Battery health state estimation method based on end-cloud cooperation strategy

Abstract

The invention relates to the field of lithium ion batteries, in particular to a battery health state estimation method based on an end-cloud cooperation strategy, which comprises the following steps: extracting battery aging characteristics; data preprocessing, including data normalization and time series data construction; constructing a battery prediction model, including constructing and training a neural network model at a cloud side and constructing a battery aging empirical model at an end side; and setting evaluation indexes. According to the method, a battery SOH estimation-oriented end-cloud collaborative hybrid prediction network is constructed, different battery SOH prediction models are deployed on the cloud side and the end side of the network respectively, SOH prediction is carried out by utilizing battery aging characteristics and mathematical relationships respectively, individual differences of battery aging and general laws of battery aging can be comprehensively considered, and the prediction accuracy is improved. Meanwhile, the model calculation speed is increased, and good precision and robustness are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium-ion battery management, and in particular to a battery health status estimation method based on a terminal-cloud collaboration strategy. Background Art

[0002] In recent years, with the rapid development of the new energy vehicle industry, electric vehicles, as the main type of new energy vehicles, have become the focus of industry attention for their power batteries, lithium-ion batteries. The performance of lithium-ion batteries will gradually decline during the use of electric vehicles. When the battery life reaches a certain value, continued use may bring safety hazards or even cause serious accidents. Therefore, accurate prediction of the health status of lithium-ion batteries can effectively avoid such accidents, reduce system management costs, and improve the safety and stability of the battery system.

[0003] At present, the prediction methods for battery health status estimation can be divided into four categories: experimental-based methods, model-based methods, data-driven methods and signal analysis-based methods.

[0004] The experimental method requires a large number of experiments to determine the relationship between battery health indicators (such as capacity, internal resistance) and external factors (such as current, temperature). This method can be divided into calendar aging model and cycle aging model according to the different battery aging pathways. The calendar aging model mainly reflects the aging phenomenon caused by the influence of factors such as ambient temperature over a long period of time. The cycle aging model involves complex external environmental factors, usually considering current, temperature, DOD (Depth of Discharge) and charge and discharge cycle range. Some studies have coupled the impedance-based electrothermal model with the calendar aging model to obtain a complete battery life model to simulate battery aging behavior. The principle of the experimental method is intuitive and easy to understand. However, in practical applications, the battery usage environment is extremely complex and changeable, and there are significant differences in the characteristics of different batteries, resulting in different aging patterns. In addition, the battery aging process is often not linear, but is affected by a variety of nonlinear factors, which limits the effectiveness of models based on experimental methods in practical applications, making it difficult for these models to make accurate predictions of the battery's health status.

[0005] Model-based methods predict battery SOH by establishing equivalent battery models, such as electrochemical model methods and equivalent circuit models. Electrochemical models can describe the internal and external characteristics of the battery, among which the single particle model (SPM) has a lower computational cost than the pseudo two-dimensional model (P2D), but is lower than the latter in terms of prediction accuracy. The equivalent circuit model establishes a mathematical model based on the electrical characteristics of the battery, reflecting the health status of the battery through the changes in model parameters. These models usually include one or more resistors, capacitors, and inductors, which simulate the dynamic response and internal electrochemical reactions of the battery. Model-based methods are valued for their mathematical simplicity and good simulation capabilities of battery dynamic behavior. The advantage of this method is that it can provide a deep understanding of battery behavior. However, its effectiveness is highly dependent on the accurate identification of model parameters and sufficient data support. Building an accurate battery model is a challenging task that usually requires a lot of time and computing resources. Therefore, the accuracy and practicality of these physical models are often difficult to guarantee.

[0006] The data-driven SOH estimation method evaluates the health status of the battery by utilizing the real-time data or historical data of the battery. This method collects the data of the battery during the aging process and extracts features from it as the input database. Then, a machine learning algorithm is used to establish a SOH estimation model, and the battery SOH is estimated through the model, and the trend of battery health status changes is predicted. The data-driven method mainly includes support vector machine regression, Gaussian process regression, extreme learning machine, and long short-term memory network. The data-driven method does not rely on the physical and chemical model of the battery, but directly uses historical or real-time data such as battery voltage, current, and temperature to predict SOH. This method is favored for its flexibility and wide applicability. However, its prediction effect depends largely on the quality and relevance of the input data. The correlation between the input data and the battery capacity is crucial to establishing an effective mapping relationship. Therefore, how to optimize the input database has become a difficulty in current research.

