A federal collaborative intelligent diagnosis method and system for lithium battery failure

Abstract

The present invention discloses a federated collaborative intelligent diagnosis method and system for lithium battery failure, the method comprising obtaining lithium battery operation data, preprocessing the data using sample entropy, performing parameter aggregation on the data using federated learning, extracting features from the data using dynamic mode decomposition (DMD), encoding the key parameters of the DMD and Autoformer models in the same framework, classifying and diagnosing lithium battery data using a random forest model and an optimized Autoformer model, co-optimizing key parameters using an energy valley optimization algorithm, and finally inputting the diagnosis results into a model library for diagnosis, and feeding back relevant response plans to users; in addition, the present invention also adopts a federated learning method, jointly establishing a global model with lithium batteries of different brands, breaking data silos, and improving diagnostic accuracy. The present invention can effectively improve diagnostic accuracy, enhance the robustness of the system, effectively improve work efficiency, and protect user privacy.

Description

Technical Field

[0001] The present invention belongs to the technical field of lithium battery fault diagnosis, and in particular relates to a federal collaborative intelligent diagnosis method and system for lithium battery faults. Background Art

[0002] As an efficient and clean energy storage device, lithium batteries are widely used in electric vehicles, mobile devices and other fields, and have long been closely related to our lives. However, as the scope of lithium battery applications gradually expands and the audience becomes wider and wider, the safety and reliability issues are becoming increasingly prominent, especially the failure of lithium batteries, which has become the key to restricting the further development of lithium battery technology. Due to the instability of the real environment, the working conditions of lithium batteries are also complex and changeable. For example, lithium batteries are prone to overheating and water immersion in bad weather such as exposure to the sun and heavy rain; the output power of lithium batteries varies greatly under bad road conditions, and they are prone to overcharging and over-discharging; when lithium batteries are exhausted, it is easy to reduce the performance of lithium batteries and shorten their service life.

[0003] Traditional lithium battery fault diagnosis has its own fault data, which will not be shared jointly. Therefore, manual fault diagnosis can only rely on experience, characterization phenomena and relevant opinions provided by each brand to conduct one-by-one troubleshooting and analysis. These methods have insufficient diagnostic accuracy and are time-consuming and labor-intensive. In addition, since data between brands are not shared, many of the same faults need to be repeatedly checked, which seriously affects work efficiency. Secondly, lithium battery fault diagnosis methods mostly use a single model or original model, relying on simple data, which is difficult to converge, resulting in deviations in the final results, making it difficult to meet the needs of enterprises and users. Finally, due to the complexity of lithium battery failures, changes in various conditions also have uncertainty in the formation of failures. Since each link in the system performs information exchange on its own, it is difficult to coordinate effectively and cannot achieve overall optimization. Therefore, a lithium battery fault diagnosis method is needed that can break data silos, improve work efficiency, and improve fault diagnosis efficiency and accuracy. Summary of the invention

[0004] Purpose of the invention: The present invention provides a federated collaborative intelligent diagnosis method and system for lithium battery faults, which realizes intelligent processing of lithium battery data and accurate identification of fault types by combining federated learning algorithm, dynamic mode decomposition, energy valley optimization algorithm and Autoformer model.

[0005] Technical solution: The federal collaborative intelligent diagnosis method for lithium battery failure described in the present invention comprises the following steps:

[0006] (1) Acquire the operating data of the lithium battery including normal data and fault data in time series;

[0007] (2) Preprocess the running data, use sample entropy to evaluate the processed data, set a threshold for sample entropy, and classify data with a sample entropy value greater than the threshold as complex, otherwise classify it as simple;

[0008] (3) Federated learning is used to aggregate parameters of complex and simple data, and a global model is established by combining different brands of lithium batteries to achieve fault parameter sharing.

[0009] (4) Extract features from the operating data using dynamic mode decomposition (DMD) to generate feature vectors;

[0010] (5) Combine the key parameters of the DMD and Autoformer models in the same framework, encode them, and construct a coding set;

[0011] (6) The DMD and Autoformer models are collaboratively optimized based on the energy valley optimization algorithm EVO, and the data classified as simple is input into the random forest model for diagnosis, and the data classified as complex is input into the optimized Autoformer model for diagnosis;

[0012] (7) The feature-extracted data is input into the random forest or optimized Autoformer model for diagnosis, and the fault diagnosis results are output.

[0013] Furthermore, the lithium battery operating data in step (1) includes voltage, current, temperature, state of charge, internal resistance, internal pressure of the lithium battery, and gas monitoring data around the lithium battery.

