rise rate of the battery environment, i.e., T s The larger the value, the faster the battery absorbs heat and the greater the probability of thermal runaway. In a battery module, the closer the battery is to the battery experiencing thermal runaway, the more likely it is to experience thermal runaway (heat diffusion). Therefore, T and T s The larger the value, the smaller the D value and the higher the probability of thermal runaway.

[0112] like Figure 5 As shown, T s = 0.1, the surface showing the thermal diffusion probability of the battery module as a function of ambient temperature and the distance between the lithium-ion battery cells; Figure 6 As shown, T s = 0.7, the surface showing the thermal diffusion probability of the battery module as a function of ambient temperature and the distance between the lithium-ion battery cells. Figure 5 and Figure 6 It can be seen that the thermal diffusion probability of lithium-ion battery modules under different values ​​of the distance between single cells, the temperature of the battery itself, and the heating mode are significantly different. This is because when the temperature of the battery itself is higher and closer to the point of thermal runaway, the heating mode of the lithium-ion battery is more direct, that is, the faster the ambient temperature rises, the more likely the battery module is to thermally diffuse. This result also verifies the effectiveness of the basic rule setting and the effectiveness of this method.

[0113] Furthermore, a fuzzy control rule with three inputs and one output is established, as shown in the following table (V, S, R, M, and D represent low, lower, normal, higher, and high, respectively).

[0114] Table 2 Fuzzy control rules table

[0115]

[0116] Furthermore, a triangular membership function is constructed for each input variable, such as Figure 7 , Figure 8 and Fig. 9 As shown, T and T s , the triangular membership function of D. Among them, the variables T and T sThe corresponding triangle membership functions have one parameter μ1 and one parameter μ2 that need to be optimized, and the triangle membership function of variable D has three parameters μ3, μ4 and μ5 that need to be optimized.

[0117] S3.4, divide the domain of the output variable into multiple subsets, and construct a fuzzy membership function for the output variable (P, heat diffusion probability), such as Fig.10 As shown in the figure, fuzzy reasoning is performed. The fuzzy membership function here adopts the triangle membership function.

[0118] Specifically, in this embodiment, the domain of the output variable is divided into 5 subsets. In the triangle membership function, the highest vertex of the triangle is a fixed value of 1, and the shape of the triangle is mainly determined by three parameters to be optimized {μ6, μ7, μ8}. In order to prevent unreasonable membership function arrangement or undivided domain from appearing in the fuzzy domain set, {μ6, μ7, μ8} needs to satisfy the increasing relationship in the domain, that is: 0<μ6<μ7<μ8<1.

[0119] S3.5, based on the available test data, an intelligent optimization algorithm is used to simultaneously optimize the eight parameters (μ1 to μ8) in the above fuzzy membership function.

[0120] Specifically, since the thermal diffusion probability of the battery module is positively correlated with the battery temperature, the accuracy of the thermal diffusion probability can be verified by the Pearson correlation coefficient of the battery's own temperature. Therefore, the objective function of the intelligent optimization algorithm is set as follows:

[0121]

[0122] Where PCC is the Pearson correlation coefficient between battery temperature (T) and battery module thermal diffusion probability (P); and are the average values ​​of battery temperature and battery module thermal diffusion probability, respectively. i is the battery temperature in the nth group of thermal runaway test data, P i It is the thermal diffusion probability of the battery module obtained based on the nth group of thermal runaway test data.

[0123] The intelligent optimization algorithm adopts the dung beetle optimization algorithm. By optimizing 8 parameters in the fuzzy membership function of the input variables and the output variables, the maximum PCC coefficient is obtained. Therefore, the dimension of the dung beetle optimization algorithm is 8; the maximum number of iterations and the number of populations are DB1 and DB2, respectively, and the recommended values ​​are 8 and 50, respectively.

[0124] In addition, in order to improve the convergence of the dung beetle optimization algorithm, a strategy is introduced in each round of iteration to achieve the optimization of parameter variables, as follows:

[0125] xnew,r (t+1)=x best,r +δ·R1·(x new,r (t)-LB S )+δ·R2·(x new,r (t)-UB S );

[0126] Where R1 and R2 are random numbers; x best,r represents the current optimal value of the rth parameter; x new,r (t+1) and x new,r (t) represents the optimal value of the rth parameter in the tth round and the t+1th round respectively; LB S and UB S They represent the lower and upper bounds of the parameter values ​​respectively; δ is the dynamic spiral search shape parameter, and the specific calculation method is as follows:

[0127] γ=e zf ·cos(2πf)

[0128]

[0129] Where: k r is a constant; f is a random number in [-1,1].

[0130] S3.6, determine the optimal parameters in each fuzzy membership function, and establish an offline heat diffusion probability evaluation model based on a multi-input single-output fuzzy inference system. And integrate this model into the existing data workstation of the energy storage power station.

[0131] S4, combining the calculation method and model provided in the previous steps, establish an online evaluation module for the thermal diffusion probability of the battery module of the energy storage power station.

