Battery remaining useful life prediction method based on working temperature correction

Abstract

The application discloses a kind of based on the remaining useful life prediction method of battery after temperature correction, by setting specific charge-discharge strategy simulating the actual operating condition of battery and carrying out accelerated aging test, collect charge-discharge cycle data for model development use;Through multilayer neural network, the voltage, current and capacity data under each cycle are extracted, and are handled into sliding matrix with 50 cycles as interval, the capacity and remaining useful life of battery are preliminarily predicted by multilayer long short-term memory neural network (Long short-term memory, LSTM);Through a large number of accelerated aging data derivation and by big data intelligent algorithm, the Arrhenius RUL prediction semi-empirical model corrected by temperature is obtained;By introducing aging acceleration factor to represent the aging difference caused by different work, the final accurate battery remaining useful life prediction is obtained by combining the preliminary prediction result and the corrected aging deviation.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of lithium ion battery capacity detection, and more particularly to a battery remaining useful life prediction method based on working temperature correction. BACKGROUND

[0002] Lithium ion batteries have become an integral part of our daily electronic devices and most advanced electronic mobile devices due to their high energy density and long life cycle. With the increasing global carbon dioxide emissions and the continuous development of lithium ion batteries, the electrification of the automotive industry is indeed the trend of the times. However, as the battery charging and discharging cycles increase, lithium ion batteries inevitably experience aging and capacity degradation, and the internal resistance increases. It is undeniable that lithium ion batteries bring convenience to various fields of application, but also pose a safety hazard, such as mobile phone battery explosion, lithium battery power bank fire, and Tesla vehicle fire. The commercialization and popularization of electric vehicles will undoubtedly require higher reliability and safety of lithium ion batteries. Therefore, it is of great research significance and practical significance to establish a prediction model for the remaining useful life of battery systems under different use conditions, and to provide a reliable basis for fault prediction and health management of the battery management system.

[0003] However, the factors affecting the capacity degradation of the battery are very complex. Different battery material systems, different working environment temperatures, and different cycle states will result in large differences in battery aging rate and capacity performance. In particular for electric vehicles, driving safety and range anxiety have always been the focus of attention. The same model of electric vehicle has a huge performance difference under different environmental temperatures, and many electric vehicles have a big discount in range after driving to the north. This is not only because the low temperature affects the activity of the active material in the battery, but also because the power of the vehicle air conditioner increases and part of the mechanical energy is lost. When the working temperature of the battery changes, many existing prediction models have difficulty in making targeted adjustments and contributing accurate prediction results. Accurate remaining useful life prediction under different working temperatures is of great significance to reduce vehicle safety hazards, alleviate range anxiety, and enhance the robustness of the automotive battery management system.

[0004] In summary, the current lithium ion battery remaining useful life prediction model does not fully consider the life aging difference caused by the change of battery working temperature, and does not explicitly explore the dependence between working temperature and battery capacity degradation. Therefore, the model does not have a sensitive response to changes in working temperature. In view of this, we propose a battery remaining useful life prediction based on working temperature correction. SUMMARY

[0005] The application provides a battery remaining service life prediction method based on working temperature correction, which aims to make up for the defects of the prior art, comprehensively calculate the life aging factors caused by battery temperature changes, study the dependence between working temperature and battery capacity attenuation, and awaken the response of the model to working temperature changes, so as to provide a battery remaining service life prediction method based on working temperature correction.

[0006] In order to solve the above technical problems and achieve the above purpose, the technical scheme provided by the application is a battery remaining service life prediction method based on working temperature correction, comprising the following steps:

[0007] S1: A special charging and discharging strategy is set to simulate the actual operation condition of the battery and accelerate the battery aging, and the charging process data and the aging acceleration experiment data are recorded;

[0008] S2: The voltage, current, temperature and historical capacity data in the charging process are respectively denoised, abnormal values are removed, and are normalized into data structures of the same shape, and are input into a time sequence feature extraction module to extract compact features of each variable under different cycle numbers;

[0009] S3: The compact features are input into a multi-layer long short-term memory neural network prediction model, the hyperparameter space is searched through a Bayesian optimization algorithm, the best hyperparameters are determined, and a preliminary capacity attenuation curve and a remaining service life prediction value are obtained;

[0010] S4: The actual working temperature data of the battery are input into an Arrhenius temperature correction model, the best semi-empirical model hyperparameters are searched through a Bayesian optimization algorithm, and an aging acceleration factor caused by different working temperatures is obtained;

[0011] S5: The preliminary prediction value of the remaining service life and the aging acceleration factor are integrated to obtain a corrected remaining service life prediction.

