Battery health state prediction method based on BMS and EMS fusion

Abstract

The invention relates to a battery health state prediction method based on BMS and EMS fusion. The method comprises the steps of obtaining and preprocessing EMS historical data; establishing a battery health state prediction model; pre-training a battery health state prediction model; the BMS collects battery data in real time and uploads the battery data to the EMS; retraining, optimizing and evaluating the trained battery health state prediction model by using the data uploaded by the BMS; and acquiring a battery health state prediction result based on data uploaded by the BMS by utilizing the retrained battery health state prediction model through evaluating the battery health state prediction model, and feeding back the battery health state prediction result to the BMS, wherein the BMS corrects the original SOH. According to the method, BMS and EMS architectures are fused, a model structure of a two-way gating cycle unit (GRU) optimized based on a particle swarm optimization (PSO) algorithm is built, meanwhile, a self-attention mechanism is introduced, the real-time online prediction of the battery health state is achieved, the model can be dynamically re-optimized according to continuously uploaded new data, and the real-time online prediction of the battery health state is achieved. Therefore, continuous improvement of prediction precision and adaptability enhancement are ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of energy storage batteries, in particular to a battery state of health prediction method based on the fusion of BMS and EMS. BACKGROUND

[0002] Lithium-ion batteries have become an important energy storage technology in power systems, electric vehicles, telecommunications, aerospace and many other industries due to their small size, high energy density, low self-discharge rate, low cost and long battery life. However, lithium-ion batteries will degrade over time and charge-discharge cycles, resulting in a decrease in their capacity and output power. In addition, improper use of batteries can accelerate their degradation, so dynamic and real-time monitoring of batteries is of great significance.

[0003] An important indicator reflecting the state of the battery is its state of health (SOH), defined as the ratio of the current capacity of the battery to its initial capacity. SOH directly reflects the degradation of the battery during its service life.

[0004] Traditional battery state of health monitoring methods often rely on offline measurement and periodic testing. This method not only consumes time and effort, but also makes it difficult to reflect the actual state of the battery. In addition, as the use time and charge-discharge cycles of the battery increase, the performance degradation pattern of the battery will also change, making it difficult for traditional monitoring methods to maintain long-term prediction accuracy.

[0005] Therefore, how to achieve real-time online prediction of the state of health of lithium-ion batteries and dynamically optimize the prediction model based on continuously uploaded new data to improve prediction accuracy and adaptability has become an important field of current research. SUMMARY

[0006] The present application mainly solves the above problems and provides a battery state of health prediction method based on the fusion of BMS and EMS. This method fuses BMS and EMS architecture and builds a model structure based on a particle swarm optimization algorithm (PSO) optimized bidirectional gated recurrent unit (GRU), while introducing a self-attention mechanism. The purpose is not only to achieve real-time online prediction of the battery state of health, but also to dynamically re-optimize the model based on continuously uploaded new data, thereby ensuring continuous improvement and adaptability of prediction accuracy.

[0007] The technical solution adopted by the present application to solve its technical problems is a battery state of health prediction method based on the fusion of BMS and EMS, comprising the following steps:

[0008] S1: Obtain EMS historical data and preprocess;

[0009] S2: Establish a battery state of health prediction model;

[0010] S3: Pre-training battery health state prediction model;

[0011] S4: BMS collects battery data in real time and uploads to EMS;

[0012] S5: Retraining, optimizing and evaluating the trained battery health state prediction model using the data uploaded by the BMS;

[0013] S6: Using the retrained and evaluated battery health state prediction model, the battery health state prediction result is obtained based on the data uploaded by the BMS and fed back to the BMS, and the original SOH is corrected.

[0014] As a preferred scheme of the above scheme, the step S1 comprises: screening battery cycle data features, the battery cycle data features including battery temperature, current, voltage, power, time and capacity; calculating the battery health state in each discharge cycle; and normalizing the data.

[0015] As a preferred scheme of the above scheme, the step S2 comprises: establishing a battery health prediction model based on a BI-GRU model and optimizing the parameters of the battery health prediction model using a PSO algorithm.

