A battery thermal management energy-saving regulation method and system based on feedforward feedback coupling

Abstract

This invention belongs to the field of lithium-ion battery thermal management technology, specifically a battery thermal management energy-saving control method and system based on feedforward-feedback coupling. The method includes: S1, collecting lithium-ion battery operating data and constructing a physical heat generation model based on the Bernardi heat generation equation; S2, constructing a hybrid heat generation prediction model to obtain a real-time predicted heat generation rate; S3, designing a feedforward-feedback dual closed-loop controller, with Q... pred The feedforward cooling power P is calculated using model predictive control as input. ff Using the measured highest temperature and maximum temperature difference ΔT as inputs, the power P is calculated and corrected via adaptive PID control. fb , will P ff With P fb The final cooling power P is obtained by weighted fusion. cool S4, according to P cool Dynamically adjust cooling temperature. This invention achieves a balance between real-time performance, precision, energy efficiency, and safety in lithium-ion battery thermal management.

Description

Technical Field

[0001] This invention relates to the field of lithium-ion battery thermal management technology, and in particular to a battery thermal management energy-saving control method and system based on feedforward feedback coupling. Background Technology

[0002] Lithium-ion batteries are widely used in new energy vehicles, large-scale energy storage, and consumer electronics due to their advantages such as high energy density, long cycle life, and low self-discharge rate. However, lithium-ion batteries generate a large amount of heat during charging and discharging. If this heat cannot be dissipated in a timely and effective manner, it will lead to an increase in battery temperature and even cause serious safety problems such as thermal runaway.

[0003] Currently, most mainstream battery thermal management control strategies are based on temperature feedback control. The basic principle is to monitor the battery temperature in real time using a temperature sensor; when the temperature exceeds a preset threshold, the cooling system is activated; and when the temperature drops to a safe range, cooling is stopped or reduced. However, this traditional temperature feedback control has the following significant drawbacks: (1) Response hysteresis: Temperature change is the result of the combined effect of heat generation and heat dissipation, and has a large thermal inertia. When the temperature exceeds the standard, a considerable amount of heat may have accumulated inside the battery. At this time, it is difficult to quickly control the temperature within a safe range when the cooling system is started. Especially under high-rate charge and discharge conditions, this hysteresis can easily lead to temperature runaway.

[0004] (2) Coarse control: Threshold-based switching control or simple PID control is difficult to achieve precise temperature regulation. The cooling system often switches between "fully on" and "fully off", resulting in large fluctuations in battery temperature, high energy consumption, and is not conducive to the control of temperature uniformity within the battery pack.

[0005] (3) Low energy efficiency: Due to the lack of prediction of battery heat generation trends, the cooling system can only respond passively after the temperature exceeds the standard, resulting in a mismatch between cooling capacity and heat generation demand. Either insufficient cooling leads to excessively high temperature, or excessive cooling causes energy waste, increasing the parasitic power consumption of the whole vehicle or energy storage system. Summary of the Invention

[0006] This invention proposes a battery thermal management energy-saving control method and system based on feedforward feedback coupling, aiming to solve the technical problem in the prior art that when the internal temperature of a lithium-ion battery is detected to exceed the standard, a large amount of heat may have accumulated inside the battery, making it difficult to control quickly.

[0007] In a first aspect, the present invention provides a battery thermal management energy-saving control method based on feedforward feedback coupling, comprising: S1 collects lithium-ion battery operating data, including current I, voltage V, and surface temperature T. s Ambient temperature Ta And a physical heat production model was constructed based on the Bernardi heat production equation; S2, Construct a hybrid heat production prediction model, which includes the physical heat production model and a data-driven model. The data-driven model predicts the heat production residual based on historical operating data using a machine learning algorithm. The output of the physical heat production model is fused with the heat production residual to obtain the real-time heat production rate prediction value Q. pred ; S3, construct a feedforward-feedback dual closed-loop controller, using the real-time heat generation rate prediction value Q. pred As input, the feedforward cooling power P is calculated using a model predictive control algorithm based on a feedforward controller. ff The highest measured temperature T max The maximum temperature difference ΔT is taken as input, and the feedback correction power P is calculated by the feedback controller based on the adaptive PID algorithm. fb and P ff With P fb The final cooling power P is obtained by weighted fusion. cool ; S4, based on the final cooling power P cool The cooling temperature is dynamically adjusted, and a safety fallback strategy is implemented based on a preset temperature safety threshold.

