A battery grading temperature control method, system and device based on mahalanobis distance early warning

Abstract

The application provides a battery grading temperature control method, system and device based on Mahalanobis distance early warning, which comprises the following steps: obtaining and preprocessing the operation parameters of the battery and each actuator, constructing a multi-dimensional state feature vector; using a preset Mahalanobis distance calculation model to calculate the average Mahalanobis distance, and mapping it as a thermal runaway comprehensive risk index; pre-judging the control state level of the current battery; obtaining the PID control parameters through a preset RBF neural network; based on the control state level of the current battery, using the PID control parameters to calculate the expected control amount vector of each actuator, and using it to calculate the actual control amount vector of each actuator; taking the actual control amount vector as a temperature control signal to control the current actuator to perform temperature control processing on the battery. Through the above method, the problems of weak temperature control ability, insufficient thermal suppression ability and poor control strategy adaptability of the traditional BMS in thermal runaway prevention and control can be solved.

Description

Technical Field

[0001] This invention relates to the field of battery safety and thermal management technology, and in particular to a battery graded temperature control method and system based on Mahalanobis distance early warning. Background Technology

[0002] As application scenarios place increasingly higher demands on battery safety and response speed, battery management systems (BMS) are evolving towards higher safety and faster response. However, traditional systems still have the following problems:

[0003] In terms of battery temperature control, traditional BMS mostly rely on surface temperature sensors or single voltage fluctuations for abnormal alarms. When the system detects an abnormal signal, the battery has often already entered an irreversible thermal runaway stage.

[0004] In terms of thermal runaway testing, existing methods usually directly heat the battery until it catches fire and explodes, making it difficult to implement effective non-destructive intervention and suppression at the "thermal runaway critical point (thermal boundary)," resulting in a lack of means for safety control research in the critical state;

[0005] At the thermal management execution level, traditional air-cooled or passive water-cooled systems have large thermal inertia and slow cooling rates. Faced with the drastic temperature rise in the early stages of local thermal runaway, they cannot provide the ability to "force thermal suppression" with instantaneous high power, making it difficult to quickly curb the spread of heat.

[0006] In terms of control strategies, traditional temperature control algorithms are mostly based on fixed parameter designs, which are difficult to adapt to the strong nonlinearity and hysteresis characteristics of heat generation during battery thermal runaway, and are prone to failure. At the same time, the traditional control architecture does not independently allocate computing power to multiple actuators, resulting in sluggish system response and easy oscillations during the temperature regulation process.

[0007] It is evident that traditional BMS has significant shortcomings in battery temperature control capabilities, testing intervention capabilities, thermal management response speed, and control strategy adaptability, making it difficult to meet the development requirements of high safety and high responsiveness. Summary of the Invention

[0008] To address the shortcomings of traditional battery management systems (BMS) such as weak temperature control, insufficient heat suppression capabilities, and poor adaptability of control strategies, this paper proposes a battery graded temperature control method, system, and device based on Mahalanobis distance early warning. This method calculates the control vectors of each actuator as temperature control signals to process the battery temperature. Furthermore, it utilizes the real-time operating parameters of the battery and each actuator to update the parameters required by the RBF neural network and incremental PID controller in real time, achieving precise temperature control based on changes in battery temperature.

[0009] In a first aspect, according to the design scheme provided by the present invention, a battery graded temperature control method based on Mahalanobis distance early warning is provided, comprising the following:

[0010] The operating parameters of the battery and each actuator are acquired, and the operating parameters are preprocessed to construct a multi-dimensional state feature vector. The average Mahalanobis distance is calculated using the multi-dimensional state feature vector and a preset Mahalanobis distance calculation model. The average Mahalanobis distance is then mapped to a comprehensive thermal runaway risk index.

[0011] Multiple control state levels are preset and different threshold ranges are set for each control state level. The current control state level of the battery is determined based on the threshold range of the thermal runaway comprehensive risk index.

[0012] PID control parameters are obtained through a preset RBF neural network; based on the current control state level of the battery, the expected control vector of each actuator is calculated using the PID control parameters, and the actual control vector of each actuator is calculated based on the expected control vector. The actual control vector is used as a temperature control signal to control the current actuator to perform temperature control processing on the battery.

[0013] Furthermore, the calculation of the average Mahalanobis distance using the multidimensional state feature vector and the preset Mahalanobis distance calculation model specifically includes:

[0014] Obtain the historical operating parameters of the battery in normal operating mode, and calculate and store the normal baseline mean vector based on the historical operating parameters. The normal baseline covariance matrix is ​​obtained based on the normal baseline mean vector. To construct a Mahalanobis distance calculation model:

[0015]

[0016] in, Represents Mahalanobis distance, Represents a multidimensional state feature vector. This represents the normal baseline covariance matrix after regularization. Indicates the discrete sampling sequence number;

[0017] Based on the Mahalanobis distance calculation model, the Mahalanobis distance of the current battery is obtained, and then the Mahalanobis distance of the current battery is smoothed by a window to obtain the average Mahalanobis distance. :

[0018]

[0019] in, This indicates the preset length of the sliding window.

[0020] Furthermore, the comprehensive risk index for thermal runaway is expressed as:

[0021]

[0022] in, This indicates the comprehensive risk index of thermal runaway. Indicates the discrete sampling sequence number. This represents the sensitivity adjustment coefficient. This represents the average Mahalanobis distance. This represents the upper limit of normal deviation determined based on normal samples.

[0023] Furthermore, the acquisition of PID control parameters through a preset RBF neural network specifically includes:

[0024] Within each control cycle, the operating parameters of the battery are normalized. The normalized operating parameters and the actual control vector of each actuator in the previous cycle are used as input data for the RBF neural network. The input data is mapped into feature vectors in a high-dimensional feature space through radial basis functions. Based on the inner product of the feature vectors and the preset RBF network weights, network prediction data is generated.

