As a researcher focused on advancing thermal management in new energy vehicles, I have observed the critical need for efficient systems that ensure both passenger comfort and battery safety. The integration of thermal management for the cabin and the battery presents a complex challenge due to dynamic coupling and conflicting objectives. In this article, I present a novel collaborative control strategy that leverages model predictive control (MPC) with neural networks to optimize system performance. Throughout this work, I emphasize the role of the battery management system (BMS) in monitoring and regulating thermal conditions, as it is pivotal for energy efficiency and longevity. The battery management system must work in harmony with other components to achieve optimal outcomes.
The growing adoption of electric vehicles worldwide has intensified the demand for innovative thermal management solutions. Traditional approaches, such as proportional-integral-derivative (PID) control, often fall short in handling the nonlinearities and multi-objective nature of integrated systems. My research addresses these limitations by proposing a synergistic framework that combines predictive modeling with real-time optimization. Central to this framework is the BMS, which provides essential data on battery temperature and state of charge, enabling precise control actions. By integrating the battery management system with cabin thermal controls, we can achieve significant energy savings and enhanced reliability.
In this article, I will detail the development of a dynamic simulation model, the incorporation of a nonlinear auto-regressive exogenous (NARX) neural network into the MPC module, and the comparative analysis with PID control under standardized driving cycles. Multiple tables and mathematical formulas are used to summarize key parameters and results, ensuring clarity and depth. The findings demonstrate that our strategy not only improves temperature stability but also reduces compressor energy consumption, highlighting the importance of an advanced battery management system in overall vehicle performance.
Introduction to Thermal Management Challenges
Thermal management in new energy vehicles is a multifaceted domain that directly impacts range, safety, and comfort. The cabin requires precise temperature control for passenger well-being, while the battery pack must operate within a narrow temperature window to prevent degradation and hazards. The battery management system plays a crucial role here, as it continuously monitors thermal states and triggers cooling or heating actions. However, standalone control of these domains often leads to suboptimal energy use. For instance, excessive cooling for the cabin might starve the battery of necessary thermal regulation, or vice versa. Therefore, a collaborative approach is essential.
My investigation begins with understanding the heat generation mechanisms in both cabin and battery. The cabin’s thermal load is influenced by solar radiation, ambient conditions, and occupant heat, whereas the battery’s heat arises from internal resistance and electrochemical reactions. The BMS typically estimates these heat fluxes using models, but integration with cabin systems adds complexity. I propose a unified model that captures these interactions, enabling coordinated control. The primary goal is to minimize energy consumption while maintaining temperatures within desired bounds. This requires sophisticated algorithms that can predict future states and adjust actuators accordingly.
To quantify the thermal dynamics, I derive fundamental equations. For the cabin, the heat balance can be expressed as:
$$ \frac{dT_c}{dt} = \frac{1}{m_c c_{p,c}} \left( \dot{Q}_{sol} + \dot{Q}_{amb} – \dot{Q}_{cool} \right) $$
where \( T_c \) is the cabin temperature, \( m_c \) is the air mass, \( c_{p,c} \) is the specific heat, \( \dot{Q}_{sol} \) is solar radiation, \( \dot{Q}_{amb} \) is ambient heat exchange, and \( \dot{Q}_{cool} \) is cooling from the air conditioning system. For the battery, the heat generation rate is given by:
$$ \dot{Q}_b = I^2 R_{int} + I T_b \frac{dU_{OCV}}{dT_b} $$
where \( I \) is current, \( R_{int} \) is internal resistance, \( T_b \) is battery temperature, and \( U_{OCV} \) is open-circuit voltage. The battery management system uses such equations to monitor conditions, but in an integrated setup, these models feed into a higher-level controller.
Background on Existing Control Strategies
Prior to delving into the proposed method, it is important to review conventional techniques. PID control has been widely used due to its simplicity and robustness. In thermal management, PID controllers adjust compressor speed or valve positions based on error signals. For example, the cabin temperature error drives the compressor, while the battery temperature error controls refrigerant flow valves. However, PID struggles with coupled systems because it reacts to past errors without anticipating future disturbances. This often results in overshoot, oscillations, and increased energy use. The BMS in such setups operates independently, leading to conflicts.