[0008] The technical effect of the battery thermal management energy-saving control method based on feedforward-feedback coupling disclosed in this invention is as follows: by constructing a hybrid heat generation prediction model to achieve advanced and accurate calorific value prediction, and by adopting a feedforward-feedback dual closed-loop control strategy combined with three-level safety net, the real-time performance, accuracy, energy saving and safety of thermal management are significantly improved.

[0009] Furthermore, the physical heat generation model in S1 is as follows: ; in, The physical model defines the heat generation rate, where E is the open-circuit voltage and T is the weighted average of multiple temperature measurement points on the battery surface. The entropy coefficient is obtained by differentiating the open-circuit voltage curves at different temperatures.

[0010] Furthermore, the physical heat production model in S2 outputs a calculated heat production rate value Q. phy The data-driven model employs a temporal convolutional network (TCN) and a multi-head attention mechanism to extract temporal features from the input features, and uses the battery's actual real-time heat generation rate Q. real Based on the actual real-time heat generation rate Q of the battery, the prediction is made. real Compared with the calculated value Q from the physical heat production model phy The heat production residual ε formed by the difference res =Q real Q phy Input features include I, V, and T at the current time. s T a The corresponding sliding window statistics, difference values, and frequency domain components are obtained; finally, the physical heat generation model is calculated using an adaptive fusion gate to obtain the value Q. phy After residual correction, the real-time predicted heat production rate Q is output. pred : ; Wherein, the fusion weight λ∈[0,1] is the value Q calculated by the fusion weight network based on the physical heat generation model under the current operating conditions. phy With the battery's actual real-time heat generation rate Q real The degree of deviation is dynamically calculated.

[0011] Furthermore, the feedforward controller in S3 employs Model Predictive Control (MPC), and its objective function is: ; Where, N p To predict the time domain, ,β, T represents the weighting coefficient. target For the target temperature, P cool (k) represents the final cooling power at time k, T pred (k) is based on a simplified thermal dynamics model and Q. pred The predicted temperature at time k, ΔP, is calculated. cool (k) represents the change in cooling power at time k.

[0012] Furthermore, the feedback controller in S3 employs an adaptive PID controller, and its output is: ; in, =T max -T target For temperature deviation, K p K i K d K ΔT K is the gain coefficient. ΔT Used for temperature difference compensation.

[0013] Furthermore, the P ff With P fb The weighted fusion method is as follows: ; Where, ω ff ω fb For dynamic weights and ω ff +ω fb =1.

[0014] Furthermore, the safety fallback strategy in S4 includes three levels of criteria: Level 1: When T max <T warn At that time, feedforward control is the primary method, and the economic cooling mode is executed. Level 2: When T warn ≤T max <T crit Or when 5℃≤ΔT<10℃, strengthen feedback control, start active balancing and limit charging and discharging power; Level 3: When T max ≥T crit If ΔT ≥ 10℃, immediately activate the maximum cooling mode and disconnect the charging / discharging circuit; Among them, T warn As the warning temperature threshold, T crit T is the critical temperature threshold. warn =40℃, T crit =50℃.

[0015] Secondly, the present invention provides a battery thermal management energy-saving control system based on feedforward feedback coupling for implementing the method, the system comprising: The data acquisition unit is used to collect operating data of the lithium-ion battery, including current I, voltage V, and surface temperature T. s Ambient temperature T a And a physical heat production model was constructed based on the Bernardi heat production equation; A hybrid prediction unit is used to construct a hybrid heat production prediction model, which includes the physical heat production model and a data-driven model. The data-driven model predicts the heat production residual based on historical operating data using a machine learning algorithm. The output of the physical heat production model is fused with the heat production residual to obtain the real-time heat production rate prediction value Q. pred ; A dual-closed-loop control unit is used to construct a feedforward-feedback dual-closed-loop controller, based on the real-time heat production rate prediction value Q. pred As input, the feedforward cooling power P is calculated using a model predictive control algorithm based on a feedforward controller. ff The highest measured temperature T max The maximum temperature difference ΔT is taken as input, and the feedback correction power P is calculated by the feedback controller based on the adaptive PID algorithm. fb and P ff With P fb The final cooling power P is obtained by weighted fusion. cool ; The execution and safety unit is configured to, based on the final cooling power P coolThe cooling temperature is dynamically adjusted, and a safety fallback strategy is implemented based on a preset temperature safety threshold.