[0007] The signal analysis-based method uses electrochemical analysis technology and signal processing technology to extract battery degradation characteristics with certain signal characteristics during the battery aging process, and conducts battery aging analysis based on these characteristics. Common characteristic signal analysis methods include electrochemical impedance spectroscopy, slow cyclic voltammetry, differential voltage analysis, incremental capacity analysis, etc. These are all in-situ non-destructive characterization methods. These methods draw curves by secondary processing of easily measured signals such as voltage, current, and temperature. The peak and valley values ​​and positions of these curves contain information about lithium ions participating in electrochemical reactions of the battery under different aging conditions. These methods can extract reliable battery degradation characteristics without destructive testing of the battery, and are a fast electrochemical characterization method that has both engineering practicality and theoretical basis. The signal analysis-based method does not require in-depth analysis of the complex reaction process inside the lithium-ion battery, but processes and draws curves of signals that are easy to measure during the battery aging process, extracts features through these curves and uses them for information prediction. This method is popular because of its good nonlinear characteristics and simple data acquisition process. However, the generalization ability of this method is relatively weak, and high-quality training data is required to ensure that the trained model has sufficient robustness and stability. Summary of the invention

[0008] To achieve the above object, the present invention provides the following technical solutions: A battery health state estimation method based on a terminal-cloud collaborative strategy, comprising: Step 1: Extract battery aging features; Step 2: Data preprocessing, including data normalization and time series data construction; Step 3: Build a battery prediction model, including building and training a neural network model on the cloud side and building an empirical model of battery aging on the device side. Step 4: Set evaluation indicators.

[0009] As a further solution of the present invention: In step 1, a lithium-ion battery charge and discharge experimental condition is set to charge and discharge the battery. This experiment is verified on an open source data set, and the data set used in the algorithm comes from the Center of Excellence in Prediction of NASA Ames Research Center. The entire data set is synthesized by a dynamic load curve consisting of different battery cycles and randomly selected charging and discharging currents. The data set contains experimental data of various batteries under charging, discharging and electrochemical impedance spectroscopy measurement conditions. Therefore, the experimental data of No. 5 battery, No. 6 battery, No. 7 battery and No. 18 battery under discharge conditions were selected for research in the experiment. The lithium-ion battery tested is a commercially available 18650-sized lithium-ion battery produced by the Idaho National Laboratory. The tested battery was subjected to degradation experiments in constant current constant voltage charging (CC-CV) mode and constant current discharge at room temperature (24°C). During the constant current charging process, the charging current was maintained at 1.5A, and the charging cutoff voltage of the four batteries was 4.2V. Afterwards, during constant voltage charging, when the charging current was lower than 20mA, the charging process stopped. The charging process of the four batteries is consistent. After charging, the four batteries are discharged at a current of 2A until the voltages of battery No. 5, battery No. 6, battery No. 7, and battery No. 18 reach the preset values ​​of 2.7V, 2.5V, 2.2V, and 2.5V, respectively. Under CC conditions, the discharge time within the same working end voltage range decreases continuously, which can be used to characterize the battery aging process. The present invention records the discharge time of the battery between the voltage range [3.4V, 3.8V] as one of the alternative features of battery aging.

[0010] As a common characteristic analysis method to describe the battery degradation process, the IC curve can characterize the battery degradation from the electrode level and has a high resolution for the battery charge and discharge platform area. The IC curve is defined as the battery capacity increment of continuous voltage increment, and its mathematical expression is: ; Where q, I, V, and t are discharge capacity, current, voltage, and time, respectively.

[0011] In order to remove noise interference, the present invention uses the SG filtering method to filter the curve. The SG filter has excellent filtering performance in high-frequency digital signal filtering and is a local polynomial regression technology. The main advantage of the SG filter is that in the process of removing noise, it can effectively keep the basic form and key features of the signal unchanged, so it is very suitable for filtering the IC curve and can ensure that the filtered curve retains important data features. The structure of the SG filter is described as follows: ; Where p(i+j) is the original input signal, y(i) is the resulting output signal, and F jis the coefficient given by the SG filter, N is the number of convolution integers, which is equal to the smoothing window size .

[0012] The IC curve of the battery of the present invention has a significant peak value throughout the degradation process, and as the battery degrades, the peak value continues to decrease and move to the left. These phenomena indicate that the peak value of the IC curve can be used as a feature to characterize battery degradation. Therefore, the present invention uses the IC peak value and its corresponding battery measurement voltage as alternative aging features.