[0014] Furthermore, the implementation process in step (3) is as follows:

[0015] With s lithium battery brands participating in the federation, all modeling tasks are regarded as the process of optimizing the loss function. The loss function min(f(a)) is the accumulation of diagnostic errors in the model library corresponding to different brands. During the error accumulation process, the impact of the number of lithium battery samples from different brands on the global diagnostic error is considered. f(a) is the global diagnostic error of the model for the current task.

[0016] The search direction of the global model constructed by federated learning is to minimize the global diagnostic error:

[0017]

[0018] In the formula, F i (a) is the failure loss of the i-th user of a brand of lithium battery; n i is the total sample size of the i-th lithium battery; a J is the random forest model parameter, a F are the Autoformer model parameters.

[0019] Furthermore, the implementation process of step (4) is as follows:

[0020] The lithium battery operation data obtained in time series consists of K pieces of time length T, and the expression at time t is:

[0021] x t =Ax t-1 +ε t

[0022] Where A is a K×K coefficient matrix;

[0023] X n For the load data of the nth column of the high-dimensional sampling data matrix, construct a sliding matrix:

[0024]

[0025] The dynamic mode decomposition expression is:

[0026] X2≈AX1

[0027] Perform a singular matrix decomposition on X:

[0028] X1=UΣV T

[0029] Where U is an n×r-dimensional singular matrix, Σ is an r×r-dimensional singular diagonal matrix, V is a (t-1)×r-dimensional singular matrix, r is the rank of X, and U, Σ, and V are all truncated singular values ​​of X1; the approximate matrix is

[0030]

[0031] In the formula, V r ,Σ r are the truncated matrices of U, V, and Σ respectively;

[0032] use calculate The eigenvalue λ and eigenvector ω of , we get the dynamic characteristics:

[0033]

[0034] Where Q is the lithium battery eigenvector matrix, Λ is the diagonal matrix of lithium battery eigenvalues, Q -1 is the inverse matrix of the lithium battery eigenvector matrix.

[0035] Furthermore, the lithium battery characteristic vector matrix Q includes dynamic characteristics of the lithium battery, specifically damping ratio, frequency phase information, real-time temperature change, SOC change, internal resistance change, real-time charge and discharge rate and battery failure mode; the characteristic value λ includes the dynamic characteristics of the lithium battery in the system to which it belongs.

[0036] Furthermore, the implementation process in step (5) is as follows:

[0037] The key parameters of DMD include three singular parameters U, V, and Σ. The key parameters of the Autoformer model include learning rate b, number of encoder layers c, and number of decoder layers d. These six parameters are encoded, parameters are generated within their respective ranges, and a coding set is constructed.

[0038] Furthermore, the implementation process of step (6) is as follows:

[0039] S1: Perform random initialization and use the key parameters of the DMD and Autoformer models as the individual positions of the particles:

[0040]

[0041] Where, X i is the key parameter of the i-th particle, i.e., after encoding, m is the number of candidate parameter sets, t is the number of key parameters of DMD and Autoformer to be optimized, is the jth decision variable of the i-th particle;

[0042] S2: The minimum comprehensive error of the lithium battery diagnosis result is used as the neutron enrichment level NEL, which is:

[0043] According to the diagnostic results corresponding to each set of parameters, the recall rate Li of lithium batteries is calculated. z , false alarm rate Li w , false negative rate Li l , the search direction of EVO optimization parameters is to minimize the comprehensive diagnostic error E, and the calculation formula of E(·) is:

[0044]

[0045] E(U,∑,V,a f )=p z Li z +p w Li w +p l Li l

[0046]

[0047]

[0048]

[0049] Where, YY means that both the real result and the test result of the lithium battery are faulty; YN means that the real result of the lithium battery is faulty and the test result is not faulty; NY means that the real result of the lithium battery is not faulty and the test result is faulty; NN means that both the real result and the test result of the lithium battery are faulty; p z ,p w ,p l They are the recall rate Li z , false alarm rate Li w , false negative rate Li l The corresponding weights are calculated using information entropy;

[0050] Calculate the average diagnostic error EB for the candidate key parameter set:

[0051]

[0052] S3: According to NEL i , determine the normalized diagnostic error SL of the candidate key parameters:

[0053]

[0054] In the formula, BS and WS represent the diagnostic errors corresponding to the optimal parameter value and the worst parameter value, respectively;

[0055] S4: When NEL i >EB, the key parameter group will decay with the strategy of α, β, γ to maintain the stability of the key parameters; randomly generate numbers in the range of [0, 1] as the local search range SB; when SL i When >SB, the key parameters decay with α and γ strategies to maintain the stability of the key parameters; the position update formula is as follows:

[0056]

[0057]

[0058]