[0016] The technical effect of the system disclosed in this invention is that, through the integrated data acquisition, hybrid prediction, dual closed-loop control and execution unit, the above-mentioned method can be effectively implemented to achieve dynamic optimization and control of battery temperature.

[0017] Thirdly, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described thereon.

[0018] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the method described thereon. Attached Figure Description

[0019] Figure 1 This is a schematic flowchart of a battery thermal management energy-saving control method based on feedforward feedback coupling proposed in an embodiment of the present invention. Figure 2 This is a flowchart illustrating the construction and operation of the hybrid heat generation prediction model in this embodiment of the invention. Figure 3 This is a flowchart illustrating the design and coordinated control of the dual closed-loop controller in an embodiment of the present invention. Figure 4 This is a schematic diagram of the safety fallback and dynamic control process in an embodiment of the present invention; Figure 5 The following are the Matlab simulation data effects in the embodiments of the present invention; wherein, (a) is a comparison chart of total energy consumption; (b) is a comparison chart of power distribution; and (c) is a comparison chart of comprehensive performance indicators. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] The purpose of this invention is to overcome the shortcomings of existing lithium-ion battery thermal management technologies and provide a battery thermal management energy-saving control method based on feedforward feedback coupling, achieving a balance between real-time performance, accuracy, energy efficiency, and safety in lithium-ion battery thermal management. (Reference) Figures 1 to 5 As shown, the specific steps include: Step S1: Data acquisition and establishment of physical heat generation model.

[0022] First, a high-precision data acquisition system is established. This embodiment uses a distributed architecture for data acquisition, targeting a typical 96S4P (96 series 4 parallel) battery pack configuration. The current acquisition module uses a closed-loop Hall current sensor (such as LEMHAH1DR), installed on the positive and negative busbars of the main circuit of the battery pack, with a sampling frequency of 10Hz, accuracy ±0.5%, and measurement range ±300A. It outputs current data I via the CAN bus. The voltage acquisition module uses an isolated differential amplifier (such as TIINA240) to measure the terminal voltage of each battery module, with a sampling frequency of 10Hz, accuracy ±0.2%, and output voltage data V. The temperature acquisition network deploys an NTC thermistor array, following the "hotspot coverage" principle, with a total of 48 temperature measurement points: one point each at the center, edge, and corner of each module surface, and six points at key connection points. The sampling frequency is 1Hz, accuracy ±0.5℃, and the output temperature data T for each measurement point is displayed. i Used to calculate surface temperature T s Average temperature T, maximum temperature T max The maximum temperature difference ΔT. The environmental monitoring module places temperature and humidity sensors at four corners inside the battery pack to measure the ambient temperature T. a The sampling frequency is 1Hz. All sensor data are synchronized using a unified time stamp with a timestamp accuracy of 1ms, ensuring time alignment of multi-source data.

[0023] The collected raw data undergoes rigorous preprocessing. First, outlier detection is performed using a modified 3σ method. The mean and standard deviation of the data within a sliding window (100 sampling points) are calculated, and data points exceeding the μ±3σ range are removed. Additional physical constraints are applied to the current data I: charging current is negative, discharging current is positive, and the absolute value does not exceed 1.5 times the battery's rated current. For missing values, single missing points are interpolated linearly; consecutive missing points (2-5 points) are interpolated cubically using cubic spline interpolation; and large numbers of missing points (>5 points) are marked as invalid data segments and removed during model training. Finally, data normalization is performed. Current I and voltage V use global minimum-maximum normalization based on historical data statistics, while temperature data (T...)... s T, T max For (e.g., ΔT), adaptive normalization based on the recent data range is adopted.