[0132] like Figure 2 As shown, in actual applications, after obtaining the online monitoring data of each battery module (including the status data of the battery cell), after being processed by the feature extraction and analysis module, the influencing factor optimization module and the thermal diffusion probability assessment model, the processing results are input into the online assessment module, and the assessment module outputs the thermal diffusion probability of each battery module in real time as the thermal diffusion risk assessment result of each battery module. Preferably, in order to facilitate risk management and operation and maintenance optimization, the assessment results are also sorted from high to low according to the thermal diffusion probability.

[0133] Furthermore, batteries with higher thermal diffusion probabilities can be taken as key monitoring targets and included in the inspection and risk warning scope of energy storage power stations, so that operating personnel can take targeted measures in a timely manner to reduce the risk of thermal diffusion accidents.

[0134] The present embodiment provides a data-driven method for evaluating the thermal diffusion probability of battery modules in energy storage power stations. Based on the historical thermal runaway test data of the collected battery modules, the LASSO regression algorithm and the random forest algorithm are first used for feature extraction and analysis, and the key influencing factors are calculated by the fuzzy entropy weight method to remove redundant factors and reduce the complexity of subsequent modeling, which has high engineering applicability. On this basis, a three-input-single-output fuzzy reasoning system is constructed, with the extracted key factors as input and the thermal diffusion probability of the battery module as output, and the fuzzy membership function is optimized by the dung beetle algorithm, and a refined thermal diffusion probability evaluation model for lithium-ion battery modules is further established, which can accurately and efficiently evaluate the thermal diffusion probability of battery modules in energy storage power stations. In particular, this scheme deeply explores the key characteristic factors that affect the thermal runaway behavior of battery cells in the battery module, and comprehensively considers the coupling effects of differentiated factors such as working conditions and battery positions (corresponding to S1-S2), which helps to improve the accuracy and completeness of the evaluation. The set fuzzy reasoning rules are obviously effective and can ensure the reliability of the evaluation results.

[0135] In addition, this solution is also adapted to the existing data workstation of the energy storage power station. The constructed offline modules and models can be integrated into the existing data workstation of the energy storage power station. Combined with the online evaluation module, it can realize the online evaluation of the thermal diffusion probability of the battery module in the electrochemical energy storage power station, which can effectively improve the early warning capability of the energy storage power station for the thermal diffusion of the battery module and provide reliable support for the safe operation of the energy storage power station. In addition, this solution can form an offline modeling and online application framework including feature extraction and analysis module - influencing factor optimization module - thermal diffusion probability evaluation model - online evaluation module, which can be embedded in the operation monitoring platform of the existing electrochemical energy storage power station, can adapt to the online monitoring needs of the operation status of the battery module of the energy storage power station, and is convenient for direct integration and deployment in practical applications. It has strong adaptability and high practical promotion and application value.

[0136] The above is only an embodiment of the present invention. The common sense such as the known specific structure and characteristics in the scheme is not described in detail here. The ordinary technicians in the relevant field are aware of all the common technical knowledge in the technical field of the invention before the application date or priority date, can obtain all the existing technologies in the field, and have the ability to apply the conventional experimental means before that date. The ordinary technicians in the relevant field can improve and implement the scheme in combination with their own abilities under the enlightenment given by this application. Some typical known structures or known methods should not become obstacles for the ordinary technicians in the relevant field to implement this application. It should be pointed out that for the technicians in this field, without departing from the structure of the present invention, several deformations and improvements can be made, which should also be regarded as the protection scope of the present invention, and these will not affect the effect of the implementation of the present invention and the practicality of the patent.

Claims

1. A data-driven method for evaluating the thermal diffusion probability of battery modules in energy storage power stations, characterized in that: The following steps are involved: S1, collect historical thermal runaway test data of battery modules in energy storage power stations; extract thermal runaway impact characteristics based on the historical thermal runaway test data; And calculate the importance index of each thermal runaway impact feature; the importance index includes a first type of importance index obtained based on the LASSO regression algorithm and a second type of importance index obtained based on the random forest algorithm; S2, calculate the weight of each importance index by fuzzy entropy weight method, and evaluate the comprehensive importance index of each thermal runaway influencing feature; based on the comprehensive importance index, screen out the key factors of thermal runaway; S3, constructing a thermal diffusion probability assessment model based on fuzzy reasoning; taking the status data of a battery cell in the battery module as input, and taking the probability of thermal runaway of the remaining battery cells in the battery module as output; wherein, the status data of the battery cell is determined according to the key factors of thermal runaway.

2. The data-driven probability evaluation method for thermal diffusion of battery modules in energy storage power stations according to claim 1 is characterized in that: In S1, the historical thermal runaway test data includes temperature data and time data of thermal runaway of battery cells under different heating modes, ambient temperatures, SOC levels, and heating area combinations.