[0025] The cross-Jacobi matrix of the actuator is calculated using the network prediction data. Based on the cross-Jacobi matrix and the preset PID adaptive learning rate, the PID control parameters of each actuator are tuned in real time using the gradient descent method.

[0026] Furthermore, the method of obtaining PID control parameters through a preset RBF neural network also includes:

[0027] The rate of change of Mahalanobis distance was calculated using the average Mahalanobis distance. Through the Mahalanobis distance change rate Dynamically adjust the learning rate of the RBF neural network weights and the adaptive learning rate of the PID controller. :

[0028]

[0029] in, , Indicates the weights of the RBF network. This represents the adaptive learning rate of the PID controller. This represents the adjustment coefficient. Indicates the discrete sampling sequence number.

[0030] Furthermore, the calculation of the desired control vector for each actuator using the PID control parameters specifically includes:

[0031] Based on the current control state level of the battery, the target temperature reached by the battery is obtained; the deviation between the current battery temperature and the target temperature reached is calculated. An incremental PID controller is input; the incremental PID controller calculates the incremental control quantity based on the PID control parameters. And through the incremental control quantity Calculate the desired control vector for each actuator.

[0032] The incremental control quantity Represented as:

[0033]

[0034] in, Indicates actuator Incremental control quantity, proportion ,integral ,differential Indicates actuator PID control parameters.

[0035] Furthermore, the calculation of the actual control vector of each actuator based on the desired control vector specifically includes:

[0036] Based on the current control state level of the battery and the comprehensive thermal runaway risk index, the control weights corresponding to each actuator are calculated using a continuous mapping function. According to each control weight Get control allocation matrix ;

[0037] Using the control allocation matrix and the expected control vector of each actuator, the actual control vector of each actuator is obtained, and the actual control vector is saturated and limited to obtain the final actual control vector.

[0038] The control weights corresponding to each actuator Represented as:

[0039]

[0040] Where S represents the control state of the battery, and S0 represents a control state level. Indicates actuator The minimum allowed weight, Indicates the actuator in state S The maximum allowable weight, R represents the comprehensive risk index of thermal runaway. Indicates actuator The risk response inflection point under state S; This indicates the sensitivity to changes in weights.

[0041] Furthermore, the calculation of the actual control vector for each actuator also includes:

[0042] The actuator includes at least three types of actuators; the heat dissipation flow rate of the first and second actuators and the hot end temperature of the third actuator are obtained; a preset dynamic safety lower limit constraint is applied to the increment of the actual control vector of the first and second actuators; the operation of each actuator is controlled using the actual control vector after the constraint is applied; so as to avoid the situation where the heat dissipation flow rate of the first and second actuators is mismatched with the rate of increase of the hot end temperature of the third actuator, resulting in the hot end temperature of the third actuator being higher than the battery temperature.

[0043] Secondly, the present invention also provides a battery grading temperature control system based on Mahalanobis distance early warning, comprising:

[0044] Data processing module: used to acquire the operating parameters of the battery and each actuator, preprocess the operating parameters to construct a multi-dimensional state feature vector; use the multi-dimensional state feature vector and a preset Mahalanobis distance calculation model to calculate the average Mahalanobis distance; and map the average Mahalanobis distance to a comprehensive thermal runaway risk index.

[0045] Control state determination module: used to pre-set multiple control state levels and set different threshold ranges for each control state level, and determine the current control state level of the battery based on the threshold range of the thermal runaway comprehensive risk index;

[0046] Temperature control module: used to obtain PID control parameters through a preset RBF neural network; based on the current control state level of the battery, calculate the expected control vector of each actuator using the PID control parameters, and calculate the actual control vector of each actuator according to the expected control vector, and use the actual control vector as a temperature control signal to control the current actuator to perform temperature control processing on the battery.

[0047] Thirdly, the present invention also provides a battery grading temperature control device based on Mahalanobis distance warning, comprising a battery box and batteries located inside the battery box, and further comprising a temperature control system based on Mahalanobis distance warning as described in the second aspect and a heat dissipation actuator located above the battery box. The heat dissipation actuator includes a thermoelectric cooler, a water-cooled plate located above the thermoelectric cooler, and a cooling fan located above the water-cooled plate. The battery grading temperature control system is electrically connected to each heat dissipation actuator and is used to send control signals to the heat dissipation actuator to adjust the temperature of the battery.

[0048] The beneficial effects of this invention are:

[0049] (1) Based on the battery temperature trend, this invention uses a temperature control method that includes the average Mahalanobis distance and the thermal runaway risk level to control multiple actuators for temperature control. It performs real-time temperature control according to the actual situation of the battery. On this basis, an anti-thermal backflow mechanism is also set up to prevent the battery from entering the irreversible thermal runaway stage, which solves the problems of weak temperature control capability and insufficient thermal suppression capability of traditional temperature control systems. At the same time, this invention sets the actual control quantity of each actuator, so that each actuator performs cooling operation according to the preset control quantity, and optimizes the control strategy of BMS by independently allocating the control quantity of multiple actuators.

[0050] (2) Real-time acquisition of the operating parameters of the battery and each actuator, and calculation of the average Mahalanobis distance required for the temperature control algorithm based on the current operating parameters; the average Mahalanobis distance is used as a dynamic parameter to adjust the temperature control parameters during the operation of the algorithm in a timely manner, so that the system can adapt to the actual working condition changes and make corresponding temperature control responses, thus solving the problem of poor adaptability of traditional BMS control strategies. Attached Figure Description

[0051] Figure 1 This is a flowchart illustrating a battery grading temperature control method based on Mahalanobis distance early warning in an embodiment of the present invention.