[0013] As a further solution of the present invention: Step 1 In order to comprehensively and quantitatively study the influence of characteristic variables, reduce computational costs, and improve the robustness of the model, Pearson correlation analysis is used to quantitatively evaluate the correlation between the extracted characteristic variables and SOH, and the characteristic variable values ​​with a high degree of correlation are screened out as the input of the prediction model. The specific formula for calculating the correlation coefficient is as follows: ; Where n is the number of sample sequences, , Each represents a variable. , Respectively represent , This method calculates the correlation coefficient of the input data and obtains a value between -1 and 1. According to the different correlation coefficients, the degree of correlation between the two variables can be quantified intuitively.

[0014] As a further solution of the present invention: In step 2, data normalization adopts the maximum and minimum normalization method to convert the data values ​​into the interval [0,1]. This process includes finding the maximum and minimum values ​​in the data set, and adjusting each data point according to these extreme values ​​to maintain their relative position in the overall data set. This function of the maximum and minimum normalization method helps to maintain the scale consistency of the data in subsequent processing, which is of great significance for improving the efficiency of model training and prediction accuracy. The normalization process for each data variable is as follows: ; in, is the normalized value, w min is the minimum value in the sequence, w max is the maximum value in the sequence, w is the original data, and in the present invention, the three extracted battery characteristics and the battery capacity in each cycle are taken respectively.

[0015] As a further solution of the present invention: in step 2, time series data is constructed. First, the batch size is selected, that is, the size of the data contained in each batch after all the data are divided into a certain number of batches. Taking battery No. 5 as an example, the batch size is selected as 16, that is, the first batch contains the battery aging characteristics from the 1st time point to the 16th time point, the second batch contains the battery aging characteristics from the 17th time point to the 32nd time point, and so on. Since the data set size of battery No. 5 has a total of 168 battery cycles, the last batch contains the battery aging characteristics from the 161st time point to the 168th time point. After the batches are divided, the windows are divided. The first window contains the battery aging characteristics from the 1st time point to the 2nd time point. The information in this window is used to predict the battery SOH value at the 3rd time point after neural network learning. The remaining window sizes are all 2 and slide backward with a step size of 1. Time series analysis can help identify the trend of data changes over time, provide historical information, and help identify the causal relationship between variables, which is crucial for predicting future changes.

[0016] As a further solution of the present invention: In step 3, on the terminal side, in order to more accurately and quickly predict the aging behavior of the battery, the present invention adopts a battery aging empirical model that combines exponential decay and the Arrhenius equation. This model can simultaneously capture the nonlinear decay characteristics of the battery aging process and the effect of temperature on the aging rate, providing a more comprehensive aging prediction framework. The model consists of two parts, one is a capacity decay model based on the Arrhenius equation, and the other is an internal resistance growth model based on double exponential decay.

[0017] The Arrhenius equation is described as follows: ; Among them, Q represents the battery capacity decay rate, is the battery activation energy, is the gas constant, is temperature, is the parameter to be fitted. This equation describes how temperature affects the battery aging rate, where the activation energy reflects the difficulty of battery aging reaction, and the temperature The exponential term indicates the acceleration or retardation of the aging rate by temperature.

[0018] The exponential decay function is described as follows: ; in, To consider the capacity when only the number of cycles is considered, is the parameter to be fitted, and n is the number of cycles. This equation simulates the change of battery capacity with the number of cycles, where and represent the initial value and final stable value of capacity, respectively, and and It describes the rate and trend of capacity change.

[0019] The battery empirical model used in the present invention combines exponential decay and the Arrhenius equation, which is specifically described as follows: ; ; in, is a parameter to be determined, To consider the capacity when the number of cycles is considered, is the battery activation energy, is the gas constant, is the temperature, n is the number of cycles, is the internal resistance when the number of cycles is considered. Equation (7) combines the exponential decay and Arrhenius equation to describe the change in battery capacity. This equation takes into account the effects of temperature, number of cycles, activation energy, and battery internal resistance on battery capacity. Equation (8) is constructed based on the exponential decay model and describes the change in battery internal resistance with the number of cycles.

[0020] The advantage of this model lies in its accuracy and applicability. It can accurately predict the aging behavior of batteries under different temperatures and cycle conditions, and the flexibility of the model enables it to adapt to different battery types and operating conditions. In addition, this model reduces experimental time and cost, and improves research efficiency by replacing a large number of repeated experiments with model predictions. The basic flow chart for establishing the model on the end side is as follows: Figure 2 shown.