[0016] As a preferred scheme of the above scheme, the optimization target of the PSO algorithm is to minimize the prediction error, and the fitness function is as follows:

[0017]

[0018] Wherein, n is the total number of samples in the time series, is the predicted value of the i-th time point, y i is the true value of the i-th time point.

[0019] As a preferred scheme of the above scheme, the step S3 comprises:

[0020] S31: Constructing a training data set, extracting health factors representing the health state of lithium ion batteries from the charge-discharge curve and evaluating the direct correlation between each health factor and the battery health state using the Pearson correlation coefficient, the health factors including average discharge voltage, voltage difference between initial voltage, average power, current and energy at the same voltage;

[0021] S32: Designing a loss function, taking mean square error as the loss function;

[0022] S33: Model training, using adaptive learning rate gradient descent method for model training.

[0023] As a preferred scheme of the above scheme, the loss function is represented as:

[0024]

[0025] Wherein, n represents the sample number of true value and predicted value, y represents the true battery health state reflected by the battery cycle data, as the label of cycle data, measures the actual state of battery performance, y' represents the predicted battery health state after fusion feature processing and input into the trained complete battery health state prediction model.

[0026] As a preferred scheme of the above scheme, the data collected in step S4 includes the voltage, current and temperature of the single battery, and the total voltage, total current and total capacity of the battery pack.

[0027] As a preferred scheme of the above scheme, step S5 includes:

[0028] S51: obtaining the data uploaded by the BMS from the EMS, and dividing the data into training set, test set and validation set in proportion;

[0029] S52: retraining the battery health state prediction model and adjusting the parameters of the model;

[0030] S53: optimizing the model using the validation set;

[0031] S54: evaluating the model using the test set, and evaluating the prediction accuracy and reliability of the model by comparing the prediction results of the model with the actual values.

[0032] As a preferred scheme of the above scheme, in step S6, the BMS corrects the original SOH, including: comparing and analyzing the model prediction result with the original SOH calculation result of the BMS system, and updating the battery health state data in the BMS based on the prediction result.

[0033] As a preferred scheme of the above scheme, further comprising updating and optimizing the parameters of the original SOH calculation module of the BMS system.

[0034] The advantages of the present application are:

[0035] 1. The historical battery data of the energy management platform (EMS) and the real-time data of the battery management system (BMS) are fused to construct a comprehensive data platform, and real-time and dynamic monitoring of the battery health state is realized.

[0036] 2, BMS and EMS are fused to realize online prediction and timing optimization, real-time data and historical data are combined, the health status of the battery is evaluated in real time, and maintenance or replacement plan is made. Model parameters are updated regularly to achieve dynamic optimization. Real-time data from BMS helps to find abnormal conditions of the battery (such as voltage imbalance, temperature anomaly), and combined with historical data from EMS, it can be determined whether the abnormality is removable, which helps to find and deal with potential problems such as battery performance degradation, thereby prolonging the service life of the battery.

[0037] 3, PSO algorithm can intelligently adjust model parameters to improve convergence speed and prediction accuracy of the model, bidirectional GRU model can capture context information in time series and more accurately reflect the trend of battery health status change, and self-attention mechanism further enhances the model's ability to extract key information. BRIEF DESCRIPTION OF DRAWINGS

[0038] Figure 1 Figure 1 is a flowchart of a battery health state prediction method based on fusion of BMS and EMS.

[0039] Figure 2 Figure 2 is a structure diagram of Bi-GRU.

[0040] Figure 3 Figure 3 is a structure diagram of GRU. DETAILED DESCRIPTION

[0041] The technical solutions of the present application will be further described below through examples and in combination with the drawings.

[0042] Example:

[0043] The battery health state prediction method based on fusion of BMS and EMS in this embodiment, as shown in Figure 1 , includes the following steps:

[0044] S1: Obtain EMS historical data and pre-process.

[0045] Screen battery cycle data features, including battery temperature, current, voltage, power, time and capacity, etc. The battery health state in the first discharge cycle is set to 1, and the health state in all subsequent discharge cycles is set to the ratio of the capacity of the current cycle to the capacity of the starting cycle. The calculation expression of the battery health state in each discharge cycle is:

[0046]

[0047] Wherein, H i is the battery health state of the current cycle, C e , C i are the discharge capacity of the starting cycle and the discharge capacity of the current cycle, respectively.