[0052] Figure 2 This is a state machine architecture diagram of an event-driven battery grading temperature control method based on Mahalanobis distance early warning in an embodiment of the present invention.

[0053] Figure 3 This is a schematic diagram of the event determination algorithm in a battery graded temperature control method based on Mahalanobis distance early warning in an embodiment of the present invention;

[0054] Figure 4 This is a schematic diagram of a battery grading temperature control system architecture based on Mahalanobis distance early warning in an embodiment of the present invention;

[0055] Figure 5 This is an overall structural diagram of a battery grading temperature control device based on Mahalanobis distance early warning in an embodiment of the present invention;

[0056] Figure 6 This is a flow topology diagram of a battery grading temperature control device based on Mahalanobis distance early warning in an embodiment of the present invention;

[0057] Figure 7 This is a cross-sectional view of the battery interior of a battery grading temperature control device based on Mahalanobis distance warning, as described in an embodiment of the present invention.

[0058] Figure 8 This is a battery temperature control curve diagram of a battery graded temperature control method based on Mahalanobis distance early warning in an embodiment of the present invention.

[0059] Figure 9 This is a comprehensive risk index warning curve of a battery graded temperature control method based on Mahalanobis distance warning in an embodiment of the present invention;

[0060] Figure 10 This is a timing diagram of the event-driven state machine transition for a battery grading temperature control method based on Mahalanobis distance early warning in an embodiment of the present invention.

[0061] In the attached diagram, the following labels are used: 1 represents the battery box; 2 represents the battery; 3 represents the thermally conductive copper plate; 4 represents the cooling fan; 5 represents the thermoelectric cooler; 6 represents the water-cooled plate; 7 represents the water-cooling inlet; 8 represents the heat sink; 9 represents the screw hole; 10 represents the battery box frame; 11 represents the flexible heating film; and 12 represents the temperature sensor. Detailed Implementation

[0062] To make the objectives, technical solutions, and advantages of this invention clearer and more understandable, the technical solutions of the embodiments of this invention will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0063] like Figure 1 As shown, this embodiment of the invention provides a battery graded temperature control method based on Mahalanobis distance warning, comprising:

[0064] S101. Obtain the operating parameters of the battery and each actuator, preprocess the operating parameters to construct a multi-dimensional state feature vector; use the multi-dimensional state feature vector and a preset Mahalanobis distance calculation model to calculate the average Mahalanobis distance; and map the average Mahalanobis distance to a comprehensive thermal runaway risk index.

[0065] Furthermore, the operating parameters of the battery to be cooled include the following physical quantities: measured values ​​from the battery surface temperature sensor array. ( ), actual concentration of characteristic gases (such as CO, VOC) Battery module charging and discharging current Battery voltage and the temperature at the interface between the hot end of the TEC and the water-cooled plate. Among them, each physical quantity is based on the sampling period. Real-time synchronous data collection.

[0066] Furthermore, the operating parameters are preprocessed, firstly by performing median filtering on the data collected by the temperature sensor:

[0067]

[0068] in, This represents the original measurement value of the j-th temperature sensor. The median filter window length can be 3 or 5; `median` indicates that the median should be used. Indicates the discrete sampling sequence number.

[0069] To avoid the average temperature masking local hot spots, the highest temperature in the temperature sensor array is used as the representative temperature of the battery surface. The battery temperature rise rate is then expressed as:

[0070]

[0071] in, The number of step sampling points calculated for the temperature rise rate. The actual calculation time interval for the temperature rise rate. Indicates the sampling period; when At that time, the battery temperature rise rate is set to 0.

[0072] Therefore, the temperature difference on the battery surface is defined as:

[0073]

[0074] Define the rate of voltage change as:

[0075]

[0076] in, The number of differential interval sampling points used to calculate the rate of voltage change. This is the actual time interval for calculating the rate of voltage change.

[0077] A real-time multidimensional state feature vector is constructed using the battery surface temperature difference, battery temperature rise rate, actual concentration of characteristic gases, battery module charging and discharging current, and voltage change rate.

[0078]

[0079] Furthermore, the calculation of the average Mahalanobis distance using the multidimensional state feature vector and the preset Mahalanobis distance calculation model specifically includes:

[0080] Continuous data collection was performed when there were no thermal triggering events, no sensor malfunctions, and the battery was in normal operating condition. Group of normal samples: And calculate the normal baseline mean vector:

[0081] ;

[0082] Based on the normal baseline mean vector, define the normal baseline covariance matrix:

[0083]

[0084] Based on this, to ensure matrix invertibility, regularization is introduced for the normal baseline covariance matrix:

[0085]

[0086] in, For a smaller regularization coefficient, It is an identity matrix.

[0087] As can be seen, the real-time Mahalanobis distance The calculation model is as follows:

[0088]

[0089] Mahalanobis distance uses the covariance matrix of normal samples to correct the correlation of features with different dimensions, and can comprehensively measure the degree of deviation of the current state from the normal state distribution. When multiple correlated features deviate from the normal distribution simultaneously, the comprehensive distance will increase accordingly.

[0090] Specifically, to prevent false triggering caused by single-point noise, the calculated Mahalanobis distance is divided into segments with a length of [missing information]. Sliding window averaging: It is the preset sliding window length, representing the number of Mahalanobis distance samples participating in the averaging, usually taken as 3-5.

[0091]

[0092] Then, the smoothed Mahalanobis distance A comprehensive risk index for continuous thermal runaway with a nonlinear mapping of 0-100%:

[0093]

[0094] in, This represents the comprehensive risk index for thermal runaway, and its value range is... , This is the sensitivity adjustment coefficient used to control the level of sensitivity in early warning systems. This is the upper limit of normal deviation determined based on normal samples.