[0021] As a further solution of the present invention: In step 3, on the cloud side, an LSTM neural network is constructed to predict the battery SOH. The LSTM network can solve the gradient problem and the long-term dependency problem well. LSTM contains three different types of gate structures, namely input gate, forget gate and output gate. The first step of LSTM is to use the forget gate to determine which information in the cell state is discarded and retained when the LSTM cell obtains data. The forget gate can be expressed as: ; in is the output of the forget gate, is the weight matrix, is the bias vector, is the hidden state of the neuron at the previous time step, is the input of the LSTM network at the current time step, is the sigmoid function.

[0022] In the second step, we decide what new information to store in the cell. This is done in two parts. First, we update the information in the cell through the input gate. Then, we use the tanh layer to get the new information. In the next step, we combine these two pieces of information to update the state. These two steps can be expressed as: ; ; in and is the weight matrix, and is the bias vector, is the output of the input gate at the current time step, is the newly generated memory state of the neuron, is the hyperbolic tangent function.

[0023] The third step is to integrate the data from the forget gate and the input gate. Multiply by the output of the forget gate , represents the process of partially discarding data, and then adding the output of the input gate Multiply by the new memory state of the neuron , changing the memory state of the neuron from Updated to , It will be passed to the next neuron, and the specific calculation process can be expressed as: , Finally, the output part is determined. The output gate is based on the unit state. , the previous cycle output and current input Generate a new hidden state The output gate can be expressed as: ; ; in and The values ​​are the weight matrix and bias vector, is the output of the output gate at the current time step, is the hidden state of the neuron at the current time step and is also the output of the current time step. Will make peace Pass it on to the next neuron.

[0024] The present invention uses the RMSprop optimization algorithm and the Dropout technology to train the LSTM network. The Dropout technology is used in the present invention to prevent neurons in the neural network from being overly dependent on the outputs of other neurons. This technology randomly "discards" (i.e., temporarily removes) a portion of neurons during the training process, so that each neuron must rely on the group behavior of the entire network rather than the output of a single neuron. Specifically, when the neural network is forward propagating, each neuron has a certain probability of being temporarily disabled, so that the disabled neurons will not have any effect on the output of the network. This mechanism in which some neurons are randomly disabled is essentially to create multiple different "scaled versions" of networks based on the original network. The model complexity of these networks is relatively low because they contain fewer neurons. In this way, the Dropout technology helps to reduce the complexity of the entire model. When dealing with overfitting problems, Dropout speeds up the convergence of the model by reducing the number of parameters involved in training, thereby improving the generalization ability of the model. The schematic diagram of the Dropout principle is shown in the figure. Figure 3 shown. Figure 3 The neurons that are hidden during the information propagation process are marked with crosses, and these neurons are temporarily removed, thereby simplifying the model complexity. The cloud SOH prediction model flow chart is as follows Figure 4 As shown, first obtain the corresponding battery aging characteristics from the battery aging data set and organize the data into time series data that conforms to the neural network input. Build a neural network based on LSTM on the cloud side, and use the powerful computing power of the cloud side to train and obtain the optimal network parameters. Build an empirical model of battery aging on the end side. The empirical model used in the present invention combines exponential decay and Arrhenius equations. This equation takes into account the effects of temperature, number of cycles, activation energy, and battery internal resistance on battery capacity, and can quickly derive the battery SOH estimate. The SOH results predicted by the two parts of the model are weighted and summed to obtain the final SOH prediction value.

[0025] As a further solution of the present invention: In step 4, two evaluation indicators are set to evaluate the performance of the model, namely, Mean Absolute Error (MAE) and Root Mean Square Error (RMSE): , ; in for The true value of SOH at the moment, for The SOH forecast value at the moment. MAE is the average of the absolute values ​​of the differences between the predicted values ​​and the actual values. It gives the same weight to all errors, the result is easy to understand, and it is insensitive to outliers. RMSE is the square root of the average of the squares of the differences between the predicted values ​​and the actual values. It gives greater weight to larger errors. Because the errors are squared, larger errors have a greater impact on RMSE. This makes RMSE more sensitive to outliers. The smaller its value, the better the prediction effect of the entire sequence.