[0012] Further, the step S1 is specifically:

[0013] S101: A lithium iron phosphate (LFP) / graphite battery is taken, the nominal capacity is 1.1 Ah, and the rated voltage is 3.3 V;

[0014] S102: The battery is charged at a large rate C1 to the SOC value of the first stage, then is charged at a second rate C2 to 80% of the rated capacity, then is charged at 1C to 3.6 V, and finally is switched to 3.6 V constant voltage charging to less than 0.05C; the temperature is taken at intervals of 20 DEG C, and the temperature interval is from 0 DEG C to 40 DEG C;

[0015] S103: The battery is randomly divided into a training set and a test set according to a ratio of 8:2.

[0016] Further, the step S102, the battery is cycled to failure in 48-channel Arbin LBT charge-discharge workstation and constant temperature box.

[0017] Further, the step S2 is specifically:

[0018] S201: remove the noise of data by data smoothing and removing outliers, and normalize the data into the same matrix shape by Akima interpolation method, and provide suitable input for the time sequence feature extraction module;

[0019] S202: the input data is generated into compact time sequence features by a multi-layer neural network.

[0020] Further, the input data is generated into compact time sequence features by a multi-layer neural network, specifically including: the input matrix is activated after being given a connection weight and then serving as an output, the value of the connection weight determines the importance of the data amount transmitted by the corresponding neuron, the activation function determines whether the state of the corresponding hidden layer neuron is activated, and the key equation of the neural network y is as follows:

[0021]

[0022] Wherein, n is the number of neuron cells, x i is the input of the neural network, w i is the weight, θ is the bias, and f(·) is the activation function, which is the hyperbolic tangent function f(x) here:

[0023]

[0024] Wherein, x generally refers to the input of the activation layer function. The training process adjusts the connection weight by iteration to make the model output approximate the best compact feature.

[0025] Further, the step S3 is specifically:

[0026] S301, process the compact features under each cycle into a sliding matrix with an interval of 50 cycles to realize the matrix input for the battery life prediction from sequence to sequence, that is, the time sequence composed of several capacity degradation data at historical time to predict the time sequence composed of several capacity estimation values at future time, and the obtained sliding matrix is as follows:

[0027]

[0028] In this context, the slip matrix represents the input and output vectors, respectively. The lengths of the input and output sequences are 1000 and 50, respectively, meaning that the battery capacity performance for the next 1001 to 1050 cycles will be predicted using 1000 historical capacity data points. Every 50 prediction steps, the capacity performance for the next 50 cycles is predicted iteratively using the historical capacity data from the previous 1000 cycles. This prediction method requires fewer iterations and accumulates less error; it's essentially a sequence-to-sequence model.

[0029] S302. The deep long short-term memory neural network processes the slip matrix and predicts the capacity in each loop after the prediction starting point. It uses a Bayesian optimization algorithm to find the optimal number of network nodes, network layers, activation function, and learning rate. The gate function and key equations of the processing within a single long short-term memory neural network cell are as follows:

[0030] f t =σ(W xf *X t +W hf *h t-1 +b f );

[0031] i t =σ(W Xi *X t +W hg *h t-1 +b i );

[0032] g t =tanh(W Xg *X t +W hg *h t-1 +b g );

[0033] o t =σ(W Xo *X t +W ho *h t-1 +b o );

[0034] C t =f t ⊙C t-1 +i t ⊙g t ;

[0035] h t =o t ⊙tanh(C t );

[0036] Among them, f t it and o t respectively represent the forget gate output, input gate output and output gate output of the LSTM neural cell at time t, and the three gates can control the transmission process of information in different ways;g t represents the cell candidate memory at time t;C t-1 represents the internal state of the cell at the previous time t-1;C t represents an intermediate state of the internal state of the cell at the current time t, which is changed through a nonlinear function;h t is the hidden layer output at time t, Xt represents the input sliding matrix at time t, W X～ and W h～ are the weights of the neural network, ~ generally refers to f, i, g and o, respectively represent the weight coefficients of the corresponding gate function, and b ～ represents the bias value of the corresponding gate function. The symbol "*" represents the multiplication operation, and the symbol "⊙" represents the vector element product symbol.