Model predictive control offers a superior alternative by using a dynamic model to forecast system behavior and optimize control inputs over a horizon. MPC can handle constraints and multi-objective tasks effectively. Recent studies have explored MPC for vehicle thermal management, but many rely on linear models that may not capture nonlinearities. My approach enhances MPC with a neural network to improve prediction accuracy, specifically for the integrated cabin-battery system. The battery management system data, such as temperature and state of charge, serve as inputs to this network, enabling better state estimation.
To illustrate the limitations of PID, consider the following comparison table based on typical performance metrics:
| Control Strategy | Overshoot | Settling Time | Energy Consumption | BMS Integration |
|---|---|---|---|---|
| PID | High | Long | High | Limited |
| MPC (Proposed) | Low | Short | Low | Comprehensive |
This table summarizes that PID control often leads to higher overshoot and longer settling times, whereas MPC reduces these issues. Moreover, the battery management system integration is more seamless in MPC, as it can account for battery states in predictive calculations.
Methodology: Integrated Model and Control Design
My methodology involves three main steps: building a dynamic simulation model, designing the MPC with neural network augmentation, and implementing PID for auxiliary actuators. The simulation model is developed using a platform analogous to AMESim, where cabin and battery thermal domains are coupled through a refrigerant circuit. The model parameters are derived from real vehicle data to ensure realism. Key components include the compressor, evaporators, expansion valves, and the battery management system module.
The cabin model uses a lumped-parameter approach, considering heat transfer from walls, windows, and ventilation. The governing equation is:
$$ m_c c_{p,c} \frac{dT_c}{dt} = h_c A_c (T_{amb} – T_c) + \dot{Q}_{occ} + \dot{Q}_{sol} – \dot{m}_r c_{p,r} (T_c – T_{evap}) $$
where \( h_c \) is the convective heat transfer coefficient, \( A_c \) is surface area, \( T_{amb} \) is ambient temperature, \( \dot{Q}_{occ} \) is occupant heat load, \( \dot{m}_r \) is refrigerant mass flow rate, \( c_{p,r} \) is refrigerant specific heat, and \( T_{evap} \) is evaporator temperature. The battery model incorporates an equivalent circuit with temperature-dependent resistance. The BMS updates the resistance based on temperature and state of charge, as per:
$$ R_{int} = \sum_{i=1}^{3} \left( a_i \beta^i T_b^{3-i} + b_i \beta^i + c_i T_b^{3-i} \right) $$
where \( \beta \) is state of charge, and \( a_i, b_i, c_i \) are empirical coefficients. This equation highlights how the battery management system captures thermal effects on electrical properties.
The integrated system state-space representation is formulated for MPC design. Let the state vector be \( \mathbf{x} = [T_c, T_b, P_{suct}]^T \), where \( P_{suct} \) is suction pressure, and the input vector be \( \mathbf{u} = [N_{comp}, \alpha_v]^T \), where \( N_{comp} \) is compressor speed and \( \alpha_v \) is valve opening. The disturbance vector includes ambient temperature and vehicle speed. The discrete-time state-space equation is:
$$ \mathbf{x}(k+1) = \mathbf{A} \mathbf{x}(k) + \mathbf{B}_u \mathbf{u}(k) + \mathbf{B}_v \mathbf{v}(k) $$
$$ \mathbf{y}(k) = \mathbf{C} \mathbf{x}(k) $$
where \( \mathbf{A} \), \( \mathbf{B}_u \), \( \mathbf{B}_v \), and \( \mathbf{C} \) are matrices identified from simulation data. The battery management system provides measurements for \( T_b \) and other battery-related states, which are fed into this model.