[0024] Based on the preprocessed data, a fundamental physical model of heat production is constructed. This embodiment is based on the Bernardi heat production equation: ; The open-circuit voltage E(SOC,T) was obtained through HPPC experiments. At a reference temperature of 25℃, mixed-pulse tests were performed at 10% SOC intervals to measure the balance voltage at each SOC point. The experiments were repeated at different temperature points (0℃, 15℃, 25℃, 35℃, 45℃) to construct a three-dimensional lookup table for E(SOC,T), where E is the open-circuit voltage (a function of SOC and temperature T, obtained through HPPC experiments at different battery states of charge and temperatures, and the three-dimensional lookup table was constructed). In practical applications, bilinear interpolation was used for fast lookup. Entropy thermal coefficient... The average temperature T was calculated by differentially analyzing the open-circuit voltage curves at different temperatures. The average temperature T was taken as the weighted average of 48 temperature measurement points on the battery surface, with the weights determined based on the thermal representativeness of the thermocouple placement (e.g., center points have a higher weight than edge points). To optimize real-time calculations, E(SOC,T) and... (SOC) pre-calculation is performed as a lookup table, stored in the controller ROM. Q can be quickly calculated by substituting the real-time acquired I, V, and T values. phy The calculation time for a single operation is less than 0.1ms.

[0025] Step S2: Construction of the Hybrid Heat Generation Prediction Model. This step constructs a hybrid heat generation prediction model that integrates a physical model and a data-driven model to output a more accurate real-time heat generation rate prediction value Q. pred .like Figure 2 As shown, the model includes a physical branch (i.e., the physical heat generation model) and a data-driven branch (data-driven model).

[0026] First, feature engineering design is performed. The input features are divided into three groups: (1) Original feature group: I(t), V(t), Ts(t), Ta(t), SOC(t) at the current time; (2) Time series feature group: extracting seven statistics of mean, standard deviation, maximum, minimum, range, skewness, and kurtosis from the original features using a sliding window (w=20); (3) Dynamic feature group: first-order difference Δx(t)=x(t)-x(t-1), rate of change feature dx / dt, and frequency domain feature Xtp=F(Xt) (obtained by extracting the original features through fast Fourier transform). After the total feature dimensions are evaluated by the random forest feature importance assessment, about 25-30 key features are retained.

[0027] The data-driven branch is built upon TCN and a multi-head attention mechanism. The TCN consists of four layers of dilated causal convolutions with dilation coefficients of 1, 2, 4, and 8, kernel size k=3, and the number of channels increases layer by layer: 32→64→128→256. Each layer includes convolution, weight normalization, ReLU activation, Dropout (0.2), and residual connections. The TCN output features are fed into the multi-head attention layer, with h=4 attention heads, and each head has query, key, and value dimensions dk=dv=32. The attention output is then processed through LayerNorm and a feedforward neural network, and finally mapped to a 1-dimensional output by a fully connected layer. That is, the residual heat production.

[0028] Physical branch output Q phy Data-driven branch output ε res The two are fused through an adaptive fusion mechanism to obtain the final predicted heat production rate, Q. pred : ; Among them, the residual heat production ε res For the battery's true real-time heat generation rate Q real With the physical model output Q phy The difference, and the fusion weights λ∈[0,1], are dynamically calculated by the fusion weight network. This mechanism guarantees the physical model Q. phy As the underlying predicted value always exists, the residual ε provided by the data-driven model... res The model is then modified to different degrees based on the confidence level of the current operating condition: when the confidence level of the physical model is high (such as steady-state operating condition), λ approaches 0, and the model relies more on the physical branch; when the confidence level of the physical model is low (such as complex dynamic operating condition), λ approaches 1, and the model introduces more residual correction of the data-driven branch, thereby achieving complementary advantages.

[0029] During model training, ±5% random noise was added to the current data I, and ±0.5℃ random offset was added to the temperature data Ts, T, etc., to enhance robustness. The loss function used was weighted MSE, with 1.5 times the weight for high-rate discharge conditions (|I|>1C) and 1.3 times the weight for high-temperature conditions (T>40℃). The optimizer used was AdamW, with an initial learning rate of 1e-3, weight decay of 1e-4, and cosine annealing learning rate scheduling and early stopping mechanisms.