[0059] In the formula, is the accumulated error between the i-th group of key parameters and the k-th group of key parameters, (X1, Y1) and (X2, Y2) are the coordinates of the key parameters in the search space; is the new position vector generated after the i-th group of key parameters decays with α as the strategy, i.e., the key parameters of DMD and Autoformer after optimization; X BSis the position vector of the optimal key parameter, AlphaIndexⅡ is a random integer in the range of [1, Alpha IndexⅠ], defining which key parameters are decayed with α as the strategy; the emitted ray is the decision variable of the key parameter, which is removed and replaced by X BS ; is the new position vector for the i-th group of key parameters to decay with the γ strategy, i.e., the key parameters of DMD and Autoformer after optimization; X NG is the position vector of the adjacent key parameters of the i-th group of key parameters, and the position vector is given by Determine; Gamma IndexⅠ is a random integer in the range of [1, t], defining the key parameters for random exploration to avoid falling into the local optimum; Gamma IndexⅡ is a random integer in the range of [1, GammaIndexⅠ], defining which groups of key parameters are randomly explored; the emitted photons are decision variables of the key parameters, which are removed and replaced by the adjacent key parameters or candidate X NG Replacement;

[0060] S5: When SL i ≤SB, the key parameters decay with β as the strategy; the calculation formula is as follows:

[0061]

[0062]

[0063]

[0064] Where, X CP is the mean value of the key parameters, The new position vector generated after the i-th group of key parameters decays with β as the strategy, that is, a new set of key parameters for DMD and Autoformer is generated. This set of key parameters will be sent to X BS and X CP A controlled change is made, r1 and r2 are two random numbers in the range of [0,1], which determine the change amount of the key parameters; another position update process is performed on the key parameters that decay with β as the strategy. At this time, the new position vector generated after the i-th group of key parameters decays with β as the strategy is That is, a new set of key parameters for DMD and Autoformer is generated, and the key parameter of the decay strategy with β as the strategy is shifted to X BS and X NG Make controlled changes, when SL i It does not affect the change process; r3 and r4 are two random numbers in the range of [0,1], which determine the change amount of the key parameters;

[0065] S6: When NEL i When ≤EB, the key parameters change in position and speed to stabilize; at this time, the key parameter update formula is as follows:

[0066]

[0067] In the formula, Generate a new position vector for the i-th group of key parameters through position and speed changes, which is the key parameter of DMD and Autoformer after optimization. r is a random number in the range of [0,1], which determines the change amount of the key parameter.

[0068] S7: After each position update, verify whether the position exceeds the preset upper or lower limit. If it exceeds, it needs to be corrected to ensure that it is within the preset range;

[0069] S8: When the EVO main loop ends, if NEL i >EB, each set of key parameters generates two new sets of key parameters, which are and Instead, only a new set of key parameters is generated, namely Then the newly generated key parameters are merged with the current key parameter set, and the best key parameter set participates in the next search cycle of the algorithm; when the algorithm reaches the maximum number of iterations, the loop is terminated and X is output BS , at this time, the output is decoded, which is the final optimal parameter of DMD and Autoformer model, realizing collaborative optimization.

[0070] The federal collaborative intelligent diagnosis system for lithium battery faults described in the present invention comprises a data acquisition module, a data preprocessing module, a feature extraction module, a fault parameter sharing module and a fault diagnosis module; wherein:

[0071] Data acquisition module, which collects lithium battery operation data, including voltage, current, temperature, state of charge, internal resistance, internal pressure of lithium battery, and gas monitoring around lithium battery;

[0072] The data preprocessing module preprocesses the running data, uses sample entropy to evaluate the processed data, sets a threshold for sample entropy, and classifies data with a sample entropy value greater than the threshold as complex, otherwise it is classified as simple;

[0073] The feature extraction module uses dynamic mode decomposition (DMD) to extract features from the operating data, generate feature vectors, and extract dynamic feature data of the lithium battery;

[0074] The fault parameter sharing module uses federated learning to aggregate parameters of complex and simple data respectively, and jointly builds a global model with lithium batteries of different brands to achieve fault parameter sharing; the key parameters of the DMD and Autoformer models are combined in the same framework, encoded, and a coding set is constructed;

[0075] The fault diagnosis module collaboratively optimizes the DMD and Autoformer models based on the energy valley optimization algorithm EVO, and inputs the data classified as simple into the random forest model for diagnosis, and inputs the data classified as complex into the optimized Autoformer model for diagnosis.

[0076] Furthermore, the system also includes a classification management module, which classifies and manages the data and results, and feeds back the results and related processing solutions to the user end for the user to make an emergency and effective response.