[0026] In summary, the present invention proposes an innovative battery SOH estimation method, which is implemented by deploying different types of models on the cloud and the end side. Specifically, a relatively complex neural network model is deployed on the cloud, while an empirical battery aging model is deployed on the end side. This strategy makes full use of the powerful computing resources of the cloud to train complex neural networks, and uses the computing power of the end side to quickly perform predictions based on mathematical relationships, thereby significantly improving the computing efficiency of the entire system. The neural network model on the cloud can deeply analyze and learn the historical aging characteristics of specific batteries and capture the complex and nonlinear patterns in the battery aging process. These models can identify specific trends and behaviors of battery aging through a large amount of data training, thereby providing rich information for SOH estimation. On the other hand, the empirical model on the end side is based on the general laws of battery aging and quickly obtains prediction results through mathematical relationships. Combining these two models can not only ensure the accuracy of SOH estimation, but also enhance the robustness of the model. The deep learning ability of the neural network model enables the prediction results to adapt to the individual differences in battery aging, while the versatility of the empirical model ensures the applicability under different batteries and different usage conditions. This fusion strategy improves the stability and reliability of the model while maintaining high accuracy.

[0027] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention constructs a hybrid prediction network for end-cloud collaboration for battery SOH estimation. The network deploys different battery SOH prediction models on the cloud side and the end side, respectively, and uses battery aging characteristics and mathematical relationships to perform SOH prediction.

[0028] 2. The present invention constructs an empirical model of battery aging for predicting battery SOH, which combines the exponential decay model and the Arrhenius equation and can quickly predict changes in battery capacity and internal resistance.

[0029] 3. The present invention performs weighted fusion of the neural network model on the cloud side and the battery empirical model on the terminal side, combining the powerful computing power of the cloud side and the fast and simple advantages of the battery aging empirical model, and can accurately predict the battery SOH.

[0030] 4. The present invention sets two indicators, the error value and the root mean square error value, to evaluate the performance of the model from a local and overall perspective respectively. By complementing each other with these two indicators, the performance of the model can be evaluated more comprehensively. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 This is a flow chart of a battery health status estimation method based on a terminal-cloud collaboration strategy in an embodiment of the present invention.

[0032] Figure 2 A flowchart for constructing an end-side battery aging empirical model in a battery health status estimation method based on an end-cloud collaborative strategy in an embodiment of the present invention.

[0033] Figure 3 The figure is a schematic diagram of the Dropout principle in a battery health status estimation method based on a terminal-cloud collaboration strategy in an embodiment of the present invention.

[0034] Figure 4 The present invention provides a flowchart for constructing a cloud-based SOH prediction model in a battery health status estimation method based on a terminal-cloud collaborative strategy in an embodiment of the present invention.

[0035] Figure 5 The present invention provides a flowchart of the overall model construction and prediction in a battery health status estimation method based on a terminal-cloud collaborative strategy in an embodiment of the present invention.

[0036] Figure 6 This is a diagram of the prediction results of four batteries in a battery health status estimation method based on a terminal-cloud collaboration strategy in an embodiment of the present invention.

[0037] Figure 7 Graphs showing four battery prediction error results in a battery health status estimation method based on a terminal-cloud collaborative strategy in an embodiment of the present invention. DETAILED DESCRIPTION

[0038] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0039] In the field of battery SOH prediction, the use of historical data is crucial. These data not only contain the performance changes of the battery under different usage conditions, but also reflect the trends and patterns of battery aging. Deploying the prediction model to the cloud side can make full use of the powerful computing power of cloud computing, thereby significantly improving the model's computing speed and ability to process large-scale data sets. The model deployed on the cloud side is able to process a large amount of historical data, which usually includes key parameters such as the number of battery charge and discharge cycles, voltage, current, and temperature. Through deep learning methods, such as long short-term memory recurrent neural networks, the model can learn the battery's life decay process, thereby accurately estimating the battery's SOH. The computing resources on the cloud side allow the model to be trained on large-scale data sets, which is difficult to achieve on traditional end-side devices because of the limited computing power and storage capacity of the end-side devices. The present invention is a battery health state estimation method based on an end-cloud collaborative strategy, which combines the end-side empirical model of battery aging and the cloud-side trained data-driven model to comprehensively evaluate the battery state of health (SOH). In this hybrid model strategy, the cloud-side neural network data-driven model uses a large amount of historical data and powerful computing power to train a model that can accurately predict the battery aging trend through a deep learning algorithm. The battery discharge time within a fixed voltage range is measured during the battery aging process, and the peak value of the battery IC curve and its corresponding measured voltage are recorded as candidate features. After correlation analysis, it is shown that these three features can be used as battery aging features. The battery aging characteristics are used as features and the battery SOH value is used as a label to construct time series data, which is input into the cloud-side neural network for training and optimization. This model can capture the nonlinear and complex patterns in the aging process of individual batteries. The end-side empirical model quickly provides a preliminary SOH estimate based on simple mathematical relationships and real-time battery operation data. The prediction of the empirical model can take into account the general laws of battery aging. Finally, the estimation results of the two models are combined in a weighted manner to obtain the final SOH estimation result. This hybrid model has strong accuracy and robustness, can reduce computing costs, and the architecture allows flexible allocation of tasks between the cloud and the end, which facilitates model expansion and maintenance.