[0030] Step S3: Design of a dual-closed-loop controller with calorific value feedforward and temperature feedback. This step designs a dual-closed-loop controller that combines the speed of feedforward with the accuracy of feedback, such as... Figure 3 As shown. Feedforward controller (MPC implementation): The feedforward controller is designed based on a simplified thermal dynamics model. The discretized form of the state equation of this model is: ; Among them, A and B u B d The system matrix identified experimentally involves parameters such as battery heat capacity C, convective heat transfer coefficient h, and heat transfer area A. The MPC optimization problem is constructed as a quadratic programming problem with the objective function: ; Where, N p To predict the value of 20 in the time domain, control the time domain N. c =5, ,β, For weighting coefficients (e.g., α=0.01, β=1.0, γ=0.1), T target For the target temperature, P cool (k) represents the final cooling power at time k, T pred (k) is based on a simplified thermal dynamics model and Q. pred The predicted temperature at time k, ΔP, is calculated. cool (k) represents the change in cooling power at time k. The optimization solution employs a hot-start sequential quadratic programming (SQP) algorithm, using the optimal solution from the previous time step as an initial guess, significantly reducing computation time. The solution frequency is 1Hz, and the single solution time is controlled within 50ms. The output is the feedforward cooling power P. ff .

[0031] Feedback Controller (Adaptive PID Implementation): The feedback controller adopts an incremental PID structure and introduces a temperature difference feedforward term. Its core calculation formula is: ; in, =T max -T target For temperature deviation, K p K i K d K ΔT K is the gain coefficient. ΔT For temperature difference compensation, an adaptive strategy is used for dynamic adjustment. For example, K p The base value is 50, and it is adjusted according to the magnitude of the error and the trend of change. K p =K p0 ×[1+0.3×| | / 10+0.2×|T max | / 1]; K i The base value is 0.5, and an integral separation strategy is adopted. When K is too large, the integral effect is weakened; d The base value is 10, considering noise suppression; K ΔTThe base value is 20, and it increases with increasing temperature difference: K ΔT =K ΔT0 ×[1+0.3×ΔT / 10]. The differential term uses a first-order low-pass filter (time constant τ=5s) to suppress noise.

[0032] Dual closed-loop coordination mechanism: final cooling power P cool Weighted synthesis of feedforward and feedback outputs: ; Where, ω ff ω fb For dynamic weights and ω ff +ω fb =1. Adjust ω based on system conditions (such as prediction confidence, temperature deviation, etc.). For example, increase ω when prediction confidence is high and temperature deviation is small. ff To leverage the energy-saving advantages of feedforward; when the temperature deviation is large or the prediction confidence is low, ω is increased. fb To enhance the robustness of the system.

[0033] Step S4: Dynamic Control and Safety Backup Strategy. Based on the calculated final cooling power P cool This step involves dynamic control of the cooling system and the implementation of a three-level safety fallback strategy, such as... Figure 4 As shown. First, the cooling power command needs to be converted into a specific actuator control signal. For liquid cooling systems, the cooling power is converted into a combination of pump speed and valve opening through a pre-calibrated mapping relationship. For air-cooled systems, it is converted into fan speed. To compensate for actuator response lag (e.g., pump lag of approximately 0.5s, fan lag of approximately 1.0s), a feedforward compensation algorithm is used: u cmd =u desired +τ×(du desired / dt), where τ is the actuator time constant, u desired For the expected instruction, u cmd This is the final instruction after compensation.

[0034] The core security feature of this invention lies in its three-level, state machine-based security fallback strategy, with specific criteria and actions shown in Table 1 below: Table 1. Triggering conditions and core actions for the three levels

[0035] The switching between different modes uses state machine logic with hysteresis. For example, the condition for entering the warning mode from the normal mode is T. max ≥41℃ or ΔT≥6℃ (hysteresis 1℃), while the condition for returning from the warning mode to the normal mode is T. maxThe system operates at temperatures below 39℃ and ΔT < 4℃ to prevent frequent oscillations near the critical point. In addition to maximum temperature control, the system also pays special attention to temperature difference control. When ΔT > 5℃, a temperature difference compensation term can be added to the base cooling power.