[0077] Beneficial effects: Compared with the prior art, the present invention has the following beneficial effects:

[0078] The present invention can effectively utilize the operating data of lithium batteries of different brands by adopting the method of federated learning, thereby improving the accuracy and robustness of fault diagnosis. By utilizing federated learning, lithium batteries of different brands are combined to construct a global model with stronger applicability and robustness, thus successfully breaking the data island phenomenon and improving the accuracy of the fault diagnosis system.

[0079] The present invention classifies fault data by building a model library, which can effectively improve work efficiency. Using multiple models for processing can more accurately determine the fault type, meet more complex situations, data characteristics and performance requirements, and solve the needs of various enterprises and users.

[0080] The present invention adopts the EVO method for collaboratively optimizing the DMD and Autoformer models, which can effectively cope with the information transmission between different links in the coordination system, and can respond to rapid changes in the environment through real-time communication and rapid decision-making of entities, thereby improving the generalization ability of the model and enhancing the robustness of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1 A flow chart of the federal collaborative intelligent diagnosis method for lithium battery failure proposed by the present invention;

[0082] Figure 2 Flowchart of optimizing DMD and Autoformer models for EVO of the present invention. DETAILED DESCRIPTION

[0083] The present invention will be further described in detail below in conjunction with the accompanying drawings.

[0084] like Figure 1 As shown, the present invention provides a federated collaborative intelligent diagnosis method for lithium battery faults, which uses the energy valley optimization algorithm (EVO) to collaboratively optimize the dynamic mode decomposition (DMD) and the Autoformer optimization model, and uses the random forest model and the Autoformer model for classification diagnosis. The lithium battery data faults can be classified and diagnosed, and the rapid response and reliability of the system are effectively improved. Specifically, the following steps are included:

[0085] Step 1: Acquire the operating data of the lithium battery including normal data and fault data in a time series, including but not limited to voltage, current, temperature, state of charge, internal resistance, internal pressure of the lithium battery, and gas monitoring around the lithium battery.

[0086] Step 2: Preprocess the collected data, use sample entropy to evaluate the processed data, set the sample entropy threshold, and classify the data with sample entropy value greater than the threshold as complex, otherwise it is classified as simple.

[0087] Step 3: Considering that operating data between different lithium battery brands are not shared, federated learning is used to aggregate parameters of complex and simple data separately, and a global model is jointly established for lithium batteries of different brands to achieve fault parameter sharing, keep user privacy data confidential, and effectively improve diagnostic accuracy.

[0088] With s automotive lithium battery brands participating in the federation, all modeling tasks are regarded as the process of optimizing the loss function. The loss function min(f(a)) is the accumulation of diagnostic errors in the model library corresponding to different brands. During the error accumulation process, the impact of the number of lithium battery samples from different brands on the global diagnostic error is considered. f(a) is the global diagnostic error of the model for the current task. The search direction of the global model constructed by federated learning is to minimize the global diagnostic error:

[0089]

[0090] In the formula, F i (a) is the failure loss of the i-th user of a brand of lithium battery, and its calculation model is the random forest model or the EVO-Autoformer model; n i is the total sample size of the i-th lithium battery; a J is the random forest model parameter, a F are the Autoformer model parameters.

[0091] Step 4: Use dynamic mode decomposition (DMD) to extract features from the collected data and generate feature vectors.

[0092] The lithium battery failure data collected in time series consists of K pieces of time length T, and the expression at time t is:

[0093] x t =Ax t-1 +ε t

[0094] Where A is a K×K coefficient matrix;

[0095] X n is the load data of the nth column of the high-dimensional sampling data matrix, so the sliding matrix can be constructed:

[0096]

[0097] The dynamic mode decomposition expression is:

[0098] X2≈AX1

[0099] Perform a singular matrix decomposition on X:

[0100] X1=UΣV T

[0101] Where U is an n×r-dimensional singular matrix, Σ is an r×r-dimensional singular diagonal matrix, V is a (t-1)×r-dimensional singular matrix, r is the rank of X, and U, Σ, and V are all truncated singular values ​​of X1;

[0102] The approximate matrix

[0103]

[0104] In the formula, V r ,Σ r are the truncated matrices of U, V, and Σ respectively;

[0105] use calculate The eigenvalue λ and eigenvector ω of , we can get the dynamic characteristics:

[0106]

[0107] Where Q is the lithium battery eigenvector matrix, Λ is the diagonal matrix of lithium battery eigenvalues, Q -1 is the inverse matrix of the lithium battery eigenvector matrix. Q contains the dynamic characteristics of the lithium battery, including damping ratio, frequency phase information, real-time temperature change, SOC change, internal resistance change, real-time charge and discharge rate and battery failure mode. The eigenvalue λ contains the dynamic characteristics of the lithium battery in the system to which it belongs.

