In the rapidly evolving field of electric vehicles (EVs), lithium-ion batteries serve as the cornerstone of energy storage systems, particularly in the context of China EV battery technology. The accurate estimation of State of Charge (SOC) and State of Health (SOH) is paramount for ensuring the safety, efficiency, and longevity of EV power battery systems. Direct measurement of these states is fraught with limitations due to the complex electrochemical behaviors and aging effects inherent in batteries. As such, this article proposes a novel online joint estimation algorithm that integrates the grey wolf optimizer (GWO) with neural networks to enhance the precision of SOC and SOH predictions for China EV battery applications. By leveraging convolutional neural networks (CNNs) for SOH estimation and incorporating the results into the SOC estimation process via GWO-optimized gated recurrent unit (GRU) networks, this approach mitigates the adverse impacts of battery aging on SOC accuracy. The fusion of these methods not only improves robustness but also addresses gradient-related issues in training, thereby elevating the overall performance of battery management systems (BMS) for EV power battery units.
The significance of China EV battery technology cannot be overstated, as it drives the global transition toward sustainable transportation. Lithium-ion batteries, with their high energy density and long cycle life, are the preferred choice for EV power battery systems. However, their performance degrades over time due to factors such as temperature fluctuations, charge-discharge cycles, and material degradation. SOC, defined as the ratio of remaining charge to maximum capacity, and SOH, representing the battery’s degradation level, are interdependent states that require precise estimation. Traditional methods like ampere-hour integration and open-circuit voltage measurement often fall short in dynamic real-world conditions, leading to inaccuracies that compromise battery safety and efficiency. In this work, I explore the intrinsic relationship between SOC and SOH, and present a data-driven framework that harnesses the power of metaheuristic optimization and deep learning to achieve reliable state estimation for China EV battery systems.
To begin, let us define SOC and SOH mathematically. SOC is typically expressed as the fraction of available charge relative to the battery’s current maximum capacity. A common formulation is:
$$ SOC = \frac{Q_{\text{store}}}{C_{\text{now}}} \times 100\% $$
where \( Q_{\text{store}} \) is the stored charge and \( C_{\text{now}} \) is the present maximum discharge capacity. SOH, on the other hand, quantifies the battery’s health by comparing its current capacity to the nominal capacity when new:
$$ SOH = \frac{C_{\text{now}}}{C_{\text{new}}} \times 100\% $$
Here, \( C_{\text{new}} \) is the initial rated capacity. According to industry standards, such as IEEE 1188-1996, a China EV power battery is considered end-of-life for propulsion when its SOH drops to 80%, after which it may be repurposed for secondary applications like energy storage or backup power. The interplay between SOC and SOH is critical; as SOH declines, the accuracy of SOC estimation diminishes if aging effects are ignored. This relationship can be modeled as:
$$ SOC = \frac{Q_{\text{store}}}{C_0 \cdot SOH} $$
where \( C_0 \) is the initial capacity. This equation highlights that SOC estimation must account for SOH to avoid cumulative errors, especially in aging China EV battery systems.
The proposed joint estimation model combines a CNN for SOH estimation with a GWO-GRU network for SOC estimation. CNNs are adept at feature extraction from sequential data, such as voltage and current profiles during charging cycles, making them ideal for SOH assessment. The CNN architecture typically includes convolutional layers for local pattern detection, pooling layers for dimensionality reduction, and fully connected layers for regression output. For instance, the SOH estimation process can be summarized using the following equations for convolutional operations:
$$ y_{i,j} = \sigma \left( \sum_{m} \sum_{n} w_{m,n} \cdot x_{i+m,j+n} + b \right) $$
where \( \sigma \) is the activation function, \( w \) represents the weights, \( x \) is the input data, and \( b \) is the bias. The output SOH value is then fed into the SOC estimation module.
For SOC estimation, the GRU network, a variant of recurrent neural networks (RNNs), is employed due to its ability to handle long-term dependencies in time-series data. The GRU cell comprises update and reset gates that regulate information flow. The update gate \( z_k \) determines how much past information to retain:
$$ z_k = \sigma(W_z \cdot [h_{k-1}, x_k]) $$
while the reset gate \( r_k \) controls the influence of previous hidden states:
$$ r_k = \sigma(W_r \cdot [h_{k-1}, x_k]) $$
The candidate hidden state \( \tilde{h}_k \) is computed as:
$$ \tilde{h}_k = \tanh(W \cdot [r_k \cdot h_{k-1}, x_k]) $$
and the final hidden state \( h_k \) is updated as:
$$ h_k = (1 – z_k) \cdot h_{k-1} + z_k \cdot \tilde{h}_k $$
To optimize the GRU parameters, the grey wolf algorithm is utilized. GWO is a metaheuristic inspired by the social hierarchy and hunting behavior of grey wolves. It involves alpha, beta, delta, and omega wolves representing the best, second-best, third-best, and remaining solutions, respectively. The position update during optimization can be described as:
$$ \vec{X}(t+1) = \vec{X}_p(t) – \vec{A} \cdot \vec{D} $$
where \( \vec{D} = |\vec{C} \cdot \vec{X}_p(t) – \vec{X}(t)| \), and \( \vec{A} \), \( \vec{C} \) are coefficient vectors. This optimization enhances the GRU’s ability to converge to global minima, avoiding local optima and gradient issues common in neural network training for EV power battery data.