[0037] Further, the step S4 is specifically:

[0038] S401: In the accelerated aging test of the battery, the Arrhenius model is introduced to represent the temperature dependence of the remaining service life, the battery capacity attenuation is approximated to a chemical reaction, the Arrhenius equation is used to describe the influence of temperature on the aging reaction and simulate the aging degree of the battery capacity, and a new constant b is introduced to represent the relationship between the aging rate and the reaction rate constant K, and the aging rate D T at the battery working temperature T is introduced as follows:

[0039] D T = b·K

[0040] At this time, the expression of the combined reaction rate and aging rate can be obtained as follows:

[0041]

[0042] Wherein, A is the rate constant of the capacity aging reaction, in order to facilitate subsequent operation, the aging reaction rate constant A T at temperature T is b·A, and after moving term processing, the following formula can be obtained:

[0043]

[0044] The derived formula shows that as long as the degradation mechanism of the battery does not exceed a specific temperature range interval, the aging rate D T and the temperature T are homogeneous, and can be characterized by the Arrhenius model, and lnDT and present a linear relationship with a slope of .

[0045] S402: Take the battery at working temperature 40℃ as the life reference, the working temperature at this time is recorded as reference temperature T0, and the aging rate at this temperature is recorded as Take 20℃ as the interval, introduce an aging acceleration factor ε to correct the remaining service life of the battery at different working temperatures, which is defined as:

[0046]

[0047] At this time, the reaction rate D T of the expression is substituted into the above formula to obtain:

[0048]

[0049] Wherein, E a is the activation energy of the capacity aging reaction, R is the gas constant of the capacity aging reaction, T is the working temperature of the capacity aging reaction, and the aging acceleration factor ε is used to correct the deviation of the service life of the battery caused by different working temperatures;

[0050] S403: Correct the capacity attenuation curve obtained in step S3 based on the working temperature, the greater the battery aging rate, the greater the capacity attenuation offset of the battery, and the offset is affected by the cycle number N, so the capacity offset is:

[0051] ΔC T = C0·ε·N n

[0052] Wherein, ΔC T is the capacity offset of the Nth cycle at the working temperature T0, C0 is the predicted capacity of the Nth cycle at the reference working temperature, and n is a model hyperparameter determined by a hyperparameter optimization algorithm.

[0053] Further, the aging acceleration factor ε is used to correct the aging reaction rate at different working temperatures and further correct the remaining service life of the battery at different working temperatures.

[0054] Further, the reaction rate K of the aging reaction is represented as:

[0055]

[0056] Wherein, A is the rate constant of the capacity aging reaction, E a is the activation energy of the capacity aging reaction, R is the gas constant of the capacity aging reaction, and T is the working temperature of the capacity aging reaction. The higher the working temperature of the battery, the faster the aging reaction rate.

[0057] Further, the step S5 is specifically:

[0058] The predicted capacity curve deviation at different temperatures is calculated to obtain the corrected battery capacity until the end of the battery life as follows:

[0059] C T，i = C 0，i - ΔC T，i

[0060] where i = 1, 2,..., N EOL , N EOL is the number of charge and discharge cycles experienced by the battery to reach the end of life, C 0，i is the capacity of the i th cycle at the reference operating temperature, ΔC T，i is the capacity deviation of the i th cycle at the real-time operating temperature. At this time, the corrected prediction value of the remaining service life can be expressed as:

[0061] N RUL = N EOL - N i

[0062] where N RUL represents the number of charge and discharge cycles experienced from the prediction starting point to the end of life, N i is the deviation value of the cycle life due to different temperatures.

[0063] The beneficial technical effects of the present application are at least the following points:

[0064] (1) The lithium ion battery remaining service life prediction method extracts features from the battery charging process data, which can avoid the influence of random working load, and has strong practicality and stability.