[0095] S102. Pre-set multiple control state levels and set different threshold ranges for each control state level. Determine the current control state level of the battery based on the threshold range of the thermal runaway comprehensive risk index.

[0096] Furthermore, to prevent frequent start-stop operations of the actuator due to slight temperature fluctuations around the threshold, a hysteresis range is introduced. A three-level rise trigger threshold is preset: (Level 1 trigger) (Level 2 trigger) (Level 3 trigger); and satisfy the following conditions: And preset three levels of descent backoff thresholds: (Level 1 trigger) (Level 2 trigger) (Level 3 trigger); and satisfy the following conditions: .

[0097] And design constraints for each threshold: , , (difference) This is the hysteresis safety range); the response level is determined by the risk value. This is determined in conjunction with the current battery control state, as follows:

[0098] Specifically, the actuator includes air-cooled, liquid-cooled, and semiconductor coolers; therefore, the battery response level during the rising trigger phase is:

[0099] when At this time, the battery control state is in monitoring state S0;

[0100] when When the battery control state transitions to state S1, it triggers a first-level response: medium power operation for air cooling, and low power operation for liquid cooling and semiconductor cooler.

[0101] when When the battery control state transitions to the S2 state, a secondary response is triggered, and the air-cooled and liquid-cooled systems operate at high power, while the semiconductor cooler operates at medium power.

[0102] when When the battery control state transitions to S3, a three-level response is triggered, which increases the composite cooling output under the safety constraint of the TEC hot end and quickly suppresses abnormal temperature rise.

[0103] Specifically, the response levels for the descent / rollback phase include: [List of levels would be inserted here] This is the temperature flag for the rollback state;

[0104] The battery control status is in S3. and When the time comes, it reverts to the S2 state, cancels the third-level response and enters the second-level response: the semiconductor cooler smoothly reduces from full load to medium power, while air cooling and liquid cooling maintain high power;

[0105] The battery control state is in S2. and When the time comes, it reverts to state S1, cancels the secondary response and enters the primary response: high-power air cooling is used as the main heat dissipation method, while liquid cooling and semiconductor coolers operate at low power.

[0106] The battery control state is in S1. and At that time, revert to The state, that is, returning to normal: air-cooled, liquid-cooled and semiconductor coolers all return to standby mode.

[0107] S103. Obtain PID control parameters through a preset RBF neural network; based on the current control state level of the battery, calculate the expected control vector of each actuator using the PID control parameters, and calculate the actual control vector of each actuator according to the expected control vector. Use the actual control vector as a temperature control signal to control the current actuator to perform temperature control processing on the battery.

[0108] Furthermore, the acquisition of PID control parameters through a preset RBF neural network specifically includes:

[0109] Within each control cycle, the operating parameters of the battery are normalized. The normalized operating parameters and the actual control vector of each actuator in the previous cycle are used as input data for the RBF neural network. The input data is mapped into feature vectors in a high-dimensional feature space through radial basis functions. Based on the inner product of the feature vectors and the preset RBF network weights, network prediction data is generated.

[0110] The cross-Jacobi matrix of the actuator is calculated using the network prediction data. Based on the cross-Jacobi matrix and the preset PID adaptive learning rate, the PID control parameters of each actuator are tuned in real time using the gradient descent method.

[0111] Specifically, to address the nonlinearity and hysteresis of thermal runaway, a multi-input multi-output (MIMO) RBF neural network is introduced to simultaneously predict the battery surface temperature. and the temperature at the interface between the hot end of the TEC and the water-cooled plate. .in, It represents the common thermal state of the TEC (Thermal Design Equipment) under the combined effects of heat generation, liquid cooling heat dissipation, and air cooling heat dissipation. It is used to determine whether there is a risk of heat accumulation and backflow at the hot end of the TEC, serves as an auxiliary safety state input for RBF (Radio Refrigerant Function) temperature prediction, and is also used to make safety corrections to the control quantities of air cooling, liquid cooling, and thermoelectric coolers.

[0112] The input vector of the RBF neural network:

[0113]

[0114] The input control quantity is the final control quantity after continuous weight allocation, safety constraints, and saturation limiting. This represents the actual heating power of the thermal induction module during the previous control cycle.

[0115] The RBF neural network predicts the output vector:

[0116]

[0117] Because temperature, heating power, and actuator control quantities have different dimensions and numerical ranges, each input variable is normalized before being input into the RBF network. For the q-th input variable:

[0118]

[0119] The normalized network input vector is then:

[0120]

[0121] in, and These are the lower and upper limits of the calibration for the q-th input variable, respectively.

[0122] Furthermore, the hidden layer radial basis function is expressed as:

[0123]

[0124] in, Indicates the center of the j-th basis function. Indicates the width of the j-th basis function;

[0125] The network prediction output is then expressed as:

[0126]

[0127] in, Let be the hidden layer output vector of the RBF neural network, and The formula for updating the output weights is:

[0128]

[0129] in, This is the output prediction error vector of the RBF network. The data is obtained after processing by sensor measurements. The learning rate is the weight. Use very small positive numbers to prevent the denominator from approaching zero.

[0130] Furthermore, by taking the partial derivative of the RBF network output, the cross-Jacobi matrix of the multiple actuators for the dual-temperature target is extracted in real time:

[0131]

[0132] Based on the first row of the Jacobian matrix (sensitivity to battery temperature), the incremental PID parameters of each actuator are tuned in real time using the gradient descent method, and the performance indicators are... The update rules are as follows:

[0133]

[0134]

[0135]

[0136] in These are the adaptive learning rates for the proportional, integral, and derivative parameters, respectively. Initial settings are provided. These are the calibration parameters for the i-th actuator under safe operating conditions. Furthermore, to prevent abnormal parameter updates, the PID parameters should be limited to a preset range. .