[0040] The specific implementation of the present invention is described in detail below in conjunction with specific embodiments.

[0041] Embodiment 1, a method for estimating a battery health state based on a terminal-cloud collaboration strategy provided by an embodiment of the present invention, comprises the following steps: Step 1: Battery aging feature extraction The lithium-ion battery charge and discharge experimental conditions were set to charge and discharge the battery. This experiment was verified on an open source data set. The data set used in the algorithm comes from the Center of Excellence in Prediction at NASA Ames Research Center. The entire data set is synthesized by a dynamic load curve consisting of different battery cycles and randomly selected charge and discharge currents. The data set contains experimental data of various batteries under charging, discharging and electrochemical impedance spectroscopy measurement conditions. Therefore, the experimental data of No. 5, No. 6, No. 7 and No. 18 batteries under discharge conditions were selected for research. The lithium-ion battery tested is a commercially available 18650-size lithium-ion battery produced by the Idaho National Laboratory. The tested battery was subjected to degradation experiments in constant current constant voltage charging (CC-CV) mode and constant current discharge at room temperature (24°C). During the constant current charging process, the charging current was maintained at 1.5A, and the charging cutoff voltage of the four batteries was 4.2V. Afterwards, during the constant voltage charging, when the charging current was lower than 20mA, the charging process stopped. The charging process of the four batteries is consistent. After charging, the four batteries are discharged at a current of 2A until the voltages of battery No. 5, battery No. 6, battery No. 7 and battery No. 18 reach the preset values ​​of 2.7V, 2.5V, 2.2V and 2.5V respectively.

[0042] Under CC conditions, the discharge time within the same working terminal voltage range decreases continuously, which can be used to characterize the battery aging process. The present invention records the discharge time of the battery between the voltage range [3.4V, 3.8V] as one of the candidate features of battery aging. At the same time, the IC curve of the battery is drawn. As a common feature analysis method for describing the battery degradation process, the IC curve can characterize the battery degradation from the electrode level and has a high resolution for the battery charge and discharge platform area.

[0043] In order to remove noise interference, the present invention adopts the SG filtering method to filter the curve. The SG filter has excellent filtering performance in high-frequency digital signal filtering and is a local polynomial regression technology. The main advantage of the SG filter is that in the process of removing noise, it can effectively keep the basic form and key features of the signal unchanged, so it is very suitable for filtering the IC curve and can ensure that the filtered curve retains important data features.

[0044] The IC curve of the battery of the present invention has a significant peak value throughout the degradation process, and as the battery degrades, the peak value continues to decrease and move to the left. These phenomena indicate that the peak value of the IC curve can be used as a feature to characterize battery degradation. Therefore, the present invention uses the IC peak value and its corresponding battery measurement voltage as alternative aging features.

[0045] In order to comprehensively and quantitatively study the influence of characteristic variables, reduce computational costs, and improve the robustness of the model, Pearson correlation analysis is used to quantitatively evaluate the correlation between the extracted characteristic variables and SOH, and select characteristic variable values ​​with higher correlation as prediction model input. The correlation coefficients between the three characteristics and the battery SOH in the present invention are shown in Table 1. The battery discharge time, IC curve peak value, and IC curve peak voltage are represented by F1, F2, and F3, respectively. It can be seen that the correlation coefficients between the three characteristics and the battery SOH are high, so the three characteristics can be used as battery aging characteristics.

[0046] Table 1. Schematic diagram of the correlation coefficients of the six extracted features

[0047] Step 2: Data Preprocessing In the process of neural network prediction on the cloud side, the network input data needs to be further processed.

