An intelligent evaluation method for second-life batteries based on electrochemical model

Abstract

The present invention discloses an intelligent evaluation method for second-life batteries based on an electrochemical model. The method first collects the operating data of the second-life battery under different operating conditions, such as voltage, current, temperature, etc., and performs cleaning, outlier processing and normalization. Based on the Thevenin model, aging and self-discharge related factors are introduced in combination with the characteristics of the second-life battery to construct an electrochemical model, and its parameters are determined using preprocessed data to obtain evaluation results. The present invention gives full play to the advantages of electrochemical and machine learning models, improves the accuracy and reliability of second-life battery performance evaluation, and provides an effective management means and decision-making basis for the secondary utilization of second-life batteries.

Description

Technical Field

[0001] The present invention relates to the field of battery evaluation technology, and in particular to an intelligent evaluation method for second-life batteries based on an electrochemical model, which aims to improve the accuracy, reliability and intelligence level of second-life battery performance evaluation and provide strong support for the secondary utilization of second-life batteries. Background Art

[0002] With the rapid development of the new energy industry, a large number of power batteries are retired after reaching the end of their service life. After screening, reorganization and other cascade utilization processes, these retired batteries can be applied to energy storage systems, low-speed electric vehicles and other fields to achieve resource reuse. However, the performance of cascade batteries varies, and the assessment of their health status (SOH) and remaining useful life (RUL) faces many challenges.

[0003] Traditional evaluation methods are mainly based on the external characteristic parameters of the battery, such as voltage, current, internal resistance, etc., but these methods are difficult to deeply reflect the complex electrochemical processes inside the battery, and the evaluation accuracy is limited. Although the electrochemical model can describe the internal reaction of the battery in principle, the model is complex, the parameters are difficult to obtain and difficult to update in real time, which limits its practical application. At the same time, data-driven machine learning methods lack an understanding of the internal mechanism of the battery, and the physical meaning of the prediction results is unclear when used alone. Therefore, there is an urgent need for an intelligent evaluation method that combines electrochemical models with machine learning to overcome the shortcomings of existing technologies.

[0004] Patent document CN202310842551.1 provides a battery health state estimation method that integrates mechanism and data-driven model in the field of battery health pipeline technology. By constructing a reduced-order electrochemical model with strong battery response behavior characterization, a rich training data set is generated; a data-driven model for estimating battery SOH is constructed; based on a large number of charging and discharging data fragments generated by the reduced-order electrochemical model, a source data set is constructed, and an adaptive momentum estimation algorithm is used to optimize the free variables within the data-driven model to complete the data-driven model pre-training; after obtaining a small amount of real current, voltage data and SOH labels, part of the source data set is mixed with the real data, and transfer learning is performed on the basis of the pre-trained data-driven model to quickly realize SOH estimation of the target domain. However, this method is difficult to characterize the correspondence between the internal parameters of the battery and the available capacity under the coupling influence of multiple aging mechanisms in actual scenarios.

[0005] Patent document CN201710226895.4 identifies four decay characteristic parameters of the model's internal SEI film resistance, initial lithium ion concentration of the negative electrode, and diffusion coefficients of the positive and negative electrodes based on an electrochemical model and an intelligent optimization algorithm, and completes SOH evaluation in combination with a decay characteristic parameter-available capacity change roadmap; however, the evolution law of the selected characteristic parameters with the decay of available capacity varies greatly under different aging modes, and the roadmap that relies on the extraction of laboratory aging experimental data makes it difficult to characterize the corresponding relationship between the internal parameters of the battery and the available capacity under the coupling influence of multiple aging mechanisms in actual scenarios. Summary of the invention

[0006] The purpose of the present invention is to provide an intelligent evaluation method for second-life batteries based on electrochemical models and machine learning, which can simplify the acquisition and update process of model parameters while ensuring the evaluation accuracy, and is suitable for the health status and remaining service life evaluation of large-scale second-life batteries.