[0108] Step 5: Combine the key parameters of the DMD and Autoformer models in the same framework and encode them.

[0109] The key parameters of DMD include three singular parameters U, V, and Σ. The key parameters of the Autoformer model include learning rate b, number of encoder layers c, and number of decoder layers d. Encoding these six parameters, generating parameters within their respective ranges of variation, constructing a coding set, and using the energy valley optimization algorithm EVO for parameter optimization can effectively improve diagnostic accuracy and improve the overall performance of the system.

[0110] Step 6: Figure 2 As shown, the DMD and Autoformer models are collaboratively optimized based on the energy valley optimization algorithm EVO, and the data classified as simple is input into the random forest model for diagnosis, and the data classified as complex is input into the optimized Autoformer model for diagnosis.

[0111] The encoding of the key parameters is input into the individual position vector of each particle of the energy valley optimization algorithm EVO. The particles undergo α, β, and γ decays to update the optimal position, i.e., the optimal solution. The optimal key parameters are obtained by decoding the optimal solution, thus achieving coordinated optimization of the key parameters of the DMD and Autoformer models.

[0112] S1: Perform random initialization and use the key parameters of the DMD and Autoformer models as the individual positions of the particles:

[0113]

[0114] In the formula, X i is the key parameter of the i-th particle, i.e., after encoding, m is the number of candidate parameter sets, t is the number of key parameters of DMD and Autoformer to be optimized, is the jth decision variable of the ith particle.

[0115] S2: The minimum comprehensive error of the lithium battery diagnosis result is used as the neutron enrichment level (NEL), specifically:

[0116] According to the diagnostic results corresponding to each set of parameters, the recall rate Li of lithium batteries is calculated. z , false alarm rate Li w , false negative rate Li l , the search direction of EVO optimization parameters is to minimize the comprehensive diagnostic error E, and the calculation formula of E(·) is:

[0117]

[0118] E(U,∑,V,a f )=p z Li z +p w Li w +p l Lil

[0119]

[0120]

[0121]

[0122] Where, YY means that both the real result and the test result of the lithium battery are faulty; YN means that the real result of the lithium battery is faulty and the test result is not faulty; NY means that the real result of the lithium battery is not faulty and the test result is faulty; NN means that both the real result and the test result of the lithium battery are faulty; p z ,p w ,p l They are the recall rate Li z , false alarm rate Li w , false negative rate Li l The corresponding weights are calculated using information entropy.

[0123] The average diagnostic error (EB) of the candidate key parameter set is calculated as follows:

[0124]

[0125] S3: According to NEL i , determine the normalized diagnostic error (SL) of the candidate key parameters:

[0126]

[0127] Where BS and WS represent the diagnostic errors corresponding to the optimal parameter value and the worst parameter value, respectively.

[0128] S4: When NEL i >EB, the key parameter group will decay with the strategy of α, β, γ to maintain the stability of the key parameters. Randomly generate numbers in the range [0, 1] as the local search range SB. When SL i When >SB, the key parameters decay with α and γ strategies to maintain the stability of the key parameters. The position update formula is as follows:

[0129]

[0130]

[0131]

[0132] In the formula, is the accumulated error between the i-th group of key parameters and the k-th group of key parameters, (X1, Y1) and (X2, Y2) are the coordinates of the key parameters in the search space. is the new position vector generated after the i-th group of key parameters decays with α as the strategy, that is, the key parameters of DMD and Autoformer after optimization. BS is the position vector of the optimal key parameter, AlphaIndexⅡ is a random integer in the range [1, Alpha IndexⅠ], defining which key parameters are decayed with α as the strategy. The emitted ray is the decision variable of the key parameter, which is removed and replaced by X BS . The new position vector for the i-th group of key parameters to decay with the γ strategy, i.e., the key parameters of DMD and Autoformer after optimization. NG is the position vector of the adjacent key parameters of the i-th group of key parameters, and the position vector is given by Gamma IndexⅠ is a random integer in the range [1, t], defining the key parameters for random exploration to avoid falling into the local optimum. Gamma IndexⅡ is a random integer in the range [1, GammaIndexⅠ], defining which groups of key parameters are randomly explored. The emitted photons are decision variables for the key parameters, which are removed and replaced by neighboring key parameters or candidate X NG Alternative.