To enhance prediction accuracy, I integrate a NARX neural network that estimates future cabin temperature based on current states and disturbances. The network has inputs such as suction pressure, discharge pressure, vehicle speed, valve openings, and evaporator outlet temperature. Its output is the predicted cabin temperature, which is used in the MPC optimization. The cost function for MPC is:
$$ J(k) = \sum_{i=1}^{p} \| w(k+i) – \hat{y}(k+i) \|^2_{\mathbf{Q}_i} + \sum_{j=1}^{m} \| \Delta \mathbf{u}(k+j) \|^2_{\mathbf{R}_j} $$
where \( w \) is the desired output (e.g., cabin temperature setpoint), \( \hat{y} \) is the predicted output, \( \Delta \mathbf{u} \) is the control increment, and \( \mathbf{Q}_i \), \( \mathbf{R}_j \) are weight matrices. The optimization minimizes this cost subject to constraints like compressor speed limits:
$$ 1000 \leq N_{comp} \leq 5000 $$
Meanwhile, PID controllers are retained for valve regulation, ensuring rapid response to local disturbances. The battery management system interacts with these controllers by providing temperature setpoints for the battery cooling loop.
Simulation Setup and Parameters
To validate the strategy, I conduct co-simulations linking the thermal model with the control algorithm. The driving cycle used is the China Light-duty Vehicle Test Cycle for Passengers (CLTC-P), which lasts 1800 seconds and includes low, medium, and high-speed phases. Ambient temperatures are varied at 35°C, 40°C, and 45°C to test robustness. The cabin temperature setpoint is 25°C, and the battery desired range is 20-40°C. The BMS is configured to prioritize safety by requesting cooling when temperature exceeds 35°C.
Key simulation parameters are summarized in the following tables. First, cabin parameters:
| Parameter | Value | Unit |
|---|---|---|
| Cabin Volume | 4 | m³ |
| Wall Heat Capacity | 7 | kJ/K |
| Solar Radiation | 700 | W/m² |
| External Surface Area | 8 | m² |
Second, battery parameters for a lithium iron phosphate pack:
| Parameter | Value | Unit |
|---|---|---|
| Nominal Capacity | 124 | Ah |
| Nominal Voltage | 3.2 | V |
| Total Energy | 270 | MJ |
| Weight | 589 | kg |
| Number of Cells | 192 | – |
The battery management system uses these parameters to compute heat generation and monitor limits. The neural network training data comes from simulations under CLTC-P, with 70% for training, 15% for validation, and 15% for testing. The network achieves a prediction error of less than 0.2°C for cabin temperature, which is acceptable for MPC.
Results and Analysis
The simulation results compare the proposed MPC strategy with a baseline PID control. Performance metrics include temperature overshoot, settling time, and energy consumption per unit temperature drop. The latter is defined as:
$$ e = \frac{P}{\Delta T} $$
where \( P \) is compressor power and \( \Delta T \) is battery temperature drop. This metric reflects the efficiency of the thermal management system, with lower values indicating better performance. The battery management system logs temperature data to compute \( \Delta T \).
At an ambient temperature of 35°C, the MPC strategy reduces cabin temperature overshoot by 0.9°C compared to PID. Moreover, MPC achieves stability around 25°C within 290 seconds, whereas PID takes over 650 seconds. The battery temperature is also better regulated, staying within 31°C to 35°C across cycles. The following table summarizes the results for different ambient temperatures:
| Ambient Temp (°C) | Control Strategy | Cabin Overshoot (°C) | Settling Time (s) | Battery Temp Range (°C) | Compressor Power (W) | Energy per Temp Drop (W/°C) |
|---|---|---|---|---|---|---|
| 35 | PID | 1.5 | >650 | 30.2-35.0 | 1063 | 221.5 |
| MPC | 0.6 | 290 | 31.0-35.0 | 822 | 205.5 | |
| 40 | PID | 1.2 | >600 | 35.5-40.0 | 1274 | 283.1 |
| MPC | 0.3 | 300 | 35.9-40.0 | 1132.6 | 276.2 | |
| 45 | PID | 1.0 | >500 | 41.8-45.0 | 1261 | 394.1 |
| MPC | 0.2 | 280 | 40.3-45.0 | 1787 | 380.2 |
The data shows that MPC consistently reduces overshoot and settling time. Notably, at 40°C and 45°C, MPC almost eliminates overshoot. The energy per temperature drop is lower for MPC across all cases, indicating improved efficiency. For instance, at 35°C, MPC reduces energy consumption by 7.2% compared to PID. This is achieved because MPC anticipates future loads and adjusts compressor speed proactively, whereas PID reacts belatedly. The battery management system benefits from this as it receives more stable cooling, reducing thermal stress on cells.