[0137] Furthermore, to improve the RBF thermal model's ability to track rapid changes in thermal state, a fixed learning rate is abandoned, and a Mahalanobis distance rate of change is introduced to dynamically adjust the learning rate of the RBF network weights. The learning rate for PID tuning specifically includes:

[0138]

[0139] in, , The adjustment coefficient; the rate of change of the Mahalanobis distance. for:

[0140]

[0141] It is evident that when the battery control state is in a safe state such as S0 or S1, The network uses a basic learning rate Operation, noise filtering, and stability assurance; when the battery control state is at level S2 or S3. As the learning rate increases, the nonlinear thermal model can be approximated quickly with a very high step size, thus shortening the delay time.

[0142] Furthermore, the calculation of the desired control vector for each actuator using the PID control parameters specifically includes:

[0143] Set target temperature Actual battery temperature ,error .

[0144] The smoothed target temperature function is given by the battery control states S1-S3:

[0145]

[0146] During initialization, Indicates battery temperature in S0 state. It is only used to record the current battery temperature for subsequent smoothing. The smoothed target temperature function can avoid instantaneous jumps in the PID control output caused by state switching. For upper and lower limit truncation functions:

[0147]

[0148] The target temperature without smoothing:

[0149]

[0150] The maximum allowable change in target temperature within a single sampling period, and satisfying the following conditions: :

[0151]

[0152] The target temperature in each state satisfies In other words, the higher the battery temperature risk level, the lower the target temperature used. For ambient temperature, The upper limit of safe temperature is determined based on the battery type.

[0153] For a certain activated actuator Set initial value Its incremental PID output is:

[0154]

[0155]

[0156] Proportion ,integral ,differential It is tuned in real time by the RBF neural network.

[0157] Furthermore, the calculation of the actual control vector of each actuator based on the desired control vector specifically includes:

[0158] Let the air-cooled, liquid-cooled, and thermoelectric coolers be represented by the subscripts F, L, and T, respectively. The original desired control vector calculated by the subsequent RBF adaptive PID controller is:

[0159]

[0160] This represents the original or theoretical expected control quantity given by the PID controller before considering risk weights and safety constraints.

[0161] Introducing a heterogeneous actuator continuous control allocation matrix jointly determined by the thermal runaway comprehensive risk index R and the battery control state. :

[0162]

[0163] in, These are the control weights for air-cooled, liquid-cooled, and thermoelectric coolers, respectively. For any actuator Its weights are calculated using a Logistic continuous mapping function with state constraints:

[0164]

[0165] In the formula, Let be the minimum allowed weight of actuator i in state S; The maximum allowable weight of actuator i in state S; This represents the risk response inflection point of actuator i in state S; For sensitivity to weight changes, This is a normalized comprehensive risk index. Therefore, the weights of each actuator always satisfy... ,and: .

[0166] The control quantity after continuous weight allocation is:

[0167]

[0168] This formula represents the control quantity after allocating the outputs of the three PID controllers based on the risk index and the state machine state. Considering the inherent physical limits of the actuators, a saturation limit is applied to the final actual control quantity:

[0169] ,

[0170] Here This refers to the final actual control quantity. This represents the maximum control quantity allowed for the i-th actuator.

[0171] As an implementation method, the risk response inflection point of the air-cooled actuator is set in a lower range: air cooling has the characteristics of fast start-up speed and low power consumption, so its risk response inflection point is set in a lower range, giving priority to the use of air cooling's low power consumption convection heat dissipation capability.

[0172] The risk response inflection point of liquid-cooled actuators is set in the medium-risk range: liquid cooling has a large continuous heat dissipation capacity, but there is a certain inertia of water pumps and pipelines. Therefore, its risk response inflection point is set in the medium-risk range, and the output gradually increases as the risk index rises.

[0173] The risk response inflection point of a semiconductor refrigerator actuator should be set in a higher range: While semiconductor refrigerators can achieve active heat pump cooling, their hot-end heat dissipation includes cold-end heat absorption and their own electrical power consumption. Therefore, their weight is constrained not only by the risk index but also by the current state and hot-end safety conditions. Their risk response inflection point should be set in a higher range, with the limiter controlling the maximum weight in states S1 and S2; and allowing for a rapid increase in output in state S3.

[0174] Furthermore, the step of calculating the actual control vector of each actuator based on the desired control vector also includes:

[0175] The actuator includes at least three types of actuators; the heat dissipation sensitivity of the first and second actuators to the hot end temperature of the third actuator and the hot end temperature of the third actuator are respectively acquired; a preset dynamic safety lower limit constraint is applied to the increment of the actual control vector of the first and second actuators using the heat dissipation sensitivity; the operation of each actuator is controlled using the actual control vector after the constraint is applied; so as to avoid the situation where the heat dissipation capacity of the first and second actuators is mismatched with the rate of increase of the hot end temperature of the third actuator, resulting in the hot end temperature of the third actuator being higher than the battery temperature.

[0176] The first and second actuators are the liquid-cooled and air-cooled actuators, respectively, and the third actuator is a semiconductor refrigeration actuator. Due to the physical characteristics of semiconductor refrigeration, the heat dissipated at its hot end includes absorbed heat and its own electrical power consumption. When the battery temperature is in a high-risk range, if the heat dissipation capacity of liquid cooling and air cooling does not match the surge in semiconductor power, the hot end temperature will far exceed the battery temperature, causing heat backflow.

[0177] In this case, the sensitivity symbol of the hot end to the actuator is defined as follows: (Heating effect) and This means setting an upper limit for the safe temperature of the hot end. , (Heat dissipation effect).