[0048] A linear function (max-min normalization) is used to normalize the data, which limits the original data within the range of [0, 1], and the specific formula is as follows:

[0049]

[0050] where x' represents the normalized data, x represents the original data, x max and x min represent the minimum and maximum values in the original data, respectively. A sliding window with a size of 10 and a step of 1 is used to further divide the normalized data.

[0051] S2: Establishing a battery health state prediction model; a battery health prediction model is established based on a BI-GRU model, and a PSO algorithm is used to optimize the parameters of the battery health prediction model.

[0052] The PSO optimization algorithm includes:

[0053] S21: Initialization of particles. The initial position of the particle in the PSO algorithm is initialized, denoted as X i (0) . The speed of the particle V i (0) is a zero vector.

[0054] S22: Initialization of other parameters. Other parameters of the PSO algorithm are set, such as the inertia weight w=0.5, the learning factor c1=1, the learning factor c2=2, and the maximum number of iterations is 100.

[0055] S23: Definition of fitness function. For time series prediction problems, when using the particle swarm optimization algorithm (PSO) to optimize the parameters of the bidirectional gated recurrent unit (GRU) model, and the optimization goal is to minimize the prediction error, the fitness function can be directly related to the prediction error. Generally, it can be represented by mean square error (MSE), mean absolute error (MAE), or other similar measures, and the embodiment uses MSE to define the fitness function. MSE is the average of the squares of the differences between the predicted values and the true values, which provides a standard for measuring the accuracy of model prediction.

[0056]

[0057] where n is the total number of samples in the time series, is the predicted value at the i-th time point, y i is the true value at the i-th time point. In the PSO algorithm, the algorithm tries to find the particle with the maximum fitness value, which means that the smaller the MSE, the smaller the prediction error.

[0058] S24: PSO Iteration. In each iteration t, for each particle i and each dimension d (corresponding to a parameter of the model), the following velocity update formula is used:

[0059] V id (t+1) =ωV id t +c1r1(Pbest id -X id t )+c2r2(Gbest d -X id t )

[0060] Calculate the fitness value of each particle and update the individual optimal position P of the particle based on the fitness value. best and the optimal position G of the group best Update the particle's position using the following position update formula:

[0061] X id (t+1) =X id t +V id (t+1)

[0062] The iteration stops when the preset maximum number of iterations Tmax is reached or the fitness meets a certain threshold.

[0063] like Figure 2 As shown, the Bi-GRU model consists of a forward GRU layer and a backward GRU layer. The unique network structure of Bi-GRU allows each hidden layer to capture information from the past and future at a specific time step, thus extracting more comprehensive degradation features. The input layer receives the raw time series data and directly passes it to the forward and backward GRU layers (both process the same input in parallel). The forward GRU layer captures dependencies in the sequence from the past to the future (forward direction), extracting historical information and the hidden state at each time step. Depends on the hidden state of the previous time step and current time step x t The formula is expressed as:

[0064]

[0065] The backward GRU layer captures dependencies in the sequence from the future to the past (in reverse order) and extracts future information. The hidden state at each time step... Depends on the hidden state of the next time step and the input x at the current time step t The formula is expressed as:

[0066]

[0067] The hidden states of the forward GRU and the backward GRU are spliced in the feature dimension to form a bidirectional fusion feature, and then the key time features are further extracted through a self-attention mechanism to weight the important time steps, and the output layer outputs the final result.

[0068] Among them, the gated recurrent unit (GRU) introduces a gating mechanism, namely the reset gate and the update gate. In addition, the GRU does not have a cell state part in the LSTM. The internal structure of the GRU is as shown in Figure 3 The update gate (z t ) in the GRU is responsible for updating the hidden state, and the input vector (x t ) and the previous moment hidden state (h t-1 ) are transmitted through the Sigmoid function to obtain the update gate, which can be expressed as the following equation:

[0069] z t =σ(W z ·[h t-1 ,x t ]+b z )

[0070] Where σ represents the Sigmoid activation function, outputting a value between 0 and 1. W z is the weight matrix of the update gate, and b z is the bias matrix of the update gate.