3. The data-driven method for evaluating the thermal diffusion probability of a battery module in an energy storage power station according to claim 2 is characterized in that: In S1, the heating mode, ambient temperature, SOC level and heating area are taken as influencing factors. The temperature and time effects of each influencing factor on battery thermal runaway are evaluated through factor analysis method, and the thermal runaway influence characteristics are extracted.

4. The data-driven method for evaluating the thermal diffusion probability of a battery module in an energy storage power station according to claim 3 is characterized in that: The factorial analysis method comprises the following sub-steps: The maximum temperature T of the battery cell when it is in thermal runaway max The time t1 when the battery cell reaches the maximum temperature of thermal runaway is used as the response variable of the linear model of factorial analysis. The influencing factors are used as input variables. The interaction between the four influencing factors is considered to form 10 thermal runaway influencing characteristics. The regression equation f1(·) of the maximum temperature of thermal runaway and the regression equation f2(·) corresponding to the time of reaching the maximum temperature of thermal runaway are constructed: Y1=f1(X1,X2,X3,X4) =β0+β1X1+β2X2+β3X3+β4X4+β5X1X2+β6X1X3+β7X1X4+β8X2X3+β9X2X4+β 10 X3X4+ò1; Y2=f2(X1,X2,X3,X4) =γ0+γ1X1+γ2X2+γ3X3+γ4X4+γ5X1X2+γ6X1X3+γ7X1X4+γ8X2X3+γ9X2X4+γ 10 X3X4+ò2; Where, Y1 is the maximum temperature of thermal runaway of the battery cell; Y2 is the time required for the battery cell to reach the maximum temperature; X1, X2, X3, and X4 are the four influencing factors of ambient temperature, heating mode, SOC level, and heating area; β0 and γ0 are the intercepts of the regression equation; β1, ..., β4 and γ1, ..., γ4 are the main effect coefficients of each input variable; β ij and γ ij is the coefficient of interaction between the input variables; ò1 and ò1 are the error terms of the two regression equations.

5. The data-driven method for evaluating the thermal diffusion probability of a battery module in an energy storage power station according to claim 1 is characterized in that: In S2, the weights of the importance indicators are calculated by the fuzzy entropy weight method, and the comprehensive importance indicators of the thermal runaway impact characteristics are evaluated, including the following sub-steps: Create a membership matrix, in which the values ​​of each element are the importance indexes corresponding to each thermal runaway impact feature; Normalize the elements in the membership matrix and calculate the information entropy of each importance index of each thermal runaway impact feature; Based on information entropy and the preference weights of energy storage power station operators, the proportional weight of each importance index in the thermal runaway impact characteristics is calculated; According to the proportional weight, the comprehensive importance index of each thermal runaway impact feature is evaluated.

6. The data-driven method for evaluating the thermal diffusion probability of a battery module in an energy storage power station according to claim 3 is characterized in that: In S2, when screening the key factors of thermal runaway, based on the comprehensive importance index, two relatively important influencing factors are extracted from the influencing factors as the key factors of thermal runaway.

7. The data-driven method for evaluating the thermal diffusion probability of a battery module in an energy storage power station according to claim 6 is characterized in that: In S3, when determining the status data of the battery cell, the parameters corresponding to the key factors of thermal runaway are mined as the status data through causal analysis.

8. The data-driven energy storage power station battery module thermal diffusion probability assessment method according to claim 7 is characterized in that: When the key factors of thermal runaway are the heating mode and ambient temperature, the corresponding parameters are the battery's own temperature, the heating mode and the distance between the battery cell and the battery cell that has not experienced thermal runaway; when the key factors of thermal runaway are the SOC level and the heating area, the corresponding parameters are the SOC and the heating area.

9. The data-driven method for evaluating the thermal diffusion probability of a battery module in an energy storage power station according to claim 1, characterized in that: In S3, when constructing a heat diffusion probability assessment model based on fuzzy reasoning, the following sub-steps are included: A fuzzifier for constructing a fuzzy inference system based on historical test data of an energy storage power station; the fuzzifier is used to convert input variables; Establish fuzzy language values ​​and fuzzy domains of input variables and output variables; Establish a fuzzy control rule with multiple inputs and a single output; and construct a fuzzy membership function for each input variable; the fuzzy membership function is a triangular membership function; The domain of output variables is divided into multiple subsets, and fuzzy membership functions are constructed for the output variables to perform fuzzy reasoning; Based on the available test data, an intelligent optimization algorithm is used to simultaneously optimize the parameters in the above fuzzy membership functions; The optimal parameters in each fuzzy membership function are determined, and a heat diffusion probability evaluation model based on a multi-input single-output fuzzy inference system is established offline.

10. The data-driven method for evaluating the thermal diffusion probability of a battery module in an energy storage power station according to claim 1, characterized in that: S1 also includes: constructing an offline feature extraction and analysis module; the feature extraction and analysis module is used to execute S1; S2 also includes: constructing an offline influencing factor optimization module; the influencing factor optimization module is used to execute S2.

Images