[0178] Introducing a heat flux forced matching constraint: to ensure increased semiconductor cooling The hot end does not heat up The control increments for air cooling and liquid cooling must meet the following dynamic safety lower limit constraints:

[0179]

[0180] If the above dynamic matching constraints are not met, the control quantities of liquid cooling and air cooling will be increased first to increase the heat dissipation capacity. If the constraints still cannot be met after both liquid cooling and air cooling have reached their maximum allowable output, the control quantity of TEC will be limited or reduced, and the hot end over-temperature protection will be triggered.

[0181] As one possible implementation method, such as Figure 2 The event-driven state machine architecture diagram shown features a central controller with built-in data caching and feature extraction layers. This transforms multi-dimensional physical and electrochemical data into event flags, driving state machine transitions. Specifically, it includes:

[0182] Standby / Monitoring (S0 state): The system is in sleep mode at this time, maintaining only low-power data acquisition.

[0183] Event E1 (heat accumulation event) triggers the S1 transition state. When it is determined that the temperature is rising slowly, it enters S1, with air cooling as the heat dissipation method and liquid cooling and low-power semiconductor cooler as auxiliary.

[0184] Event E2 (high load overheating event) triggers state S2. When a large temperature rise is detected, state S2 is entered, with air cooling and liquid cooling as the primary heat dissipation methods, and medium-power auxiliary heat dissipation in the thermoelectric cooler.

[0185] Event E3 (thermal runaway critical event) triggers state S3. When a sudden increase in characteristic gas, a sharp drop in voltage, and a predicted temperature exceeding the threshold are detected, the system transitions to state S3, triggering a three-stage forced cooling suppression. This instantly and significantly increases the power of the semiconductor cooler and, in conjunction with liquid cooling and air cooling, forcibly suppresses the temperature.

[0186] Hysteresis rollback mechanism: When the extreme suppression takes effect, the temperature drops and the environment is safe, the current event level is re-determined through status feedback, and the system rolls back to S2, S1 and S0 step by step to prevent the system from frequently starting and stopping and oscillating.

[0187] like Figure 3 The event determination algorithm flowchart shown includes collecting normal baseline data to calculate the Mahalanobis distance calculation model, then performing multi-dimensional feature extraction to calculate the current Mahalanobis distance, then smoothing the Mahalanobis distance with a sliding window to obtain the average Mahalanobis distance, mapping the average Mahalanobis distance to a risk index mapping, and performing threshold comparison on the risk index mapping to update the event flag and battery temperature control status level, thereby controlling the battery temperature or outputting prompt information.

[0188] This invention utilizes a temperature control method incorporating average Mahalanobis distance and thermal runaway risk levels to control multiple actuators for temperature control based on battery temperature trends. Real-time temperature control is performed according to the actual battery conditions. Furthermore, an anti-backflow mechanism is implemented to prevent the battery from entering an irreversible thermal runaway phase. Simultaneously, the invention sets the actual control values ​​for each actuator, allowing each actuator to perform cooling operations according to preset control values, thus enabling independent allocation of control values ​​across multiple actuators. Additionally, a heating film is used during the testing phase to simulate temperature rise in the temperature-controlled battery, implementing a gentler heating method and improving the testing temperature rise effect.

[0189] Example 2

[0190] like Figure 4 As shown, the present invention also provides a battery grading temperature control system based on Mahalanobis distance early warning, comprising:

[0191] Data processing module: used to acquire the operating parameters of the battery and each actuator, preprocess the operating parameters to construct a multi-dimensional state feature vector; use the multi-dimensional state feature vector and a preset Mahalanobis distance calculation model to calculate the average Mahalanobis distance; and map the average Mahalanobis distance to a comprehensive thermal runaway risk index.

[0192] Control state determination module: used to pre-set multiple control state levels and set different threshold ranges for each control state level, and determine the current control state level of the battery based on the threshold range of the thermal runaway comprehensive risk index;

[0193] Temperature control module: used to obtain PID control parameters through a preset RBF neural network; based on the current control state level of the battery, calculate the expected control vector of each actuator using the PID control parameters, and calculate the actual control vector of each actuator according to the expected control vector, and use the actual control vector as a temperature control signal to control the current actuator to perform temperature control processing on the battery.

[0194] As one possible implementation method, the specific implementation process of this system may include:

[0195] First, thermal induction is initiated; the central controller activates the heating film to simulate localized thermal runaway and temperature rise in single or multiple battery cells. A sliding window buffer extracts the voltage drop rate and temperature gradient in real time. The heating film is a polyimide heating film. (This step is used to test the temperature control algorithm; if the heating process is not required, you can skip to the next step.)

[0196] Next, an event transition and alarm mechanism is implemented. Through the data processing module and the control state determination module, the thermal runaway critical event (E3) is determined, and the control state of the state machine instantly transitions from S0 to S3. Simultaneously, a multi-dimensional monitoring and prediction process is carried out. The BMS board transmits real-time data on individual cell voltage, internal resistance, operating current, and equalization anomalies. The Mahalanobis distance algorithm integrates multiple feature quantities to output and feed back the risk index in real time.

[0197] Finally, adaptive temperature control is implemented. The heterogeneous controller in the temperature control module receives instructions and dynamically allocates the weights of each actuator based on the current risk index (Sigmoid mapping). The RBF neural network tunes the PID parameters online based on the surge in temperature error, historical control values, and the rate of change of the risk index, driving the multi-channel refrigeration mechanism for composite control and forced cooling. Simultaneously, heat backflow prevention constraints monitor the heat dissipation gap in real time, and if necessary, use a penalty function to forcibly increase the liquid cooling / air cooling output.

[0198] It should be noted that the battery grading temperature control system based on Mahalanobis distance early warning provided in this embodiment of the invention can be specifically referred to in the above method embodiment to achieve the functions described above.