[0007] In order to achieve the above object, the present invention provides an intelligent evaluation method for a second-life battery based on an electrochemical model, comprising the following steps:

[0008] S1, collects the operating data of the second-life battery under different working conditions, cleans, processes outliers and normalizes the collected data;

[0009] S2, electrochemical model construction and optimization: Based on the Thevenin model, according to the characteristics of the cascade battery, the electrochemical model is established by introducing aging and self-discharge related factors, the electrochemical model parameters are determined by the pre-processed data in step S1, and the battery evaluation results are obtained through the electrochemical model;

[0010] S3, machine learning model construction and training: select appropriate machine learning algorithm construction and machine learning model, take the electrochemical model output features and preprocessed data features as input, and the battery health status as output, build a training data set, train the model through cross-validation, select the best model structure, and obtain the battery evaluation results through the machine learning model;

[0011] S4, fusion strategy design: A series fusion strategy is used to fuse the calculation results of the electrochemical model and the prediction results of the machine learning model, and the fusion weight is adjusted according to the battery usage scenario to finally obtain the final evaluation result of the battery.

[0012] Furthermore, the evaluation result is the battery's state of health (SOH).

[0013] Further, the electrochemical model is specifically:

[0014] Internal resistance change function: R0(t) = α·e βN+γ·T+δ, where R0(t) represents the ohmic internal resistance that changes with time, N is the number of charge and discharge cycles, T is the temperature, and α, β, γ, and δ are parameters obtained by fitting a large amount of experimental data;

[0015] Battery aging function: A(t) is the comprehensive aging degree, P i (t) represents the i-th internal resistance change rate, w is a parameter obtained by fitting a large amount of data;

[0016] Self-discharge function: S d is the self-discharge rate, κ1, κ2, κ3 are coefficients obtained by fitting the experimental data;

[0017] The relationship function between ohmic internal resistance, temperature and SOC: A, B, C, and D are parameters obtained by fitting a large amount of experimental data.

[0018] Furthermore, in step S3, the machine learning algorithm includes one or more of a support vector machine, a random forest, a neural network, or a long short-term memory network.

[0019] Furthermore, the outputs of the electrochemical model and the machine learning model can be fused by weighted averaging. The specific weight adjustment mechanism is as follows:

[0020]

[0021] Among them: y is the final prediction result; y ECM is the prediction result of the electrochemical model; y ML is the prediction result of the machine learning model; w1 and w2 are the weights of the electrochemical model and the machine learning model respectively, satisfying w1+w2=1.

[0022] Furthermore, in step S4, the model weight is adjusted according to the battery usage scenario:

[0023] As a backup power source in communication base stations, the physical mechanism advantage of the electrochemical model is more obvious, so the weight of the electrochemical model can be increased and the weight of the machine learning model can be reduced, setting w1=0.7 and w2=0.3;

[0024] As a power source in low-speed electric vehicles, the machine learning model can better capture the nonlinear behavior of the battery, so the weight of the machine learning model can be increased and the weight of the electrochemical model can be reduced, setting w1 = 0.3 and w2 = 0.7;

[0025] As energy storage batteries in the energy storage system, the secondary batteries in the energy storage system need to have a certain power response capability to quickly track changes in grid load. The machine learning model can automatically adjust parameters according to the data, so the weight of the machine learning model can be appropriately increased, setting w1=0.4, w2=0.6.

[0026] The present invention has the following beneficial effects:

[0027] (1) The electrochemical model and the machine learning model are combined to give full play to the advantages of the electrochemical model in its ability to explain the internal mechanism of the battery, and the advantages of the machine learning model in its ability to capture the nonlinear behavior of the battery and its strong adaptability and generalization capabilities. The model weights are adjusted according to the battery application scenarios, which can adapt to the performance changes of cascade batteries in different application scenarios, effectively making up for the limitations of a single model and improving the accuracy and reliability of battery status prediction.

[0028] (2) The electrochemical model is optimized based on the traditional Thevenin model. Aiming at the characteristics of second-life batteries, the internal resistance change law caused by factors such as changes in the electrode material structure, decomposition and drying of the electrolyte during battery aging is considered, and a dynamic internal resistance change function is introduced. In addition, aging and self-discharge related factors are fully introduced to establish a comprehensive aging model and self-discharge function, so that the model is more in line with the actual performance of second-life batteries and can more accurately reflect the status of the battery under different working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0030] Figure 1 The figure is a flow chart of an intelligent evaluation method for second-life batteries based on an electrochemical model in the present invention.

[0031] Figure 2 Schematic diagram of the structure of the traditional Thevenin model. DETAILED DESCRIPTION

[0032] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0033] It should be understood that when used in this specification and the appended claims, the term "comprising" indicates the presence of described features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or collections.

