With the rapid expansion of the electric car industry, a large number of power batteries are entering their end-of-life phase, posing significant challenges for resource management and environmental protection. In this study, we address the critical issues of predicting retirement times, evaluating recycling benefits, and optimizing system scheduling for retired batteries from China EV. We develop integrated models that leverage multi-source data fusion, economic and environmental assessment indicators, and collaborative optimization techniques to enhance decision-making accuracy and operational efficiency in battery recycling processes. Our approach aims to provide a scalable framework for managing the entire lifecycle of electric car batteries, supporting sustainable development in the China EV market.
The retirement of power batteries in electric cars is not determined by a single factor but results from the interplay of multiple variables. Key factors include the State of Health (SoH), cycle count, average operating temperature, cumulative mileage, and discharge rate. SoH, typically set at 80% of the initial rated capacity as a retirement threshold, serves as a core parameter for assessing remaining battery performance. Cycle count and temperature influence the rate of capacity degradation and thermal stability, while data from Battery Management Systems (BMS) enable quantitative analysis of these variables. We employ Principal Component Analysis (PCA) to identify correlations among these factors and construct multivariate regression models to clarify their explanatory power for retirement behavior. This enhances the precision and generalizability of our prediction models. The data used for variable selection are derived from operational electric car datasets and samples from recycling enterprises, covering the period from 2018 to 2022. The statistical characteristics and correlation analysis of these variables are summarized in Table 1.
| Variable Name | Mean | Standard Deviation | Correlation Coefficient (to SoH) |
|---|---|---|---|
| Cycle Count, cycles | 1125 | 312 | -0.83 |
| Average Temperature, °C | 31.7 | 4.5 | -0.69 |
| Cumulative Mileage, km | 86200 | 19800 | -0.75 |
| Discharge Rate (C-rate) | 1.3 | 0.4 | -0.56 |
| Initial Capacity, A·h | 92.6 | 3.1 | 0.12 |
From Table 1, it is evident that cycle count and cumulative mileage are the dominant variables influencing retirement status, showing significant negative correlations with SoH. This indicates that as usage intensity increases, battery health declines continuously. Therefore, subsequent modeling should prioritize highly correlated variables and establish differentiated retirement determination models for various battery types in electric cars.
For medium- to long-term prediction of retired battery volumes from China EV, time series models are widely adopted due to their sensitivity to historical trends and modeling simplicity. We apply the Autoregressive Integrated Moving Average (ARIMA) model to forecast annual retirement quantities. Based on cumulative sales data of electric cars in China from 2015 to 2022, and considering the typical battery lifecycle of approximately 5 to 8 years, we construct a retirement volume prediction model. Let \( Y_t \) represent the retirement volume in year \( t \). After confirming stationarity via the Augmented Dickey-Fuller (ADF) test, the fitted model is ARIMA(1,1,1), expressed as:
$$ \Delta Y_t = \phi_1 \Delta Y_{t-1} + \theta_1 \varepsilon_{t-1} + \varepsilon_t $$
The model parameters are estimated using the least squares method, and the residuals satisfy the white noise assumption. Prediction results show that by 2026, the annual retirement volume will exceed 950,000 units, indicating a clear upward trend with policy implications. The original data and prediction trends are illustrated in the context of electric car growth.
To improve the accuracy and adaptability of retirement predictions for electric car batteries, we integrate multi-source heterogeneous data, including vehicle operation behavior, environmental factors, and battery status. We introduce a Long Short-Term Memory (LSTM) neural network to build a fusion model that captures complex nonlinear relationships and temporal features. Model inputs include cycle count, remaining capacity, and voltage fluctuations recorded by BMS; average speed and terrain changes from GPS data; and external temperature and humidity from meteorological systems. Each feature sequence is standardized using Z-score normalization before being fed into the network. The loss function for the model is the Root Mean Square Error (RMSE):
$$ \text{RMSE} = \sqrt{\frac{1}{n} \sum_{i=1}^{n} (y_i – \hat{y}_i)^2} $$
where \( y_i \) is the actual retirement label and \( \hat{y}_i \) is the predicted value. Compared to traditional univariate models, the fusion model dynamically reflects the retirement trends of individual electric cars under different environmental conditions and usage intensities, making it suitable for large-scale fleet retirement risk assessment in the China EV sector. Model training is conducted in a Python environment and validated using historical measured datasets from 2017 to 2022.