[0065] (2) The lithium ion battery remaining service life prediction method uses long short-term memory neural network to predict RUL based on the compact charging characteristics of each cycle, which effectively alleviates the traditional long sequence gradient explosion and long-term dependence problem.

[0066] (3) The lithium ion battery remaining service life prediction method first derives the dependence relationship between battery operating temperature, battery aging rate, and cycle number based on the Arrhenius equation, and establishes an accurate semi-empirical model to correct the deviation of temperature on RUL prediction.

[0067] (4) The lithium ion battery remaining service life prediction method proposes to regard the battery aging decay as a whole reaction, establishes a complete reaction kinetics equation, and proposes to use the aging acceleration factor ε to correct the battery life at different operating temperatures, which improves the response sensitivity of the model to temperature. BRIEF DESCRIPTION OF DRAWINGS

[0068] The application is further illustrated by the accompanying drawings, but the embodiments in the drawings do not constitute any limitation to the application, and other drawings can be obtained by those of ordinary skill in the art without creative labor on the basis of the following drawings.

[0069] Figure 1 A step flow chart of a battery remaining useful life prediction method based on working temperature correction according to the application;

[0070] Figure 2 A capacity attenuation curve diagram of 124 battery data sets used in the embodiment of the application;

[0071] Figure 3 A schematic diagram of a neuron network structure in the embodiment of the application;

[0072] Figure 4 A schematic diagram of an LSTM cell structure in the embodiment of the application.

[0073] Figure 5 An effect and precision diagram based on working temperature correction when performing RUL prediction in the embodiment of the application. DETAILED DESCRIPTION

[0074] The drawings are only used for illustrative description and cannot be understood as a limitation to the patent;

[0075] For those skilled in the art, it is understandable that some well-known structures in the drawings and their descriptions can be omitted.

[0076] The technical solutions of the application are further described below in combination with the drawings and embodiments.

[0077] As shown in the drawings, Figure 1 A battery remaining useful life prediction method based on working temperature correction, specifically comprising the following steps:

[0078] Step 1, set a special charging and discharging strategy to simulate the actual running condition of the battery and perform battery aging acceleration, and record the charging process data and aging acceleration experiment data.

[0079] Specifically, step 1.1, take a lithium iron phosphate (LFP) / graphite battery, with a nominal capacity of 1.1 Ah and a rated voltage of 3.3 V.

[0080] Step 1.2, the battery used was charged at a constant current (CC1) at a high rate (C1) to a first stage SOC value, then at a second rate (C2) to 80% of the rated capacity, followed by a current charge at 1C (CC3) to 3.6V, and finally switched to a 3.6V constant voltage (CV) charge until the current was less than 0.05C. The operating temperature was in 20℃ intervals, with intervals from 0℃ to 40℃. The battery was cycled to failure in a 48-channel Arbin LBT charge-discharge workstation and a thermostat. As shown in FIG. 1, the capacity decay curve of 124 battery data sets is given. Figure 2

[0081] Step 1.3, the data set was randomly divided into training set and test set in the ratio of 8:2.

[0082] Step 2, the voltage, current, temperature and historical capacity data during charging were respectively denoised, abnormal values were removed, and the data structure was normalized to the same shape, and input into the time series feature extraction module to extract the compact features of each variable at different cycle numbers.

[0083] Specifically, step 2.1, the noise of the data was removed by data smoothing and removing abnormal values, and the data was normalized to the same matrix shape by akima interpolation method to provide suitable input for the time series feature extraction module.

[0084] Step 2.2, the input data generates compact time series features through a multi-layer neural network. The input matrix is activated after being given a connection weight and is used as output. The value of the connection weight determines the importance of the data amount transmitted by the corresponding neuron, and the activation function determines whether the state of the corresponding hidden layer neuron is activated. The key equation of the neural network y is as follows:

[0085]

[0086] As shown in FIG. 2, the neuron network structure is given. Figure 3

[0087] Where n is the number of neuron cells, x i is the input of the neural network, w i is the weight, θ is the bias, and f(·) is the activation function, which is the hyperbolic tangent function here:

[0088]

[0089] Where x generally refers to the input of the activation layer function. The training process adjusts the connection weight through iteration to make the model output approximate the best compact features.