[0034] In order to simplify the drawings, only the parts related to the present invention are schematically shown in each figure, and they do not represent the actual structure of the product. In addition, in order to simplify the drawings and facilitate understanding, in some figures, only one of the parts with the same structure or function is schematically drawn or marked. In this article, "one" not only means "only one", but also means "more than one".

[0035] It should be further understood that the term “and / or” used in the specification and appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0036] In the embodiments shown in the drawings, the indications of directions (such as up, down, left, right, front and back) used to explain the structure and movement of the various components of the present invention are not absolute but relative. These descriptions are appropriate when these components are in the positions shown in the drawings. If the descriptions of the positions of these components change, the indications of these directions also change accordingly.

[0037] In addition, in the description of the present application, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0038] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the specific implementation methods of the present invention will be described below with reference to the accompanying drawings. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings and other implementation methods can be obtained based on these drawings without creative work.

[0039] See also Figure 1 , the method in this embodiment mainly includes:

[0040] Data collection and preprocessing

[0041] Build a highly integrated battery test platform with multi-condition simulation capabilities. The platform can accurately simulate various complex working conditions that second-life batteries may encounter in actual applications, such as different charge and discharge rates (covering from low rates of 0.1C to high rates of 5C or even higher), wide temperature range (from low temperature of -40℃ to high temperature of 60℃), dynamic load changes (simulating scenarios such as frequent start and stop of electric vehicles and power mutation of energy storage systems). On the test platform, deploy a high-precision sensor array, including voltage sensors (accuracy up to ±0.001V), current sensors (accuracy up to ±0.01A), temperature sensors (accuracy up to ±0.5℃), internal resistance meters (accuracy up to ±0.001Ω) and counters that can accurately record the number of charge and discharge cycles. Through these sensors, real-time collection of operating data of second-life batteries under different working conditions, including voltage, current, temperature, number of charge and discharge cycles, internal resistance changes and other information, to ensure the comprehensiveness and accuracy of the data.

[0042] The collected data is first subjected to strict data cleaning operations, using advanced filtering algorithms (such as Kalman filtering, wavelet transform filtering, etc.) to remove noise interference in the data, and combining statistical outlier detection methods (such as 3σ criteria, box plot method, etc.) to identify and properly handle abnormal data points. Subsequently, normalization technology is used to uniformly map data of different dimensions to a specific interval, improve data consistency and availability, and lay the foundation for subsequent model training and analysis.

[0043] Electrochemical model construction and optimization

[0044] The Thevenin model is selected as the basic electrochemical model, and the model structure is optimized according to the characteristics of the cascade battery. The Thevenin model, also known as the equivalent circuit model, is a battery model widely used in battery research and battery management systems (BMS). Based on circuit theory, it treats the battery as a combination of a voltage source and several circuit elements such as resistors and capacitors to simulate the electrochemical characteristics and dynamic behavior of the battery.

[0045] Figure 2It is a structural diagram of the traditional Thevenin model, which usually includes the following parts: open circuit voltage (OCV), ohmic internal resistance (R0), polarization resistance and capacitance (R1, C1). The open circuit voltage (OCV) reflects the electromotive force of the battery in the open circuit state. Its size is closely related to the remaining capacity (SOC) and temperature of the battery. It is an intuitive reflection of the conversion of the battery's electrochemical energy into electrical energy. The ohmic internal resistance covers the obstacles to current conduction caused by factors such as electrode materials, electrolytes, diaphragms, and contact resistance between components, which directly affects the energy loss and power output capacity of the battery during charging and discharging. The RC network composed of polarization resistance and polarization capacitance in parallel is used to describe the complex polarization phenomenon inside the battery. This phenomenon originates from the delayed effect of charge transfer and ion diffusion in the electrochemical reaction kinetics process, and plays a key role in the dynamic change of the battery terminal voltage.