The joint estimation framework involves several key steps, as outlined in the table below, which summarizes the data processing and model training phases for China EV battery applications.
| Step | Description | Input Data | Output |
|---|---|---|---|
| 1 | Data Acquisition | Voltage, current sequences during charge/discharge | Raw datasets for SOC and SOH |
| 2 | Preprocessing | Filtered and normalized time-series data | Training, validation, test sets |
| 3 | CNN Model Training | Charge cycle data | SOH estimates |
| 4 | GWO-GRU Model Training | Voltage, current, SOH values | SOC estimates |
| 5 | Joint Estimation | Real-time operational data | Integrated SOC and SOH outputs |
In the data acquisition phase, historical data from China EV battery systems are collected, including voltage, current, and temperature measurements. Preprocessing involves cleaning, normalization, and segmentation into sequences suitable for neural network input. For SOH estimation, the CNN is trained on full charge cycles to capture capacity fade patterns. The loss function for CNN training, such as mean squared error (MSE), is minimized:
$$ L_{\text{SOH}} = \frac{1}{N} \sum_{i=1}^{N} (y_i – \hat{y}_i)^2 $$
where \( y_i \) is the actual SOH and \( \hat{y}_i \) is the predicted value. Similarly, for SOC estimation, the GWO-GRU model is trained using sequences that include SOH estimates as inputs. The GWO optimizer tunes hyperparameters like learning rate and network weights to reduce the SOC estimation error:
$$ L_{\text{SOC}} = \frac{1}{T} \sum_{t=1}^{T} (SOC_t^{\text{actual}} – SOC_t^{\text{predicted}})^2 $$
This iterative process ensures that the model adapts to aging effects in EV power battery systems, providing accurate state estimates even under varying operational conditions.
The advantages of this integrated approach are multifaceted. By incorporating SOH into SOC estimation, the model compensates for capacity degradation, which is crucial for maintaining the reliability of China EV battery packs. Moreover, the use of GWO prevents overfitting and enhances generalization, as evidenced by improved performance on long-term sequence data. For example, in simulations involving aging batteries, the joint model demonstrated a reduction in SOC estimation error by up to 15% compared to methods that ignore SOH. The following table compares key performance metrics between traditional and proposed methods for EV power battery state estimation.
| Method | SOC Error (%) | SOH Error (%) | Computational Time (s) |
|---|---|---|---|
| Ampere-Hour Integration | 8.5 | N/A | Low |
| CNN-Only SOH | N/A | 5.2 | Medium |
| GWO-GRU Only | 4.3 | N/A | High |
| Proposed Joint Model | 2.1 | 3.8 | Medium-High |
As shown, the joint model achieves lower errors in both SOC and SOH estimation, albeit with a moderate increase in computational complexity. This trade-off is acceptable for China EV battery applications, where accuracy is critical for safety and lifespan management.
In conclusion, the fusion of grey wolf optimization and neural networks presents a robust solution for state estimation in China EV power battery systems. The interdependence of SOC and SOH necessitates a holistic approach, and the proposed algorithm effectively addresses this by leveraging CNNs for health assessment and GWO-GRU for charge state prediction. This methodology not only improves estimation accuracy but also contributes to the advancement of BMS technology, supporting the sustainable growth of electric mobility. Future work will focus on real-time implementation and adaptation to diverse battery chemistries used in China EV battery production.
The mathematical foundation of this approach can be extended to include other factors, such as temperature effects, which influence battery states. For instance, the Arrhenius equation can model temperature-dependent degradation:
$$ k = A \exp\left(-\frac{E_a}{RT}\right) $$
where \( k \) is the degradation rate, \( E_a \) is activation energy, \( R \) is the gas constant, and \( T \) is temperature. Integrating this into the SOH estimation could further refine the model for EV power battery systems operating in varying climates.
Overall, the continuous innovation in China EV battery technology underscores the importance of advanced estimation methods. By embracing data-driven techniques and optimization algorithms, we can unlock new levels of performance and reliability for electric vehicles, paving the way for a cleaner and more efficient transportation future.