[0090] ​​Step 3, the compact features are input into a multi-layer long short-term memory neural network prediction model, and the hyperparameter space is searched through a Bayesian optimization algorithm to determine the optimal hyperparameters. Through the sequence-to-sequence method, the preliminary capacity decay curve and the remaining useful life prediction value are obtained, and the cumulative error of traditional sequence prediction is reduced.

[0091] Specifically, step 3.1, the compact features under each cycle are processed into a sliding matrix with an interval of 50 cycles to provide a suitable matrix input for battery life prediction from sequence to sequence, i.e., a time series composed of several capacity degradation data at historical time to predict a time series composed of several capacity estimation values at future time. The resulting sliding matrix is as follows:

[0092]

[0093] In the sliding matrix here, the lengths of the input sequence and the output sequence are 1000 and 50, respectively, meaning that the capacity performance of the battery from 1001 to 1050 cycles in the future will be predicted from 1000 historical capacity information. Every 50 steps of prediction, the next 50 cycles of capacity performance are iteratively predicted using the 1000 historical capacity before the prediction starting point. Such a prediction method has fewer iterations and smaller cumulative errors, i.e., a sequence-to-sequence model.

[0094] Step 3.2, these sliding matrices will be processed by a deep long short-term memory neural network to predict the capacity at each cycle after the prediction starting point. The Bayesian optimization algorithm is used to find the optimal number of network nodes, number of network layers, activation function, and learning rate. The key equations of the gate function and processing process in a single long short-term memory neural network cell are as follows:

[0095] f t =σ(W xf *X t +W hf *h t-1 +b f );

[0096] i t =σ(W xi* X t +W hg *h t-1 +b i );

[0097] g t =tanh(W Xg *X t +W hg *h t-1 +b g );

[0098] ot =σ(W xo *X t +W ho *h t-1 +b o );

[0099] C t =f t ⊙C t-1 +i t ⊙ g t ;

[0100] h t =o t ⊙ tanh(C t );

[0101] Among them, f t i t and o t These represent the forgetting gate output, input gate output, and output gate output of the LSTM neuron at time t, respectively. These three gates can control the information transmission process in different ways; g t Represents candidate cellular memory at time t; C t-1 This represents the internal state of the cell at the previous time step t-1; C t This represents an intermediate state of the cell's internal state at time t, which is derived from a nonlinear function; h t X is the hidden layer output at time t. t W represents the slip matrix of the input at time t. X～ and W h～ These are the weights of the neural network, generally referring to f, i, g, and o, which represent the weight coefficients of the corresponding gate functions. Similarly, b... ～ This represents the bias value of the corresponding gate function. The symbol "*" represents the product operation, and "⊙" is the product symbol for vector elements.

[0102] like Figure 4 The diagram shown is a schematic of the structure of an LSTM cell.

[0103] Step 4: Input the actual operating temperature data of the battery into the Arrhenius temperature correction model, and use the Bayesian optimization algorithm to search for the best semi-empirical model hyperparameters to obtain the aging acceleration factor characterized by different operating temperatures.

[0104] Specifically, in step 4.1, the Arrhenius model is introduced into the accelerated aging test of the battery to characterize the temperature dependence of its remaining lifespan. The battery capacity decay is approximated as a chemical reaction, and the Arrhenius equation is used to describe the effect of temperature on this aging reaction and to simulate the degree of battery capacity aging. The chemical reaction rate constant of this reaction can be expressed as:

[0105]

[0106] wherein A, E a , R and T are the rate constant, activation energy, gas constant and working temperature of the capacity aging reaction, respectively. From the equation, the higher the working temperature of the battery, the faster the rate of the aging reaction. At this time, a new constant b is introduced to represent the relationship between the aging rate and the reaction rate constant K, and the aging rate D T , as follows:

[0107] D T = b·K

[0108] At this time, the expression of the combined reaction rate and aging rate can be obtained as follows:

[0109]

[0110] For the convenience of subsequent operation, let A T = b·A, and move the term to obtain:

[0111]

[0112] The derived formula shows that as long as the battery aging degradation mechanism does not exceed a specific temperature range interval, the aging rate D T and the temperature T are homogeneous, and can be characterized by the Arrhenius model, and lnDT and are in a linear relationship with a slope of .