[0199] Example 3

[0200] like Figure 5 as well as Figure 6 As shown, the present invention also provides a battery grading temperature control device based on Mahalanobis distance warning, comprising a battery box 1 and a battery 2 located inside the battery box 1, and further comprising a battery grading temperature control system based on Mahalanobis distance warning as described in Embodiment 2, and a heat dissipation actuator located above the battery box 1. The heat dissipation actuator includes a semiconductor cooling chip 5, a water-cooled plate 6 located above the semiconductor cooling chip 5, and a cooling fan 4 located above the water-cooled plate 6; the battery grading temperature control system is electrically connected to each heat dissipation actuator and is used to send control signals to the heat dissipation actuator to adjust the temperature of the battery.

[0201] The central controller drives the heat dissipation actuator to work based on the temperature control signal fed back by the temperature control system, and can also drive the heating film to heat up.

[0202] Specifically, it also includes a flexible heating film 11 and a multi-dimensional state sensing array, specifically including:

[0203] The flexible heating film 11 is attached to the side and bottom of the real battery module to artificially simulate the high-temperature local hot spots generated by the internal short circuit or high-rate discharge of the battery. The flexible heating film 11 includes a polyimide heating film.

[0204] like Figure 7As shown, the multi-dimensional state sensing array is distributed around the battery module, including smoke sensors and characteristic gas sensors (such as CO and VOC) for detecting the spatial environment; it also includes an array of temperature sensors 12 attached to the battery surface; and a BMS board directly connected to the battery (for real-time acquisition of individual cell voltage, internal resistance, charging and discharging current, and equalization status).

[0205] The temperature control device includes three types of actuators: a semiconductor cooling chip 5, a water-cooled plate 6, and a cooling fan 4. When the battery temperature changes, each actuator performs temperature control processing through the received control signal, which can solve the problem of poor temperature control capability of traditional temperature control devices.

[0206] To verify the effectiveness of the method proposed in this invention, the following experiments were also conducted:

[0207] like Figure 8 As shown, the battery experiences a base temperature rise (slowly increasing from 25°C to approximately 28°C) under normal load, at which point the temperature difference is small and the rate of temperature rise is extremely low. For example... Figure 9 As shown, the algorithm of this invention can effectively filter out normal fever characteristics, and the calculated comprehensive risk index remains consistently below 20%. Figure 10 As shown, the state machine remains stably in the S0 (standby monitoring) state, and the cooling actuator remains in sleep mode. It can be seen that this algorithm has the ability to resist interference during normal temperature rise and does not produce false alarms.

[0208] At the 60-second mark, a sudden high-power, high-load injection is simulated (inducing thermal runaway). Although the battery temperature does not reach the high-temperature absolute threshold of a traditional BMS, its spatial temperature gradient and temperature rise rate undergo a nonlinear abrupt change. The spatial state fusion early warning algorithm instantly captures this anomaly, driving the comprehensive risk index (…). Figure 9 Within seconds, the risk index spiked exponentially to nearly 90%. This high-risk index instantly triggered the state machine. Figure 10 The system jumps to the highest alert level S3 (thermal runaway critical). The central controller instantly activates the composite temperature control execution module to enter the full-power limit suppression state, successfully containing the temperature at around 52°C, achieving a non-destructive physical prevention of early thermal runaway.

[0209] With forced cooling taking effect, the battery temperature gradually decreases. Around 125 seconds, the risk index drops below 65%, and the state machine degrades from S3 back to S2. Around 160 seconds, the temperature decrease briefly slows due to the reduced cooling power, causing a slight rebound in the risk index. A multi-level safety threshold and a dual-condition hysteresis fallback mechanism are introduced, and the state machine ( Figure 10 The temperature remained stable in state S2 without any frequent fluctuations. This stage verifies that the proposed method can eliminate the hardware loss problem caused by frequent start-stop oscillations at the threshold edge in traditional temperature control methods.

[0210] Within approximately 250 seconds, the battery temperature was suppressed to below the 40°C safety threshold. The risk index then experienced a sharp drop before smoothly returning to the 20% safety range, and the state machine smoothly transitioned to state S1. At this point, the heterogeneous continuous control scheduling algorithm controlled the cooling module's output, achieving a safe dynamic thermal equilibrium. This ensured absolute battery safety while avoiding energy waste caused by over-cooling.

[0211] As can be seen, this experiment verifies that the temperature control method provided by the present invention has the ability to resist interference during normal temperature rise, does not produce false alarms, and achieves non-destructive physical blocking of early thermal runaway.

[0212] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A battery graded temperature control method based on Mahalanobis distance early warning, characterized in that, Include: The operating parameters of the battery and each actuator are acquired, and the operating parameters are preprocessed to construct a multi-dimensional state feature vector. The average Mahalanobis distance is calculated using the multi-dimensional state feature vector and a preset Mahalanobis distance calculation model. The average Mahalanobis distance is then mapped to a comprehensive thermal runaway risk index. Multiple control state levels are preset and different threshold ranges are set for each control state level. The current control state level of the battery is determined based on the threshold range of the thermal runaway comprehensive risk index. PID control parameters are obtained through a preset RBF neural network; based on the current control state level of the battery, the expected control vector of each actuator is calculated using the PID control parameters, and the actual control vector of each actuator is calculated based on the expected control vector. The actual control vector is used as a temperature control signal to control the current actuator to perform temperature control processing on the battery.

2. The battery graded temperature control method based on Mahalanobis distance early warning according to claim 1, characterized in that, The calculation of the average Mahalanobis distance using the multidimensional state feature vector and the preset Mahalanobis distance calculation model specifically includes: Obtain the historical operating parameters of the battery in normal operating mode, and calculate and store the normal baseline mean vector based on the historical operating parameters. The normal baseline covariance matrix is ​​obtained based on the normal baseline mean vector. To construct a Mahalanobis distance calculation model: in, Represents Mahalanobis distance, Represents a multidimensional state feature vector. This represents the normal baseline covariance matrix after regularization. Indicates the discrete sampling sequence number; Based on the Mahalanobis distance calculation model, the Mahalanobis distance of the current battery is obtained, and then the Mahalanobis distance of the current battery is smoothed by a window to obtain the average Mahalanobis distance. : in, This indicates the preset length of the sliding window.