[0133] S5: When SL i ≤SB, the key parameters decay with β as the strategy. Due to the high level of instability of these key parameters, the key parameters decay with β as the strategy and change significantly within the range. The calculation formula is as follows:

[0134]

[0135]

[0136]

[0137] In the formula, X CP is the mean value of the key parameters, The new position vector generated after the i-th group of key parameters decays with β as the strategy, that is, a new set of key parameters for DMD and Autoformer is generated. This set of key parameters will be sent to X BS and X CP To make controlled changes, r1 and r2 are two random numbers in the range of [0,1], which determine the amount of change of the key parameters. In order to improve the development and exploration stage of this algorithm, another position update process is performed on the key parameters that decay with β as the strategy. At this time, the new position vector generated after the i-th group of key parameters decays with β as the strategy is That is, a new set of key parameters for DMD and Autoformer is generated, and the key parameter of the decay strategy with β as the strategy is shifted to X BS and X NG Make controlled changes, when SL i It does not affect the change process. r3 and r4 are two random numbers in the range [0,1] that determine the amount of change of the key parameters.

[0138] S6: When NEL i When ≤EB, the key parameters are more likely to change in position and speed to stabilize. At this time, the key parameter update formula is as follows:

[0139]

[0140] In the formula, Generate a new position vector for the i-th group of key parameters through position and speed changes, that is, the key parameters of DMD and Autoformer after optimization. r is a random number in the range of [0,1], which determines the change amount of the key parameters.

[0141] S7: After each position update, it is necessary to verify whether the position exceeds the preset upper or lower limit. If it exceeds, it is necessary to correct it to ensure that it is within the preset range.

[0142] S8: When the EVO main loop ends, if NEL i >EB, each set of key parameters generates two new sets of key parameters, which are and Instead, only a new set of key parameters is generated, namely Then the newly generated key parameters are merged with the current key parameter set, and the best key parameter set participates in the next search cycle of the algorithm; when the algorithm reaches the maximum number of iterations, the loop is terminated and X is output BS , at this time, the output is decoded, which is the final optimal parameter of DMD and Autoformer model, realizing collaborative optimization.

[0143] The data after feature extraction is input into the model in the model library for diagnosis, the diagnosis results are output and the relevant response plans are fed back to the user end.

[0144] The present invention also proposes a federal collaborative intelligent diagnosis system for lithium battery failures, including a data acquisition module, a data preprocessing module, a feature extraction module, a fault parameter sharing module, a fault diagnosis module and a classification management module; wherein:

[0145] Data acquisition module, which collects lithium battery operation data, including voltage, current, temperature, state of charge, internal resistance, internal pressure of lithium battery, and gas monitoring around lithium battery;

[0146] The data preprocessing module preprocesses the running data, uses sample entropy to evaluate the processed data, sets a threshold for sample entropy, and classifies data with a sample entropy value greater than the threshold as complex, otherwise it is classified as simple;

[0147] The feature extraction module uses dynamic mode decomposition (DMD) to extract features from the operating data, generate feature vectors, and extract dynamic feature data of the lithium battery;

[0148] The fault parameter sharing module uses federated learning to aggregate parameters of complex and simple data respectively, and jointly builds a global model with lithium batteries of different brands to achieve fault parameter sharing; the key parameters of the DMD and Autoformer models are combined in the same framework, encoded, and a coding set is constructed;

[0149] The fault diagnosis module performs collaborative optimization of the DMD and Autoformer models based on the energy valley optimization algorithm EVO, and inputs the data classified as simple into the random forest model for diagnosis, and inputs the data classified as complex into the optimized Autoformer model for diagnosis;

[0150] The classification management module classifies and manages data and results, and feeds back the results and related processing solutions to the user end so that the user can make emergency and effective responses.

[0151] The above embodiments are only for illustrating the technical concept and features of the present invention, and their purpose is to enable people familiar with the technology to understand the content of the present invention and implement it accordingly, and they cannot be used to limit the protection scope of the present invention. Any equivalent transformation or modification made according to the spirit of the present invention should be included in the protection scope of the present invention.