[0113] Step 4.2, take the battery at working temperature 40℃ as the life reference, and the working temperature at this time is recorded as reference temperature T0, and the aging rate at this temperature is recorded as D Take 20 degrees Celsius as an interval, introduce an aging acceleration factor, which is used to correct the aging reaction rate at different working temperatures, and thus correct the remaining service life of the battery at different working temperatures, which is defined as:

[0114]

[0115] At this time, the expression of the reaction rate D T is substituted into the above formula to obtain:

[0116]

[0117] The aging acceleration factor can be used to correct the deviation of the battery service life caused by different working temperatures.

[0118] Step 4.3, the preliminary capacity curve obtained in step three is corrected based on the working temperature. The greater the battery aging rate, the greater the capacity attenuation of the battery, and the greater the offset of the capacity attenuation is affected by the cycle number N, so the capacity offset is:

[0119] ΔC T = C0·ε·N n

[0120] wherein, ΔC T is the capacity offset of the Nth cycle at the working temperature T0, C0 is the predicted capacity of the Nth cycle at the reference working temperature; n is a model hyperparameter, which is determined by a hyperparameter optimization algorithm.

[0121] Step 5, the preliminary prediction value of RUL is integrated with the aging acceleration factor to obtain the corrected RUL prediction.

[0122] Specifically, step 5.1, the predicted capacity curve deviation at different temperatures is calculated to obtain the corrected battery capacity until the end of life (EOL) as follows:

[0123] C T,i = C 0,i -ΔC T,i

[0124] wherein, i = 1, 2,..., N EOL , N EOL is the number of charge and discharge cycles experienced by the battery to reach the end of life, C 0，i is the capacity of the ith cycle at the reference working temperature, and ΔC T，i is the capacity deviation of the ith cycle at the real-time working temperature. At this time, the corrected prediction value of the remaining service life can be expressed as:

[0125] N RUL = N EOL -N i

[0126] wherein, N RUL represents the number of charge and discharge cycles experienced from the prediction starting point to the end of life, and N i is the cycle life deviation value due to different temperatures. As Figure 5 described, the life prediction effect and accuracy of the method disclosed in the present application after working temperature correction are given (two batteries in the experimental data set are taken to show).

[0127] At present, the method disclosed by the application makes a preliminary prediction on the capacity curve of the battery after the prediction starting point, with the battery running at 40 DEG C as the reference, the mean absolute percentage error (MAPE) of the prediction is 0.105 Ah, and the MAPE of the RUL prediction is within 20 cycles.

[0128] The method disclosed by the application successfully establishes an Arrhenius semi-empirical model based on temperature correction through a large amount of data deduction and super parameter optimization of big data intelligent algorithm, when the working temperature of the battery changes, the model method can correct the preliminary life prediction result according to the temperature difference, the MAPE of the corrected capacity prediction is 0.028 Ah, the MAPE of the RUL prediction is within 5 cycles, and the precision is greatly improved.

[0129] Although the embodiments of the application have been shown and described, those skilled in the art can understand that various changes, modifications, replacements and deformations can be made to the embodiments without departing from the principles and purposes of the application, and the scope of the application is defined by the claims and their equivalents.