[0046] Since second-life batteries have gone through a certain period of use, they are significantly different from new batteries in terms of material structure and performance. For example, the electrode materials inside the battery may undergo structural changes and active material loss, resulting in battery capacity decay, increased internal resistance, and changes in charge and discharge characteristics. When selecting the Thevenin model for second-life battery evaluation, the model structure needs to be optimized based on these characteristics. On the one hand, depending on the degree of aging of the second-life battery, the modeling of the ohmic internal resistance needs to be more detailed, considering its change over time and the number of charge and discharge cycles, and introducing a function or parameter adjustment mechanism that can reflect the trend of internal resistance growth. On the other hand, the polarization phenomenon may be more complex in second-life batteries, so the parameter setting and dynamic adjustment strategy of the polarization resistance and polarization capacitance need to be reviewed to more accurately capture the changes in the polarization characteristics of the battery at different aging stages and operating conditions.

[0047] However, since the operating conditions of second-life batteries are more complicated, in order to evaluate the batteries more comprehensively and accurately, new factors related to battery aging and self-discharge are added to the Thevenin model to establish an electrochemical model.

[0048] On the basis of the traditional Thevenin model, the modeling of ohmic internal resistance is further refined, considering the change of internal resistance caused by factors such as changes in electrode material structure, decomposition and drying of electrolyte during battery aging, and introducing a dynamic internal resistance change function, which is related to factors such as the number of battery charge and discharge cycles, cumulative discharge capacity, and temperature history. The following internal resistance change function is established:

[0049] R0(t)=α·e βN +γ·T+δ, where R0(t) represents the ohmic internal resistance that changes with time, N is the number of charge and discharge cycles, T is the temperature, and α, β, γ, and δ are parameters obtained by fitting a large amount of experimental data;

[0050] At the same time, aging and self-discharge related factors are fully introduced. In terms of aging factors, in addition to considering capacity attenuation and internal resistance increase, the influence of factors such as crystal transformation of active materials inside the battery and thickening of SEI film on electrode surface on battery performance are also included to establish a comprehensive aging model. A comprehensive aging index A(t) is defined, which is associated with the aging related parameters of the battery. The battery aging function is: A(t is the comprehensive aging degree, P i (t) represents the i-th internal resistance change rate, w is a parameter obtained by fitting a large amount of data, assuming that the initial value of the battery internal resistance is R0, after t time, the internal resistance becomes R0(t), and the internal resistance change rate is By conducting long-term experimental tests on a large number of batteries of the same type, the internal resistance data at different time points or cycle numbers are collected, and the parameters are obtained by data fitting based on the corresponding battery performance attenuation indicators (such as capacity retention rate, power attenuation, etc.).

[0051] For example, in the study of lithium-ion batteries, internal resistance data is obtained by performing electrochemical impedance spectroscopy tests on batteries at different aging stages, while monitoring battery capacity and other performance parameters. After multiple experiments and data processing, parameter values ​​for a specific type of lithium-ion battery may be obtained, such as w=0.5. In this way, the function can be used to evaluate the degree of battery aging based on the rate of change of internal resistance, providing a basis for battery health management and remaining life prediction.

[0052] In terms of self-discharge factors, a multivariate functional relationship between the self-discharge rate and battery aging, temperature and other factors is established to accurately simulate the self-discharge process and construct a self-discharge function: S d is the self-discharge rate, κ1, κ2, and κ3 are coefficients obtained by fitting experimental data. In actual research, a large number of experiments are required. For example, for the same type of battery, its self-discharge rate is measured at different aging stages (different aging degrees are achieved by controlling the number of cycles or storage time) and under different temperature environments. After sorting these data, the values ​​of the above coefficients are determined using methods such as nonlinear regression analysis, thereby obtaining a specific functional relationship between the self-discharge rate and the battery aging degree and temperature applicable to this type of battery;

[0053] Further establish a high-precision relationship model between parameters and working conditions (temperature, SOC, charge and discharge rate, etc.). Construct the relationship function between ohmic internal resistance and temperature and SOC: A, B, C, and D are parameters obtained by fitting a large amount of experimental data.