[0036] In addition, the system integrates a complete fault diagnosis and handling mechanism, including sensor fault diagnosis (consistency check), communication fault handling (using the previous cycle data to keep running after CAN bus timeout), controller watchdog, and dual-machine hot standby function, to ensure the reliable operation of the system throughout its entire life cycle.

[0037] Technical effectiveness verification: A simulation model was built on the Matlab / Simulink platform for verification (simulation results are shown below). Figure 5 (As shown). The results show that, compared with traditional threshold-based switching control, the method proposed in this invention has the following significant advantages: (1) System energy consumption: Under typical operating conditions, the total energy consumption of the thermal management system is reduced by about 26%.

[0038] (2) Temperature control effect: The battery high temperature time is reduced by more than 40%.

[0039] (3) Power distribution: The high power time was reduced by more than 40%.

[0040] (4) Safety: Under simulated extreme abuse conditions, the three-level safety fallback strategy can trigger emergency protection within milliseconds, effectively preventing the spread of thermal runaway.

[0041] Based on the same inventive concept, embodiments of the present invention also provide a battery thermal management energy-saving control system based on feedforward feedback coupling for implementing the above method, the system comprising: The data acquisition unit is used to collect operating data of the lithium-ion battery, including current I, voltage V, and surface temperature T. s Ambient temperature T a And a physical heat production model was constructed based on the Bernardi heat production equation; A hybrid prediction unit is used to construct a hybrid heat production prediction model, which includes the physical heat production model and a data-driven model. The data-driven model predicts the heat production residual based on historical operating data using a machine learning algorithm. The output of the physical heat production model is fused with the heat production residual to obtain the real-time heat production rate prediction value Q. pred ; A dual-closed-loop control unit is used to construct a feedforward-feedback dual-closed-loop controller, based on the real-time heat production rate prediction value Q. pred As input, the feedforward cooling power P is calculated using a model predictive control algorithm based on a feedforward controller. ff The highest measured temperature Tmax The maximum temperature difference ΔT is taken as input, and the feedback correction power P is calculated by the feedback controller based on the adaptive PID algorithm. fb and P ff With P fb The final cooling power P is obtained by weighted fusion. cool ; The execution and safety unit is configured to, based on the final cooling power P cool The cooling temperature is dynamically adjusted, and a safety fallback strategy is implemented based on a preset temperature safety threshold.

[0042] Based on the same inventive concept, embodiments of the present invention also provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described thereon.

[0043] Based on the same inventive concept, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described thereon.

[0044] Example embodiments have been disclosed herein, and while specific terminology has been used, it is for illustrative purposes only and should be construed as such, and is not intended to be limiting. In some instances, it will be apparent to those skilled in the art that features, characteristics, and / or elements described in conjunction with particular embodiments may be used alone, or in combination with features, characteristics, and / or elements described in conjunction with other embodiments, unless otherwise expressly indicated. Therefore, those skilled in the art will understand that various changes in form and detail may be made without departing from the scope of the invention as set forth in the appended claims.

Claims

1. A battery thermal management energy-saving control method based on feedforward-feedback coupling, characterized in that, include: S1 collects lithium-ion battery operating data, including current I, voltage V, and surface temperature T. s Ambient temperature T a And a physical heat production model was constructed based on the Bernardi heat production equation; S2, Construct a hybrid heat production prediction model, which includes the physical heat production model and a data-driven model. The data-driven model predicts the heat production residual based on historical operating data using a machine learning algorithm. The output of the physical heat production model is fused with the heat production residual to obtain the real-time heat production rate prediction value Q. pred ; S3, construct a feedforward-feedback dual closed-loop controller, using the real-time heat generation rate prediction value Q. pred As input, the feedforward cooling power P is calculated using a model predictive control algorithm based on a feedforward controller. ff The highest measured temperature T max The maximum temperature difference ΔT is taken as input, and the feedback correction power P is calculated by the feedback controller based on the adaptive PID algorithm. fb and P ff With P fb The final cooling power P is obtained by weighted fusion. cool ; S4, based on the final cooling power P cool The cooling temperature is dynamically adjusted, and a safety fallback strategy is implemented based on a preset temperature safety threshold.