Claims

1. A battery remaining useful life prediction method based on operating temperature correction, characterized by, Comprise the following steps: S1: set the charging and discharging strategy, simulate the actual operation of the battery and carry out the battery aging acceleration, record the charging process data and aging acceleration experiment data; S2: the voltage, current, temperature and historical capacity data in the charging process are respectively denoised, the abnormal values are removed, the data structures of the same shape are regularized, and are input into the time sequence feature extraction module, and the compact features of each variable under different cycle numbers are extracted; S3: the compact features are input into the multi-layer long short-term memory neural network prediction model, the hyperparameter space is searched through the Bayesian optimization algorithm, the best hyperparameter is determined, and the preliminary capacity attenuation curve and the remaining useful life prediction value are obtained; S4: the actual working temperature data of the battery are input into the Arrhenius temperature correction model, the best semi-empirical model hyperparameter is searched through the Bayesian optimization algorithm, and the aging acceleration factor caused by different working temperatures is obtained; S5: the preliminary prediction value of the remaining useful life and the aging acceleration factor are combined to obtain the corrected remaining useful life prediction; The step S4 is specifically: S401: in the accelerated aging test of the battery, the Arrhenius model is introduced to represent the temperature dependence of the remaining useful life, the capacity attenuation of the battery is approximated to a chemical reaction, and the Arrhenius equation is used to describe the influence of temperature on the aging reaction and simulate the aging degree of the battery capacity; S402: Take the battery at working temperature 40℃ as the life reference, and the working temperature at this time is recorded as Take 20℃ as the interval, introduce an aging acceleration factor to correct the remaining service life of the battery at different working temperatures; S403: the capacity attenuation curve obtained in the step S3 is corrected based on the working temperature; The aging acceleration factor is used to correct the aging reaction rate under different working temperatures and then correct the remaining useful life of the battery under different working temperatures; the reaction rate of the aging reaction is represented by: wherein, k is the rate constant for the capacity aging reaction, Ea is the activation energy for the capacity aging reaction, R is the gas constant for the capacity aging reaction, and T is the operating temperature for the capacity aging reaction; The step S5 is specifically: The predicted capacity curve deviation under different temperatures is calculated to obtain the corrected battery capacity until the end of the battery life as follows: in, , The number of charge-discharge cycles a battery undergoes before reaching the end of its lifespan. For the reference operating temperature, the first The capacity of the circle For the first time under real-time operating temperature Given the capacity deviation of the coil, the corrected prediction of the remaining service life can be expressed as: wherein, represents the number of charge and discharge cycles experienced from the beginning of the prediction to the end of the life, is the deviation value for the cycle life due to the temperature difference.

2. The battery remaining life prediction method based on operating temperature correction according to claim 1, wherein The step S1 is specifically: S101: take a lithium iron phosphate / graphite battery, the nominal capacity is 1.1 Ah, and the rated voltage is 3.3 V; S102: the battery is charged at a large rate C1 to the SOC value of the first stage, then charged at a second rate C2 to 80% of the rated capacity, then charged at 1C to 3.6 V, and finally switched to 3.6 V constant voltage charging until the current is less than 0.05C; the temperature interval is 20 DEG C, and the temperature interval is from 0 DEG C to 40 DEG C; S103: the training set and the test set are randomly divided according to the proportion of 8:

2.

3. The battery remaining life prediction method based on operating temperature correction according to claim 2, characterized by, The step S102, the battery is cycled to failure in a 48-channel Arbin LBT charging and discharging workstation and a thermostat.

4. The battery remaining life prediction method based on operating temperature correction according to claim 1, wherein, The step S2 is specifically: S201: the data noise is removed through data smoothing and abnormal value removal, the data is regularized into the same matrix shape through Akima interpolation method, and is input into the time sequence feature extraction module; S202: the input data generate compact time sequence features through a multi-layer neural network.

5. The battery remaining life prediction method based on operating temperature correction according to claim 4, wherein, The input data is generated into compact time sequence features by the multi-layer neural network, specifically including: the input matrix is activated by a connection weight and then an activation function, the value of the connection weight determines the importance of the data amount transmitted by the corresponding neuron, the activation function determines whether the state of the corresponding hidden layer neuron is activated, and the neural network The key equation is as follows: wherein, is the number of neuron cells, is the input of the neural network, is the weight, is the bias, is the activation function, here the hyperbolic tangent function : wherein, Generally refers to the input of the activation layer function, and the training process adjusts the connection weight through iteration to make the model output approach the best compact features.

6. The battery remaining life prediction method based on operating temperature correction according to claim 1, wherein, The step S3 is specifically: S301, the compact features under each cycle are processed into a sliding matrix with an interval of 50 cycles to realize the matrix input of the sequence-to-sequence battery life prediction, that is, a time sequence composed of several capacity degradation data at a historical time is used to predict a time sequence composed of several capacity estimation values at a future time; S302, the deep long short-term memory neural network processes the sliding matrix, and the prediction of the capacity at each cycle after the prediction starting point is realized through the Bayesian optimization algorithm to find the best network node number, network layer number, activation function and learning rate.

Images