[0048] The present invention relates to a data normalization method, which aims to perform normalization processing on collected raw data. Specifically, a maximum and minimum normalization method is adopted to convert data values ​​into the interval [0, 1].

[0049] After normalizing the data, the time series data is constructed. First, the batch size is selected, that is, the amount of data contained in each batch after all the data are divided into a certain number of batches. In this experiment, the batch size is selected as 16, that is, the first batch contains the battery aging features from the 1st time point to the 16th time point, the second batch contains the battery aging features from the 17th time point to the 32nd time point, and so on. Since the data set size has a total of 168 battery cycles, the last batch contains the battery aging features from the 161st time point to the 168th time point. After the batches are divided, the windows are divided. The first window contains the battery aging features from the 1st time point to the 2nd time point. The information in this window is used to predict the battery SOH value at the 3rd time point after neural network learning. The remaining windows are all 2 in size and slide backward with a step size of 1. Time series analysis can help identify the trend of data changes over time, provide historical information, and help identify the causal relationship between variables, which is crucial for predicting future changes.

[0050] Step 3: Build a battery prediction model The hybrid model of the present invention includes two prediction models on the cloud side and the terminal side. A neural network model is constructed and trained on the cloud side for SOH prediction, and an empirical model of battery aging is constructed on the terminal side for prediction.

[0051] On the end side, in order to more accurately and quickly predict the aging behavior of the battery, the present invention adopts an empirical battery aging model that combines exponential decay and the Arrhenius equation. This model can simultaneously capture the nonlinear decay characteristics of the battery aging process and the effect of temperature on the aging rate, providing a more comprehensive aging prediction framework. The model consists of two parts, one is a capacity decay model based on the Arrhenius equation, and the other is an internal resistance growth model based on double exponential decay.

[0052] The advantages of this model are its accuracy and applicability. It can accurately predict the aging behavior of batteries under different temperatures and cycle conditions, and the flexibility of the model enables it to adapt to different battery types and working conditions. In addition, this model reduces experimental time and cost, and improves research efficiency by replacing a large number of repeated experiments with model predictions.

[0053] On the cloud side, an LSTM neural network is constructed to predict the battery SOH. The LSTM network can solve the gradient problem and the long-term dependency problem well. The present invention uses the RMSprop optimization algorithm and the dropout technology to train the LSTM network. In the present invention, the dropout technology is used to prevent the neurons in the neural network from being overly dependent on the output of other neurons. This technology randomly "discards" (i.e., temporarily removes) a part of the neurons during the training process, so that each neuron must rely on the group behavior of the entire network rather than the output of a single neuron. Specifically, when the neural network is forward propagated, each neuron has a certain probability of being temporarily disabled, so that the disabled neurons will not have any effect on the output of the network. This mechanism in which some neurons are randomly disabled is essentially to create multiple different "scaled" networks based on the original network, and the model complexity of these networks is relatively low because they contain a small number of neurons. In this way, the dropout technology helps to reduce the complexity of the entire model. When dealing with overfitting problems, dropout speeds up the convergence of the model by reducing the number of parameters involved in training, thereby improving the generalization ability of the model.

[0054] Finally, the prediction results of the two models are weighted and summed. The weights of the SOH prediction values ​​of the cloud-side and terminal-side prediction models in the present invention are 0.7 and 0.3, respectively.

[0055] Step 4: Evaluation indicator setting process.

[0056] In this step, two evaluation indicators are set to evaluate the performance of the model, namely the mean absolute error and the root mean square error. MAE is the average of the absolute values ​​of the differences between the predicted values ​​and the actual values. It gives the same weight to all errors, the results are easy to understand, and it is insensitive to outliers. RMSE is the square root of the average of the squares of the differences between the predicted values ​​and the actual values. It gives greater weight to larger errors. Because the errors are squared, larger errors have a greater impact on RMSE. This makes RMSE more sensitive to outliers. The smaller its value, the better the prediction effect of the entire sequence.

[0057] The first 50% of the data set is used as the training set, and the rest is used as the test set for model training and verification. Finally, the error value and root mean square error value are used as the result evaluation indicators. The prediction results and error results of the four batteries are shown in Figure 6 shown. Figure 7 The error results of four types of batteries are shown, and the prediction errors of the four types of batteries are all below 1%.

[0058] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. A battery health status estimation method based on a terminal-cloud collaboration strategy, characterized in that: The following steps are involved: Step 1: Extract battery aging features; Step 2: Data preprocessing, including data normalization and time series data construction; Step 3: Build a battery prediction model, including building and training a neural network model on the cloud side and building an empirical model of battery aging on the device side. Step 4: Set evaluation indicators.