[0071] The output of the reset gate is multiplied by the hidden state, and the input vector is added. This result is transmitted through the tanh function to pass the new memory content, which can be expressed as the following equation:

[0072]

[0073] Where W h and b h are the weight matrix and bias matrix of the new cell state of the new memory unit state. Once the memory content is obtained, the output of the current neuron can be derived using the equation:

[0074]

[0075] S3: a pre-trained battery state of health prediction model, comprising:

[0076] S31: Constructing training dataset. The historical data involved in this embodiment is 300 cycles. The relevant parameters are extracted from the charge-discharge curve as health factors representing the state of health of lithium-ion batteries. The following parameters are selected as health factors: average discharge voltage, voltage difference from initial voltage, power, current, and energy at the same voltage.

[0077] To further verify the rationality of the selected health parameters, the Pearson correlation coefficient is used to evaluate the correlation between them and the state of health of the battery. The Pearson correlation coefficient is used to represent the linear relationship and direction between two variables. The closer the value is to 1, the stronger the correlation. The calculation formula of the Pearson correlation coefficient is as follows:

[0078]

[0079] where n is the sample size, X i and Y i represent the sample values, and represent the mean of the sample, respectively.

[0080] S32: Designing loss function. The model takes the battery health state represented by the battery cycle data as the label, which is associated with the capacity of the battery at the current cycle. Through the learning of historical data, the model predicts the battery health state at the current cycle. To evaluate the model performance and optimize it, the model uses mean square error (MSE) as the loss function, which intuitively reflects the difference between the predicted value and the actual value. The specific representation is:

[0081]

[0082] where n represents the sample size of the true value and the predicted value. y represents the true battery health state reflected by the battery cycle data, which is the label of the cycle data and measures the actual state of the battery performance. y' represents the predicted battery health state after fusion feature processing and input into the trained PSO-BiGRU composite neural network model.

[0083] S33: Model training. In the model training process, the adaptive learning rate gradient descent method (Adam) is used for model training, which can effectively handle the optimization problem of large-scale data. In order to fully train the model and achieve better convergence effect, the training period (epoch) is set to 15 to obtain better model performance.

[0084] S4: BMS collects battery data in real time and uploads to EMS. The battery management system (BMS) first collects various key data of the battery pack in real time and accurately through the built-in sensor network. These data include but are not limited to the voltage, current and temperature of single battery, as well as the total voltage, total current and total capacity of the battery pack, etc. These data are the key indicators of battery health factors. After data collection is completed, BMS uploads the collected battery data to the energy management system (EMS) through a dedicated communication protocol and channel. This step requires efficient, secure and complete data transmission to ensure that data can be accurately and correctly delivered to EMS for further processing.

[0085] S5: Load the model pre-trained by EMS based on historical cycle data to EMS, and retrain, optimize and evaluate the trained battery health state prediction model using the data uploaded by BMS. Specifically, it includes:

[0086] S51: Obtain the latest battery data uploaded by BMS from EMS, including newly collected battery characteristics and corresponding health state labels. The data set is divided into training set, test set and validation set according to the ratio of 80%, 10% and 10%. These data will be used for model retraining, optimization and model evaluation.

[0087] S52: Load the updated data into the model and retrain. During the training process, the model's parameters and learning rate and other hyperparameters can be adjusted to further adjust the model's performance.

[0088] S53: After the training is completed, the model is optimized using the validation set data. By adjusting the structure and parameters of the model, the model has better generalization performance and robustness while maintaining high prediction accuracy.

[0089] S54: After optimization, the model is evaluated using the test set. By comparing the model's prediction results and actual values, the model's prediction accuracy and reliability are evaluated.

[0090] S6: After verifying that the model evaluation results are correct, the model will be used to predict the data uploaded by the BMS, generating the state of health (SOH) prediction results of the battery. These prediction results will serve as important reference information and be fed back to the BMS system in a timely manner. By comparing and analyzing with the original SOH calculation results of the BMS system, the health status of the battery can be more accurately understood, providing strong support for subsequent battery maintenance and management. After obtaining the prediction results provided by the EMS system, the original SOH calculation results in the BMS system will be corrected. This step will update and optimize the battery health status data in the BMS based on the prediction results of the EMS, ensuring that the BMS system can reflect the actual health status of the battery in real time and accurately. At the same time, we will also make necessary adjustments and optimizations to the relevant parameters of the BMS system to improve the overall performance and accuracy of the system.