Further analysis involves the predicted mean vote (PMV) for thermal comfort. PMV is calculated based on cabin temperature, humidity, and other factors. With MPC, PMV stabilizes between 0.4 and 0.78 after 600 seconds, corresponding to a neutral-to-slightly-warm sensation, which is acceptable for comfort. In contrast, PID leads to wider PMV fluctuations. This underscores the advantage of predictive control in maintaining comfort while saving energy.
The compressor speed profiles under MPC are smoother than under PID. The following equation approximates the compressor power consumption:
$$ P_{comp} = \eta \cdot \dot{m}_r \cdot (h_{discharge} – h_{suction}) $$
where \( \eta \) is efficiency, \( \dot{m}_r \) is refrigerant mass flow rate, and \( h \) is enthalpy. MPC optimizes this by modulating speed to match cooling demand, whereas PID causes frequent spikes. The battery management system contributes by providing accurate battery heat load predictions, which are incorporated into the MPC optimization.
Discussion on System Integration and BMS Role
The success of the proposed strategy hinges on seamless integration between the thermal management system and the battery management system. The BMS is not merely a monitor; it actively participates in control decisions by sharing data on battery temperature, state of charge, and health. In our framework, the battery management system sends setpoints and constraints to the MPC, ensuring battery safety is never compromised. For example, if the battery temperature approaches an upper limit, the BMS can request prioritized cooling, and MPC will adjust refrigerant allocation accordingly.
This collaborative approach addresses common issues in electric vehicles, such as range anxiety and battery degradation. By optimizing thermal management, we reduce parasitic loads from the compressor, thereby extending driving range. Moreover, keeping the battery within ideal temperature ranges slows aging processes. The battery management system tracks aging indicators like capacity fade, which can be mitigated through consistent thermal regulation.
From an implementation perspective, the neural network within MPC requires computational resources, but modern vehicle controllers are capable of running such algorithms in real-time. The BMS already employs similar computations for state estimation, so the additional burden is manageable. Future work could involve adaptive neural networks that learn from vehicle usage patterns, further enhancing prediction accuracy.
To quantify the benefits, consider the overall energy savings over a typical driving cycle. Using the energy per temperature drop metric, the MPC strategy saves an average of 4.4% across the tested ambient conditions. This translates to extended range and lower operating costs. The battery management system plays a key role in achieving these savings by enabling precise control.
Conclusion and Future Directions
In this article, I have presented a collaborative optimization strategy for new energy vehicle thermal management systems. By integrating MPC with a NARX neural network and leveraging data from the battery management system, we achieve superior temperature control and energy efficiency compared to traditional PID. The simulation results under CLTC-P cycles validate the approach, showing reduced overshoot, faster settling, and lower compressor energy consumption. The BMS is central to this success, as it provides critical thermal and electrical data that inform predictive control.
Future research will focus on hardware-in-the-loop testing to validate real-world performance. Additionally, incorporating weather and traffic predictions could further enhance MPC’s anticipatory capabilities. The battery management system could be expanded to include health-aware control, where thermal management adapts to battery degradation stages. Another avenue is integrating heat pump modes for both heating and cooling, which would require even more sophisticated coordination.
In summary, the synergy between advanced control algorithms and robust battery management systems is pivotal for the next generation of electric vehicles. As a researcher, I believe that continued innovation in this domain will drive sustainability and user acceptance, making electric vehicles more efficient and reliable.