Claims

1. A federal collaborative intelligent diagnosis method for lithium battery failure, characterized in that: The following steps are involved: (1) Acquire the operating data of the lithium battery including normal data and fault data in time series; (2) Preprocess the running data, use sample entropy to evaluate the processed data, set a threshold for sample entropy, and classify data with a sample entropy value greater than the threshold as complex, otherwise classify it as simple; (3) Federated learning is used to aggregate parameters of complex and simple data, and a global model is established by combining different brands of lithium batteries to achieve fault parameter sharing. (4) Extract features from the operating data using dynamic mode decomposition (DMD) to generate feature vectors; (5) Combine the key parameters of the DMD and Autoformer models in the same framework, encode them, and construct a coding set; (6) The DMD and Autoformer models are collaboratively optimized based on the energy valley optimization algorithm EVO, and the data classified as simple is input into the random forest model for diagnosis, and the data classified as complex is input into the optimized Autoformer model for diagnosis; (7) Input the feature-extracted data into the random forest or optimized Autoformer model for diagnosis and output the fault diagnosis results; The implementation process in step (5) is as follows: The key parameters of DMD include three singular parameters: U, V, and Σ. The key parameters of the Autoformer model include learning rate b, number of encoder layers c, and number of decoder layers d. Encode these six parameters, generate parameters within their respective ranges, and construct a coding set. The implementation process of step (6) is as follows: S1: Perform random initialization and use the key parameters of the DMD and Autoformer models as the individual positions of the particles: Where, X i is the key parameter of the i-th particle, i.e., after encoding, m is the number of candidate parameter sets, t is the number of key parameters of DMD and Autoformer to be optimized, is the jth decision variable of the i-th particle; S2: The minimum comprehensive error of the lithium battery diagnosis result is used as the neutron enrichment level NEL, which is: According to the diagnostic results corresponding to each set of parameters, the recall rate Li of lithium batteries is calculated. z , false alarm rate Li w , false negative rate Li l , the search direction of EVO optimization parameters is to minimize the comprehensive diagnostic error E, and the calculation formula of E(·) is: E(U,∑,V,a f )=p z Li z +p w Li w +p l Li l Where, YY means that both the actual result and the test result of the lithium battery are faulty; YN means that the actual result of the lithium battery is faulty and the test result is not faulty; NY means the actual result of the lithium battery is no fault, but the test result is faulty; NN means both the actual result and the test result of the lithium battery are no fault; p z ,p w ,p l They are the recall rate Li z , false alarm rate Li w , false negative rate Li l The corresponding weights are calculated using information entropy; Calculate the average diagnostic error EB for the candidate key parameter set: S3: According to NEL i , determine the normalized diagnostic error SL of the candidate key parameters: In the formula, BS and WS represent the diagnostic errors corresponding to the optimal parameter value and the worst parameter value, respectively; S4: When NEL i >EB, the key parameter group will decay with the strategy of α, β, γ to maintain the stability of the key parameters; randomly generate numbers in the range of [0, 1] as the local search range SB; when SL i When >SB, the key parameters decay with α and γ strategies to maintain the stability of the key parameters; the position update formula is as follows: In the formula, is the accumulated error between the i-th group of key parameters and the k-th group of key parameters, (X1, Y1) and (X2, Y2) are the coordinates of the key parameters in the search space; is the new position vector generated after the i-th group of key parameters decays with α as the strategy, i.e., the key parameters of DMD and Autoformer after optimization; X BS is the position vector of the optimal key parameter, Alpha IndexⅡ is a random integer in the range of [1, Alpha IndexⅠ], defining which key parameters are decayed with α as the strategy; the emitted ray is the decision variable of the key parameter, which is removed and replaced by X BS ; is the new position vector for the i-th group of key parameters to decay with the γ strategy, i.e., the key parameters of DMD and Autoformer after optimization; X NG is the position vector of the adjacent key parameters of the i-th group of key parameters, and the position vector is given by Determine; Gamma IndexⅠ is a random integer in the range of [1, t], defining the key parameters for random exploration to avoid falling into the local optimum; Gamma IndexⅡ is a random integer in the range of [1, Gamma IndexⅠ], defining which groups of key parameters are randomly explored; the emitted photons are the decision variables of the key parameters, which are removed and replaced by the adjacent key parameters or candidate X NG Replacement; S5: When SL i ≤SB, the key parameters decay with β as the strategy; the calculation formula is as follows: Where, X CP is the mean value of the key parameters, The new position vector generated after the i-th group of key parameters decays with β as the strategy, that is, a new set of key parameters for DMD and Autoformer is generated. This set of key parameters will be sent to X BS and X CP A controlled change is made, r1 and r2 are two random numbers in the range of [0,1], which determine the change amount of the key parameters; another position update process is performed on the key parameters that decay with β as the strategy. At this time, the new position vector generated after the i-th group of key parameters decays with β as the strategy is That is, a new set of key parameters for DMD and Autoformer is generated, and the key parameter of the decay strategy with β as the strategy is shifted to X BS and X NG Make controlled changes, when SL i It does not affect the change process; r3 and r4 are two random numbers in the range of [0,1], which determine the change amount of the key parameters; S6: When NEL i When ≤EB, the key parameters change in position and speed to stabilize; at this time, the key parameter update formula is as follows: In the formula, Generate a new position vector for the i-th group of key parameters through position and speed changes, which is the key parameter of DMD and Autoformer after optimization. r is a random number in the range of [0,1], which determines the change amount of the key parameter. S7: After each position update, verify whether the position exceeds the preset upper or lower limit. If it exceeds, it needs to be corrected to ensure that it is within the preset range; S8: When the EVO main loop ends, if NEL i >EB, each set of key parameters generates two new sets of key parameters, which are and Instead, only a new set of key parameters is generated, namely Then the newly generated key parameters are merged with the current key parameter set, and the best key parameter set participates in the next search cycle of the algorithm; when the algorithm reaches the maximum number of iterations, the loop is terminated and X is output BS , at this time, the output is decoded, which is the final optimal parameter of DMD and Autoformer model, realizing collaborative optimization.