[0091] The specific embodiments described herein are merely illustrative of the spirit of the present application. Those skilled in the art of the present application can make various modifications or supplements to the described specific embodiments or replace them with similar ways, without deviating from the spirit of the present application or exceeding the scope defined by the appended claims.

Claims

1. A battery health state prediction method based on the fusion of BMS and EMS, characterized by: Includes the following steps: S1: Obtain historical EMS data and preprocess it; S2: Establish a battery health status prediction model; S3: Pre-trained battery health status prediction model; S4: BMS collects battery data in real time and uploads it to EMS; S5: Use the data uploaded by the BMS to retrain, optimize and evaluate the trained battery health status prediction model; S6: Using the retrained and evaluated battery health state prediction model based on the data uploaded by BMS, obtain the battery health state prediction result and feed it back to BMS, which corrects the original SOH.

2. The battery health state prediction method based on the fusion of BMS and EMS according to claim 1, characterized in that: Step S1 includes: screening battery cycle data features, which include battery temperature, current, voltage, power, time, and capacity; calculating the battery health status in each discharge cycle; and normalizing the data.

3. The battery health state prediction method based on the fusion of BMS and EMS according to claim 1, characterized in that: Step S2 includes: establishing a battery health prediction model based on the BI-GRU model and optimizing the parameters of the battery health prediction model using the PSO algorithm.

4. The battery health status prediction method based on the fusion of BMS and EMS according to claim 3, characterized in that: The optimization objective of the PSO algorithm is to minimize the prediction error, and the fitness function is as follows: Where n is the total number of samples in the time series. y is the predicted value at the i-th time point. i It is the actual value at the i-th time point.

5. The battery health state prediction method based on the fusion of BMS and EMS according to claim 1, characterized in that: Step S3 includes: S31: Construct a training dataset, extract health factors representing the health status of lithium-ion batteries from charge-discharge curves, and use Pearson correlation coefficient to evaluate the direct correlation between each health factor and the battery health status. The health factors include the average discharge voltage, the voltage difference between the discharge voltage and the initial voltage, and the average values ​​of power, current and energy at the same voltage. S32: Design the loss function, using the mean squared error as the loss function; S33: Model training, using adaptive learning rate gradient descent for model training.

6. The battery health state prediction method based on the fusion of BMS and EMS according to claim 5, characterized in that: The loss function is expressed as: Where n represents the number of samples of true and predicted values, y represents the true battery health status reflected by the battery cycling data, serving as a label for the cycling data and measuring the actual state of battery performance, and y' represents the predicted battery health status after feature fusion processing and input into the fully trained battery health status prediction model.

7. The battery health state prediction method based on the fusion of BMS and EMS according to claim 1, characterized in that: The data collected in step S4 includes the voltage, current, and temperature of individual cells, as well as the total voltage, total current, and total capacity of the battery pack.

8. The battery health state prediction method based on the fusion of BMS and EMS according to claim 1, characterized in that: Step S5 includes: S51: Obtain the data uploaded by BMS from EMS and divide it into training set, test set and validation set according to the proportion; S52: Retrain the battery health status prediction model and adjust the model parameters; S53: Optimize the model using the validation set; S54: Use a test set to evaluate the model. By comparing the model's predictions with the actual values, assess the model's prediction accuracy and reliability.

9. The battery health state prediction method based on the fusion of BMS and EMS according to claim 1, characterized in that: In step S6, the correction of the original SOH by the BMS includes: comparing and analyzing the model prediction results with the original SOH calculation results of the BMS system, and updating the battery health status data in the BMS based on the prediction results.

10. The battery health state prediction method based on the fusion of BMS and EMS according to claim 9, characterized in that: It also includes updating and optimizing the parameters of the original SOH calculation module of the BMS system.

Images