To validate the effectiveness of different retirement prediction models, we systematically compare ARIMA, Support Vector Regression (SVR), and LSTM based on prediction accuracy, fitting capability, and stability. Evaluation metrics include RMSE, Mean Absolute Percentage Error (MAPE), and the coefficient of determination (\( R^2 \)), using the same historical dataset of estimated retirement volumes from 2015 to 2022 for training and testing. The evaluation results are presented in Table 2.
| Model Type | RMSE | MAPE (%) | \( R^2 \) |
|---|---|---|---|
| ARIMA | 7.61 | 14.3 | 0.861 |
| SVR | 5.47 | 10.1 | 0.908 |
| LSTM | 3.82 | 6.7 | 0.972 |
From Table 2, LSTM outperforms the other models across all metrics, with an RMSE of 3.82, significantly lower than ARIMA and SVR. Additionally, its \( R^2 \) value of 0.972 indicates strong fitting performance, effectively capturing nonlinear trends in retirement volumes. ARIMA is suitable for macro-trend predictions but is sensitive to outliers and lacks stability. SVR shows moderate performance but amplifies deviations from anomalies. Therefore, for large-scale operational scenarios involving electric cars, we recommend prioritizing deep learning models like LSTM.
In assessing the recycling of retired power batteries from electric cars, we establish a comprehensive evaluation system that considers technical characteristics and practical recycling chain requirements. This system is structured into three layers: objective, criterion, and indicator. The criterion layer includes four categories: economic benefits, environmental impact, safety assurance, and resource recovery rate. The indicator layer further breaks down into 12 core evaluation indicators, such as unit recycling cost, carbon emission intensity, thermal runaway risk, and nickel-cobalt recovery rates. Indicator selection is based on actual operational data and industry standards; for instance, unit recycling cost is derived from financial data of major recycling companies in 2022, and carbon emissions are quantified using Life Cycle Assessment (LCA) methods. The indicators are independent and clearly structured, reflecting the多元特性 of the entire recycling process and providing a quantitative foundation for comprehensive multi-indicator assessment in the context of China EV development.
The benefit evaluation of retired power battery recycling should encompass both economic and environmental dimensions. We compare two typical pathways: cascade use and再生利用, focusing on unit recycling cost, profit margin, and carbon reduction benefits. Economic data are based on average financial reports from companies like Ganfeng Lithium and Guanghua Tech in 2022, while environmental data are sourced from Ecoinvent 3.6 and Chinese LCA databases. The evaluation results are summarized in Table 3.
| Path Type | Recycling Cost, USD/(kW·h) | Profit Margin (%) | Carbon Reduction, kg CO2e/(kW·h) |
|---|---|---|---|
| Cascade Use | 152.8 | 12.4 | 37.6 |
| 再生利用 | 113.5 | 21.7 | 18.2 |
From Table 3, cascade use has a unit recycling cost of 152.8 USD/(kW·h) and a carbon reduction benefit of 37.6 kg CO2e/(kW·h), highlighting its environmental advantages in extending battery lifecycle. In contrast,再生利用 offers lower costs at 113.5 USD/(kW·h) and higher profit margins of 21.7%, making it more economically attractive, though with lower carbon reduction of 18.2 kg CO2e/(kW·h). Thus, cascade use is suitable for scenarios emphasizing long-term environmental benefits, while the再生利用path aligns better with value-driven business models in the electric car industry. Both pathways have distinct advantages, and recycling strategies should be selected based on technical conditions and policy directions for China EV.
To comprehensively evaluate the resource value and environmental burden of retired electric car batteries under different recycling pathways, we employ Life Cycle Assessment (LCA). The functional unit is defined as processing 1 kW·h of retired battery, with system boundaries covering transportation, preprocessing, recycling, and remanufacturing. The assessment results are presented in Table 4.
| Path Type | Carbon Emissions, kg CO2e | Resource Saving Value, USD | Material Recovery Value, USD |
|---|---|---|---|
| Cascade Use | 48.7 | 12.4 | 8.1 |
| 再生利用 | 61.3 | 6.7 | 29.6 |
From Table 4, cascade use excels in extending service life, effectively reducing the consumption of virgin resources, with a lifecycle carbon emission intensity of 48.7 kg CO2e/(kW·h) and a resource saving benefit of 12.4 USD/(kW·h). Although再生利用has slightly higher carbon emissions (61.3 kg CO2e/(kW·h)), it offers greater value in recovering critical metals like nickel and cobalt, with a lifecycle recovery value of 29.6 USD/(kW·h). Cascade use is ideal for usage-extension scenarios, while再生利用suits resource-oriented recycling systems, demonstrating complementary lifecycle values for electric car batteries.