2. The federal collaborative intelligent diagnosis method for lithium battery failure according to claim 1 is characterized in that: The lithium battery operating data in step (1) includes voltage, current, temperature, state of charge, internal resistance, internal pressure of the lithium battery, and gas monitoring data around the lithium battery.

3. The federal collaborative intelligent diagnosis method for lithium battery failure according to claim 1 is characterized in that: The implementation process in step (3) is as follows: With s lithium battery brands participating in the federation, all modeling tasks are regarded as the process of optimizing the loss function. The loss function min(f(a)) is the accumulation of diagnostic errors in the model library corresponding to different brands. During the error accumulation process, the impact of the number of lithium battery samples from different brands on the global diagnostic error is considered. f(a) is the global diagnostic error of the model for the current task. The search direction of the global model constructed by federated learning is to minimize the global diagnostic error: In the formula, F i (a) is the failure loss of the i-th user of a brand of lithium battery; n i is the total sample size of the i-th lithium battery; a J is the random forest model parameter, a F are the Autoformer model parameters.

4. The method for federal collaborative intelligent diagnosis of lithium battery failure according to claim 1, characterized in that: The implementation process of step (4) is as follows: The lithium battery operation data obtained in time series consists of K pieces of time length T, and the expression at time t is: x t =Ax t-1 +ε t Where A is a K×K coefficient matrix; X n For the load data of the nth column of the high-dimensional sampling data matrix, construct a sliding matrix: The dynamic mode decomposition expression is: X2≈AX1 Perform a singular matrix decomposition on X: X1=UΣV T Where U is an n×r-dimensional singular matrix, Σ is an r×r-dimensional singular diagonal matrix, V is a (t-1)×r-dimensional singular matrix, r is the rank of X, and U, Σ, and V are all truncated singular values ​​of X1; the approximate matrix is In the formula, V r ,Σ r are the truncated matrices of U, V, and Σ respectively; use calculate The eigenvalue λ and eigenvector ω of , we get the dynamic characteristics: Where Q is the lithium battery eigenvector matrix, Λ is the diagonal matrix of lithium battery eigenvalues, Q -1 is the inverse matrix of the lithium battery eigenvector matrix.

5. The method for federal collaborative intelligent diagnosis of lithium battery failure according to claim 4 is characterized in that: The lithium battery characteristic vector matrix Q includes the dynamic characteristics of the lithium battery, specifically damping ratio, frequency phase information, real-time temperature change, SOC change, internal resistance change, real-time charge and discharge rate and battery failure mode.

6. The method for federal collaborative intelligent diagnosis of lithium battery failure according to claim 4, characterized in that: The characteristic value λ includes the dynamic characteristics of the lithium battery in the system to which it belongs.

7. A federal collaborative intelligent diagnosis system for lithium battery failure using the method as claimed in claims 1 to 6, characterized in that: It includes a data acquisition module, a data preprocessing module, a feature extraction module, a fault parameter sharing module and a fault diagnosis module; wherein: Data acquisition module, which collects lithium battery operation data, including voltage, current, temperature, state of charge, internal resistance, internal pressure of lithium battery, and gas monitoring around lithium battery; The data preprocessing module preprocesses the running data, uses sample entropy to evaluate the processed data, sets a threshold for sample entropy, and classifies data with a sample entropy value greater than the threshold as complex, otherwise it is classified as simple; The feature extraction module uses dynamic mode decomposition (DMD) to extract features from the operating data, generate feature vectors, and extract dynamic feature data of the lithium battery; The fault parameter sharing module uses federated learning to aggregate parameters of complex and simple data respectively, and jointly builds a global model with lithium batteries of different brands to achieve fault parameter sharing; the key parameters of the DMD and Autoformer models are combined in the same framework, encoded, and a coding set is constructed; The fault diagnosis module collaboratively optimizes the DMD and Autoformer models based on the energy valley optimization algorithm EVO, and inputs the data classified as simple into the random forest model for diagnosis, and inputs the data classified as complex into the optimized Autoformer model for diagnosis.

8. The federal collaborative intelligent diagnosis system for lithium battery failure according to claim 7 is characterized in that: The system also includes a classification management module for classifying and managing data and results, and feeding back the results and related processing solutions to the user end for the user to make an emergency and effective response.

Images