2. The method according to claim 1, characterized in that, The physical heat generation model in S1 is as follows: ； in, The physical model defines the heat generation rate, where E is the open-circuit voltage and T is the weighted average of multiple temperature measurement points on the battery surface. The entropy coefficient is obtained by differentiating the open-circuit voltage curves at different temperatures.

3. The method according to claim 2, characterized in that, The physical heat production model in S2 outputs the calculated heat production rate Q. phy The data-driven model employs a temporal convolutional network (TCN) and a multi-head attention mechanism to extract temporal features from the input features and predicts the actual real-time heat generation rate Q of the battery. real Compared with the calculated value Q from the physical heat production model phy The heat generation residual ε res =Q real Q phy Input features include I, V, and T at the current time. s T a The data includes the corresponding sliding window statistics, difference values, and frequency domain components; finally, the real-time predicted heat production rate Q is output through an adaptive fusion gate. pred : ； Where λ∈[0,1] is the adaptive fusion weight, which is dynamically calculated by the fusion weight network according to the current working conditions.

4. The method according to claim 1, characterized in that, The feedforward controller in S3 employs Model Predictive Control (MPC), and its objective function is: ; Where, N p To predict the time domain, ,β, T represents the weighting coefficient. target For the target temperature, P cool (k) represents the final cooling power at time k, T pred (k) is based on a simplified thermal dynamics model and Q. pred The predicted temperature at time k, ΔP, is calculated. cool (k) represents the change in cooling power at time k.

5. The method according to claim 4, characterized in that, The feedback controller in S3 uses an adaptive PID controller, and its output is: ； in, =T max -T target For temperature deviation, K p K i K d K ΔT K is the gain coefficient. ΔT Used for temperature difference compensation.

6. The method according to claim 5, characterized in that, The P ff With P fb The weighted fusion method is as follows: ； Where, ω ff ω fb For dynamic weights and ω ff +ω fb =1.

7. The method according to claim 1, characterized in that, The safety fallback strategy in S4 includes three levels of criteria: Level 1: When T max <T warn At that time, feedforward control is the primary method, and the economic cooling mode is executed. Level 2: When T warn ≤T max <T crit Or when 5℃≤ΔT<10℃, strengthen feedback control, start active balancing and limit charging and discharging power; Level 3: When T max ≥T crit If ΔT ≥ 10℃, immediately activate the maximum cooling mode and disconnect the charging / discharging circuit; Among them, T warn As the warning temperature threshold, T crit T is the critical temperature threshold. warn =40℃, T crit =50℃.

8. A battery thermal management energy-saving control system based on feedforward feedback coupling for implementing the method of any one of claims 1-7, characterized in that, The system includes: The data acquisition unit is used to collect operating data of the lithium-ion battery, including current I, voltage V, and surface temperature T. s Ambient temperature T a And a physical heat production model was constructed based on the Bernardi heat production equation; A hybrid prediction unit is used to construct a hybrid heat production prediction model, which includes the physical heat production model and a data-driven model. The data-driven model predicts the heat production residual based on historical operating data using a machine learning algorithm. The output of the physical heat production model is fused with the heat production residual to obtain the real-time heat production rate prediction value Q. pred ; A dual-closed-loop control unit is used to construct a feedforward-feedback dual-closed-loop controller, based on the real-time heat production rate prediction value Q. pred As input, the feedforward cooling power P is calculated using a model predictive control algorithm based on a feedforward controller. ff The highest measured temperature T max The maximum temperature difference ΔT is taken as input, and the feedback correction power P is calculated by the feedback controller based on the adaptive PID algorithm. fb and P ff With P fb The final cooling power P is obtained by weighted fusion. cool ; The execution and safety unit is configured to, based on the final cooling power P cool The cooling temperature is dynamically adjusted, and a safety fallback strategy is implemented based on a preset temperature safety threshold.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the method described in any one of claims 1-7.

Images