In multi-path recycling evaluation, relying on a single indicator is insufficient to reflect technical superiority and practical adaptability. Therefore, we introduce an integrated decision-making model combining the Analytic Hierarchy Process (AHP) and entropy weight method to assign quantitative weights and optimize schemes across multiple dimensions, including economy, environmental impact, resource recovery value, and safety. The results show that carbon emissions and material recovery value have weights of 0.31 and 0.27, respectively, indicating the industry’s emphasis on low-carbon practices and resource recycling. Cascade use performs better in environmental dimensions, while再生利用excels in economy and metal recovery. This method provides a structured, quantifiable basis for policy formulation and enterprise technology route selection in the China EV sector.
After retirement, the recycling process for electric car batteries typically includes five core stages: collection access, status assessment, path determination, technical sorting, and final treatment. Different battery types (e.g., lithium iron phosphate, ternary lithium) and usage states significantly influence the selection of工艺路径. For cascade use, emphasis is placed on residual capacity and health state screening to suit medium-to-low rate applications such as energy storage and electric two-wheelers. For再生利用, the focus is on material disassembly efficiency and purification process adaptation; for example, ternary batteries have high metal recovery rates in hydrometallurgy, making them suitable for heavy metal extraction, while lithium iron phosphate batteries, with lower lithium content, are often processed using pyrometallurgical methods due to economic constraints. Additionally, PACK structure, cell packaging, and thermal management systems directly affect disassembly difficulty and recycling technology choices. Integrated battery packs require automated thermal disassembly and intelligent recognition equipment, as traditional manual methods cannot meet efficiency and safety requirements for electric car batteries.
The power battery recycling system involves a multi-node network including collection points, transfer centers, and processing plants, characterized by multiple paths, resource constraints, and spatial imbalances. Simulating the Yangtze River Delta region as an example, we construct a network system with 20 collection points, 4 transfer centers, and 2 recycling plants. Input data include regional annual recycling volumes for 2022, node processing capacities, and a transportation distance matrix, with a carbon emission coefficient set at 0.072 kg CO2e/(t·km). Simulation results show that under capacity constraints and time window limitations, the optimized network reduces total path mileage by 18.4%, decreases transportation carbon emissions by 13.2%, and shortens average single dispatch time by approximately 9.5%. Node load distribution becomes more balanced, effectively alleviating bottlenecks in transfer processes. This method provides feasible solutions for regional recycling system planning and resource allocation, demonstrating good engineering adaptability and调度鲁棒性 for China EV applications.
To enhance the operational efficiency and coordination of retired battery recycling systems, we develop a discrete-event simulation platform based on Python and SimPy, incorporating a Multi-Objective Particle Swarm Optimization (MOPSO) scheduling model to jointly optimize task response time, vehicle idle rate, and plant processing load. The scheduling process includes dynamic task collection, multi-level priority classification, and time window matching, using a rolling optimization approach for real-time decision-making. In a typical scenario based on 2022 operational data from East China (involving 100 recycling vehicles and 8 processing stations), simulation results show a 16.8% reduction in idle rate, a 12.3% shortening of average task cycle, and an increase in processing capacity utilization to 92.4%. The coordination mechanism, through task clustering and path restructuring, significantly alleviates node congestion and resource waste during peak periods. This method combines stability and responsiveness, making it suitable for building intelligent scheduling systems in regionalized, multi-center retired battery management for electric cars, and providing technical support for enterprises to develop efficient operational strategies.
In conclusion, this study addresses the prediction and recycling evaluation of waste batteries from electric cars by systematically developing prediction models based on variable identification and multi-source data fusion, establishing a multi-indicator evaluation system covering economic, environmental, and resource values, and proposing adaptability analysis and path collaborative optimization methods. This forms a complete technical process from status determination to system scheduling, tailored for the China EV market. Future research could further integrate real-time operational data and policy constraints to enhance model dynamic response capabilities, improve the practicality and robustness of scheduling algorithms in complex recycling networks, and promote efficiency in industrial chain coordination for sustainable electric car development.