2. According to claim 1, a method for estimating battery health status based on a terminal-cloud collaborative strategy is characterized in that: In step 1, the battery aging feature extraction is to extract the battery aging feature using the IC curve, and the mathematical representation of the IC curve is: ； In the formula, q, I, V, and t are discharge capacity, current, voltage, and time, respectively. The SG filtering method is used to filter the IC curve. The construction of the SG filter is described as follows: ； Where p(i+j) is the original input signal, y(i) is the resulting output signal, and F j is the coefficient given by the SG filter, N is the number of convolution integers, which is equal to the smoothing window size .

3. The method for estimating the battery health status based on the end-cloud collaborative strategy according to claim 2 is characterized in that: In step 1, Pearson correlation analysis is used to quantitatively evaluate the correlation between the extracted characteristic variables and SOH, and characteristic variable values ​​with a high degree of correlation are screened out as inputs to the prediction model. The specific formula for calculating the correlation coefficient is as follows: ； Where n is the number of sample sequences, , Each represents a variable. , Respectively represent , The mean of .

4. The method for estimating the battery health status based on the end-cloud collaborative strategy according to claim 1 is characterized in that: The data normalization in step 2 adopts the maximum and minimum normalization technology. The normalization process for each data variable is as follows: ； in, is the normalized value, w min is the minimum value in the sequence, w max is the maximum value in the sequence, w is the original data, and in the present invention, the three extracted battery characteristics and the battery capacity in each cycle are taken respectively.

5. The method for estimating the battery health status based on the end-cloud collaborative strategy according to claim 4 is characterized in that: The steps of constructing the time series data in step 2 are as follows: the normalized data set is batched and windowed to obtain several data batches, and are input into the neural network one by one.

6. The method for estimating battery health status based on a terminal-cloud collaborative strategy according to claim 1, characterized in that: The steps of building and training the neural network model on the cloud side in step 3 are as follows: The battery empirical model consists of two parts, one is a capacity decay model based on the Arrhenius equation, and the other is an internal resistance growth model based on double exponential decay. The description of the Arrhenius equation is as follows: ； Among them, Q represents the battery capacity decay rate, is the battery activation energy, is the gas constant, is temperature, is the parameter to be fitted; the exponential decay function is described as follows: ； in, To consider the capacity when only the number of cycles is considered, is the parameter to be fitted, n is the number of cycles, and the comprehensive battery empirical model is described as follows: ； ； in, is a parameter to be determined, To consider the capacity when the number of cycles is considered, is the battery activation energy, is the gas constant, is the temperature, n is the number of cycles, is the internal resistance when the number of cycles is taken into account.

7. The method for estimating battery health status based on the end-cloud collaborative strategy according to claim 6 is characterized in that: In step 3, the steps of constructing the empirical model of battery aging on the terminal side are as follows: construct an LSTM neural network to predict the battery SOH, and use a forget gate to determine which information in the cell state is discarded and retained when the LSTM cell obtains data. The forget gate is expressed as: ； in is the output of the forget gate, is the weight matrix, is the bias vector, is the hidden state of the neuron at the previous time step, is the input of the LSTM network at the current time step, is a sigmoid function; the information in the unit is updated through the input gate, and the tanh layer is used to obtain new information. ; ； in and is the weight matrix, and is the bias vector, is the output of the input gate at the current time step, is the newly generated memory state of the neuron, is the hyperbolic tangent function; the old state Multiply by the output of the forget gate , represents the process of partially discarding data, and then adding the output of the input gate Multiply by the new memory state of the neuron , changing the memory state of the neuron from Updated to , It will be passed to the next neuron, and the calculation process is expressed as: , the output gate depends on the unit state , the previous cycle output and current input Generate a new hidden state The output gate is represented as: ； ； in and The values ​​are the weight matrix and bias vector, is the output of the output gate at the current time step, is the hidden state of the neuron at the current time step and is also the output of the current time step. Will make peace Pass it on to the next neuron.

8. The method for estimating battery health status based on a terminal-cloud collaborative strategy according to claim 1, characterized in that: In step 4, the mean absolute error (MAE) and the root mean square error (RMSE) are set, and the calculation formula is as follows: ， ； in for The true value of SOH at the moment, for The predicted SOH value at the moment.

Images