3. The battery graded temperature control method based on Mahalanobis distance early warning according to claim 1, characterized in that, The comprehensive risk index for thermal runaway is expressed as follows: in, This indicates the comprehensive risk index of thermal runaway. Indicates the discrete sampling sequence number. This represents the sensitivity adjustment coefficient. This represents the average Mahalanobis distance. This represents the upper limit of normal deviation determined based on normal samples.

4. The battery graded temperature control method based on Mahalanobis distance early warning according to claim 1, characterized in that, The specific steps of obtaining PID control parameters through a preset RBF neural network include: Within each control cycle, the operating parameters of the battery are normalized. The normalized operating parameters and the actual control vector of each actuator in the previous cycle are used as input data for the RBF neural network. The input data is mapped into feature vectors in a high-dimensional feature space through radial basis functions. Based on the inner product of the feature vectors and the preset RBF network weights, network prediction data is generated. The cross-Jacobi matrix of the actuator is calculated using the network prediction data. Based on the cross-Jacobi matrix and the preset PID adaptive learning rate, the PID control parameters of each actuator are tuned in real time using the gradient descent method.

5. A battery grading temperature control method based on Mahalanobis distance early warning according to claim 4, characterized in that, The method of obtaining PID control parameters through a preset RBF neural network further includes: The rate of change of Mahalanobis distance was calculated using the average Mahalanobis distance. Through the Mahalanobis distance change rate Dynamically adjust the learning rate of the RBF neural network weights and the adaptive learning rate of the PID controller. : in, , Indicates the weights of the RBF network. This represents the adaptive learning rate of the PID controller. This represents the adjustment coefficient. Indicates the discrete sampling sequence number.

6. The battery graded temperature control method based on Mahalanobis distance early warning according to claim 1, characterized in that, The calculation of the desired control vector for each actuator using the PID control parameters specifically includes: Based on the current control state level of the battery, the target temperature reached by the battery is obtained; the deviation between the current battery temperature and the target temperature reached is calculated. An incremental PID controller is input; the incremental PID controller calculates the incremental control quantity based on the PID control parameters. And through the incremental control quantity Calculate the desired control vector for each actuator. The incremental control quantity Represented as: in, Indicates actuator Incremental control quantity, proportion ,integral ,differential Indicates actuator PID control parameters.

7. The battery graded temperature control method based on Mahalanobis distance early warning according to claim 1, characterized in that, The calculation of the actual control vector of each actuator based on the desired control vector specifically includes: Based on the current control state level of the battery and the comprehensive thermal runaway risk index, the control weights corresponding to each actuator are calculated using a continuous mapping function. According to each control weight Get control allocation matrix ; Using the control allocation matrix and the expected control vector of each actuator, the actual control vector of each actuator is obtained, and the actual control vector is saturated and limited to obtain the final actual control vector. The control weights corresponding to each actuator Represented as: Where S represents the control state of the battery, and S0 represents a control state level. Indicates actuator The minimum allowed weight, Indicates the actuator in state S The maximum allowable weight, R represents the comprehensive risk index of thermal runaway. Indicates actuator The risk response inflection point under state S; This indicates the sensitivity to changes in weights.

8. A battery grading temperature control method based on Mahalanobis distance early warning according to claim 1, characterized in that, The calculation of the actual control vector of each actuator based on the desired control vector further includes: The actuator includes at least three types of actuators; the heat dissipation sensitivity of the first and second actuators to the hot end temperature of the third actuator and the hot end temperature of the third actuator are respectively acquired; a preset dynamic safety lower limit constraint is applied to the increment of the actual control vector of the first and second actuators using the heat dissipation sensitivity; the operation of each actuator is controlled using the actual control vector after the constraint is applied; so as to avoid the situation where the heat dissipation capacity of the first and second actuators is mismatched with the rate of increase of the hot end temperature of the third actuator, resulting in the hot end temperature of the third actuator being higher than the battery temperature.

9. A battery grading temperature control system based on Mahalanobis distance early warning, characterized in that, Include: Data processing module: used to acquire the operating parameters of the battery and each actuator, preprocess the operating parameters to construct a multi-dimensional state feature vector; use the multi-dimensional state feature vector and a preset Mahalanobis distance calculation model to calculate the average Mahalanobis distance; and map the average Mahalanobis distance to a comprehensive thermal runaway risk index. Control state determination module: used to pre-set multiple control state levels and set different threshold ranges for each control state level, and determine the current control state level of the battery based on the threshold range of the thermal runaway comprehensive risk index; Temperature control module: used to obtain PID control parameters through a preset RBF neural network; based on the current control state level of the battery, calculate the expected control vector of each actuator using the PID control parameters, and calculate the actual control vector of each actuator according to the expected control vector, and use the actual control vector as a temperature control signal to control the current actuator to perform temperature control processing on the battery.

10. A battery grading temperature control device based on Mahalanobis distance warning, comprising a battery box and batteries located inside the battery box, characterized in that, It also includes a battery grading temperature control system based on Mahalanobis distance warning as described in claim 9, and a heat dissipation actuator located above the battery box. The heat dissipation actuator includes a thermoelectric cooler, a water-cooled plate located above the thermoelectric cooler, and a cooling fan located above the water-cooled plate. The battery grading temperature control system is electrically connected to each heat dissipation actuator and is used to send control signals to the heat dissipation actuator to adjust the temperature of the battery.

Images