[0054] In order to accurately determine the model parameters, a series of experiments with strictly controlled conditions were carried out. In different SOC ranges (such as 0-20%, 20-40%, 40-60%, 60-80%, 80-100%) and temperature ranges (such as -20℃-0℃, 0℃-20℃, 20℃-40℃, 40℃-60℃), charge and discharge tests were carried out at various charge and discharge rates (such as 0.1C, 0.5C, 1C, 2C), and the battery terminal voltage, current, temperature and time data were accurately recorded at regular time intervals (such as 5 seconds). The model parameters were determined by using the rich experimental data collected and combining advanced parameter identification algorithms (such as genetic algorithms, particle swarm optimization algorithms, simulated annealing algorithms, etc.). These algorithms can search for the optimal solution in a complex parameter space and effectively handle the nonlinear characteristics of the model. By real-time monitoring of the working conditions of the battery, the electrochemical model parameters are corrected in real time using the relationship model to ensure that the model can accurately reflect the actual performance of the battery under different working conditions.

[0055] Machine learning model building and training

[0056] Select a machine learning algorithm suitable for the prediction of battery performance from support vector machine (SVM), random forest (RF), neural network (NN), long short-term memory network (LSTM), etc. Build a model structure based on the selected algorithm and determine the number of nodes and connection method of the input layer, hidden layer (if any) and output layer.

[0057] The preprocessed data is used as the input of the machine learning model, and the battery's state of health (SOH) or remaining useful life (RUL) is used as the output to construct a training data set. The cross-validation technique is used to divide the data set into a training set, a validation set, and a test set. The training set is used to train the machine learning model, and the prediction error (such as mean square error, mean absolute error, etc.) is minimized by adjusting the model parameters (such as weights and biases in the neural network). The validation set is used for model selection and hyperparameter tuning to ensure the generalization ability of the model.

[0058] Evaluation result fusion strategy design

[0059] A tandem fusion strategy was adopted to combine the electrochemical model evaluation results and the machine learning model evaluation results.

[0060] Electrochemical model: Based on the principles of physical chemistry, it can accurately describe the charge transfer, lithium ion diffusion, interface reaction and other processes inside the battery. The advantage of the electrochemical model is that it has a strong ability to explain the internal mechanism of the battery and is suitable for performance prediction under steady-state conditions. However, the electrochemical model has strong parameter dependence and limited prediction accuracy under dynamic conditions (such as rapid charging and discharging, temperature changes, etc.).

[0061] Machine learning model: Based on data drive, it can automatically learn the dynamic behavior of the battery through data and has strong adaptability and generalization ability. The advantage of machine learning model is that it can capture the nonlinear behavior of the battery under different working conditions and is suitable for performance prediction in complex environments. However, the lack of understanding of the internal mechanism of the battery may lead to poor interpretability of the prediction results.

[0062] The core idea of ​​the series fusion strategy is to connect the electrochemical model and the machine learning model in series to form a hybrid model. Specifically, the electrochemical model is responsible for providing prior knowledge of the internal state of the battery as the input feature of the machine learning model; the machine learning model is dynamically adjusted according to the actual operating data and outputs the final prediction results. Through this series structure, the physical mechanism advantages of the electrochemical model and the data-driven capabilities of the machine learning model can be fully utilized to improve the prediction accuracy and robustness of the model.

[0063] In the series fusion strategy, the electrochemical model, as the front-end model, first receives the real-time operating data of the battery, such as voltage, current, temperature, etc., and calculates the electrochemical state variables inside the battery, such as the distribution of lithium internal resistance, electrode potential changes, polarization resistance and capacitance, etc., based on its internal electrochemical principles and mathematical models. The state variables output by these electrochemical models are passed to the back-end machine learning model as features. The machine learning model predicts the battery's state of health (SOH), remaining service life (RUL) or other target performance indicators based on these electrochemical features and other related features extracted from historical data (such as the number of battery charge and discharge cycles, the average charge and discharge rate over a period of time, etc.). For example, when predicting the battery SOH, the electrochemical model provides the real physical state information inside the current battery, and the machine learning model uses this information combined with the long-term trends and patterns in the historical data to give a more accurate SOH prediction value.

[0064] In order to adjust the model weight according to the battery usage scenario, it is first necessary to identify and classify the battery usage scenarios. Common second-life battery usage scenarios include:

[0065] As a backup power source in communication base stations: the battery is generally in a stable charge and discharge state, with small changes in current and temperature.

[0066] As a power source in low-speed electric vehicles: the battery is in a state of rapid charging and discharging or large temperature fluctuations, and the current and temperature change dramatically.

[0067] As energy storage batteries in energy storage systems: The secondary batteries in energy storage systems need to have a certain power response capability to quickly track changes in grid load. When there is an instantaneous power fluctuation in the grid (such as a sudden load change caused by the start or stop of large industrial equipment), the battery needs to be able to quickly adjust the charge and discharge power to maintain the stability of the grid.

[0068] In the series fusion model, the outputs of the electrochemical model and the machine learning model can be fused by weighted averaging. The specific weight adjustment mechanism is as follows:

[0069]

[0070] Among them: y is the final prediction result; y ECM is the prediction result of the electrochemical model; y ML is the prediction result of the machine learning model; w1 and w2 are the weights of the electrochemical model and the machine learning model respectively, satisfying w1+w2=1.

[0071] According to the characteristics of different battery usage scenarios, the weights of the electrochemical model and the machine learning model in the fusion process are dynamically adjusted. Specifically:

[0072] The specific method for adjusting the model weight according to the battery usage scenario is:

[0073] When the communication base station is used as a backup power source, the battery is generally in a stable charging and discharging state, and the current and temperature change little. The physical mechanism advantage of the electrochemical model is more obvious. Therefore, the weight of the electrochemical model can be increased and the weight of the machine learning model can be reduced. Set w1=0.7 and w2=0.3.

[0074] When used as a power source in low-speed electric vehicles, the driving conditions of low-speed electric vehicles (such as electric bicycles, small electric vehicles, etc.) are complex and changeable. When driving on urban roads, the vehicle needs to start and stop frequently, and the acceleration and deceleration processes are relatively frequent, which causes the battery load to change dramatically. For example, when starting and accelerating, the battery needs to provide a large current to drive the motor. At this time, the battery is in a high-load discharge state; when braking or coasting, the motor may act as a generator to charge the battery (energy recovery), but the recovered current and power are relatively small. In addition, when the vehicle climbs a slope, it needs to overcome gravity to do work, and the output power demand of the battery will increase; when driving at a constant speed on a flat road, the load is relatively stable. Therefore, in this scenario, the battery is in a state of rapid charging and discharging or large temperature fluctuations, and the current and temperature change dramatically. The machine learning model can better capture the nonlinear behavior of the battery. Therefore, the weight of the machine learning model can be increased, and the weight of the electrochemical model can be reduced. Set w1=0.3, w2=0.7;

[0075] When used as energy storage batteries in energy storage systems, the secondary batteries in the energy storage system need to have a certain power response capability to quickly track changes in grid load. When there is an instantaneous power fluctuation in the grid (such as a sudden load change caused by the start or stop of large industrial equipment), the battery needs to be able to quickly adjust the charge and discharge power to maintain the stability of the grid. Once the internal parameters of the battery change, the prediction accuracy of the electrochemical model may decrease, and the machine learning model can automatically adjust the parameters according to the data. Therefore, the weight of the machine learning model can be appropriately increased, setting w1 = 0.4, w2 = 0.6.

[0076] Example 1: Evaluation of second-life batteries in energy storage systems

[0077] In a typical energy storage system application scenario, a batch of retired lithium-ion power batteries are selected for cascade utilization. First, according to the above data collection and preprocessing method, high-precision sensors and data acquisition modules are integrated into the battery management unit (BMS) of the energy storage system, and the data collection frequency is set to once every 10 seconds to ensure that the performance changes of the battery during the charging and discharging process can be accurately captured.

[0078] According to the operating temperature range of the energy storage system (usually 10℃-35℃) and the charge and discharge rate requirements (generally between 0.2C-1C), the improved Thevenin model is parameter identified and optimized. In the laboratory, professional battery testing equipment is used to conduct deep charge and discharge tests on the cascade battery at three temperature points of 10℃, 20℃, and 30℃ and three charge and discharge rates of 0.2C, 0.5C, and 1C. Each test condition is repeated 5 times to collect a large amount of experimental data. The particle swarm optimization algorithm is used to identify the model parameters, and the model parameters are determined after multiple iterative optimizations. Finally, the parameter equation of the ohmic internal resistance at different temperatures and SOC is obtained as follows:

[0079] The long short-term memory network (LSTM) was selected as the machine learning model. The characteristics of the lithium ion concentration change rate and electrode potential difference output by the electrochemical model were taken together with the preprocessed historical data (including the average charge and discharge rate, maximum and minimum voltage difference, etc. of the past 100 charge and discharge cycles) as input to construct a training data set. The cross-validation technique was used to divide the data set into a training set, a validation set, and a test set in a ratio of 7:2:1. The LSTM model was trained using the training set, the number of hidden layer nodes was set to 64, the initial value of the learning rate was 0.001, and the Adam optimizer was used for parameter optimization. After 500 training cycles, the validation set was used for model selection and hyperparameter tuning, and the optimal model structure and parameters were finally determined.

[0080] The battery health SOH is calculated by the parameter equation of ohmic internal resistance at different temperatures and SOCs, and then the battery SOH is predicted by the machine learning model. The final battery prediction structure is obtained by the fusion model. In the model fusion process, y is the final prediction result; y ECM is the prediction result of the electrochemical model; y ML is the prediction result of the machine learning model; w1 and w2 are the weights of the electrochemical model and the machine learning model, respectively, satisfying w1+w2=1. According to the stable operation characteristics of the energy storage system, the initial weights are set to 0.4 for the electrochemical model and 0.6 for the LSTM model.

[0081] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. An intelligent evaluation method for second-life batteries based on an electrochemical model, characterized in that: The following steps are involved: S1, collects the operating data of the second-life battery under different working conditions, cleans, processes outliers and normalizes the collected data to obtain pre-processed data; S2, based on the Thevenin model, according to the characteristics of the cascade battery, introducing aging and self-discharge related factors to establish an electrochemical model, determining the electrochemical model parameters through the preprocessed data, and obtaining the battery evaluation results through the electrochemical model; S3, select the appropriate machine learning algorithm to build and the machine learning model, take the preprocessed data features as input, and the battery health status as output, build a training data set, train the model through cross-validation, select the best model structure, and obtain the battery evaluation results through the machine learning model; S4 uses a series fusion strategy to fuse the calculation results of the electrochemical model and the prediction results of the machine learning model, and adjusts the fusion weight according to the battery usage scenario to finally obtain the final evaluation result of the battery.

2. The evaluation method according to claim 1, characterized in that: The evaluation result is the battery's state of health (SOH).

3. The evaluation method according to claim 1, characterized in that: In step S2, the electrochemical model is specifically: Internal resistance change function: R0(t) = α·e βN +γ·T+δ, where R0(t) represents the ohmic internal resistance that changes with time, N is the number of charge and discharge cycles, T is the temperature, and α, β, γ, and δ are parameters obtained by fitting a large amount of experimental data; Battery aging function: A(t) is the comprehensive aging degree, P i (t) represents the i-th internal resistance change rate, w is a parameter obtained by fitting a large amount of data; Self-discharge function: S d is the self-discharge rate, κ1, κ2, κ3 are coefficients obtained by fitting the experimental data; The relationship function between ohmic internal resistance, temperature and SOC: a, B, C, and D are parameters obtained by fitting a large amount of experimental data.

4. The evaluation method according to claim 1, characterized in that: In step S3, the machine learning algorithm includes one or more of a support vector machine, a random forest, a neural network, or a long short-term memory network.

5. The evaluation method according to claim 1, characterized in that: In step S4, the outputs of the electrochemical model and the machine learning model can be fused by weighted averaging. The specific weight adjustment mechanism is as follows: Among them: y is the final prediction result; y ECM is the prediction result of the electrochemical model; y ML is the prediction result of the machine learning model; w1 and w2 are the weights of the electrochemical model and the machine learning model respectively, satisfying w1+w2=1.

6. The evaluation method according to claim 1, characterized in that: In step S4, the model weight is adjusted according to the battery usage scenario: As a backup power source in communication base stations, the physical mechanism advantage of the electrochemical model is more obvious, so the weight of the electrochemical model can be increased and the weight of the machine learning model can be reduced, setting w1=0.7 and w2=0.3; As a power source in low-speed electric vehicles, the machine learning model can better capture the nonlinear behavior of the battery, so the weight of the machine learning model can be increased and the weight of the electrochemical model can be reduced, setting w1 = 0.3 and w2 = 0.7; As energy storage batteries in the energy storage system, the secondary batteries in the energy storage system need to have a certain power response capability to quickly track changes in grid load. The machine learning model can automatically adjust parameters according to the data, so the weight of the machine learning model can be appropriately increased, setting w1=0.4, w2=0.6.

Images