A new energy automobile electric drive axle bearing temperature control method and system

Abstract

The application discloses a new energy automobile electric drive axle bearing temperature control method and system, and belongs to the technical field of high-rotating-speed bearing temperature control of electric drive axle. The method comprises the following steps: collecting bearing temperature signals by using a temperature sensor and constructing a characteristic vector; inputting the characteristic vector into a self-adaptive deep confidence network-model predictive controller to output a control variable; selecting an active forced lubrication cooling mode or a passive splash lubrication cooling mode according to the control variable and the running state of the bearing by the self-adaptive deep confidence network-model predictive controller; periodically collecting the actual output of a cooling execution mechanism and the real-time temperature of the bearing, calculating a temperature control error, and performing real-time feedback and strategy optimization. The application realizes rapid response, low overshoot and high-precision control of bearing temperature, and takes into account energy saving and reliability, and effectively solves the bearing temperature control problem under high-rotating-speed variable working conditions.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of electric drive axle control, and particularly relates to a new energy vehicle electric drive axle bearing temperature control method and system. BACKGROUND

[0002] As a core transmission component of new energy vehicles, the high-speed bearing (speed usually >= 3000r / min) of the electric drive axle is prone to generate a large amount of heat due to friction and load during operation. If the temperature is too high (exceeding 80℃ will lead to lubrication failure, and exceeding 120℃ will cause bearing burning), it will seriously affect the reliability and service life of the electric drive axle.

[0003] At present, under the background of pursuing high power density and long endurance, the electric drive axle system is often operated under complex conditions of high speed and variable load, which puts higher requirements on the temperature control performance of the bearing. However, the existing temperature control technology has the following outstanding problems:

[0004] Firstly, in terms of control strategy, the existing bearing temperature control mostly adopts traditional PID control method, which adjusts the cooling system through preset fixed PID parameters. However, under high speed and variable conditions, the bearing temperature is greatly affected by interference factors such as load fluctuation, speed mutation, and environmental temperature change. The traditional PID control has inherent defects such as poor parameter adaptability, weak anti-interference ability, and insufficient dynamic performance due to its fixed parameter mechanism: specifically, the response lags when the speed suddenly rises, and the temperature overshoot is significant (>= 15%); when the environment fluctuates or the load suddenly changes, the temperature control accuracy decreases (steady-state error > 2℃); when starting or switching conditions, the temperature regulation time is long (>= 8s), and the temperature rise cannot be quickly suppressed.

[0005] Secondly, in terms of actuator level, the traditional cooling system often adopts a single cooling mode (such as splash lubrication), and lacks an intelligent mechanism for on-demand adjustment. This will cause the cooling system to run at high power continuously under low load or steady-state conditions, resulting in unnecessary energy waste and being very detrimental to the improvement of vehicle energy efficiency; at the same time, it cannot provide sufficient cooling strength under high temperature and high speed conditions, and there is a contradiction between insufficient cooling and excessive cooling, which affects the temperature control accuracy and system reliability.

[0006] Therefore, there is an urgent need for a comprehensive control method that can balance dynamic response speed and steady-state accuracy, has a dual-mode cooling mechanism, and can intelligently adjust the cooling strength according to the working conditions, to solve the shortcomings of existing technology in adaptability, energy efficiency, and reliability. SUMMARY

[0007] To overcome the shortcomings of existing technologies, this invention provides a method and system for controlling the temperature of electric drive axle bearings in new energy vehicles. This method aims to achieve rapid response, low overshoot, and high-precision control of bearing temperature by using an adaptive deep belief network-model prediction controller to output the control quantity, combined with a dual-mode cooling mechanism, real-time feedback, and strategy optimization, while simultaneously improving system energy efficiency and operational reliability.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] A method for temperature control of electric drive axle bearings in new energy vehicles includes the following steps:

[0010] Step S1: Temperature sensors are installed on the inner ring, outer ring, and rolling elements of the electric drive axle bearing of the new energy vehicle to collect temperature signals. The temperature signals are preprocessed and the average temperature and temperature change rate are calculated. A feature vector is constructed based on the average temperature, temperature change rate, bearing speed, and bearing load.

[0011] Step S2: Design an adaptive deep belief network-model prediction controller, input the feature vector into the adaptive deep belief network-model prediction controller, and output the control quantity;

[0012] Step S3: The adaptive deep belief network-model predictive controller selects the active forced lubrication cooling mode or the passive splash lubrication cooling mode according to the output control quantity and the bearing's operating state, and linearly maps the control quantity to the corresponding cooling actuator action;

[0013] Step S4: Adaptive deep belief network-model predictive controller periodically collects the actual output of the cooling actuator and the real-time temperature of the bearing, calculates the temperature control error, and performs real-time feedback and strategy optimization.

[0014] Furthermore, the adaptive deep belief network-model predictive controller in step S2 is divided into two stages: deep belief network and model predictive control. The deep belief network stage includes offline training of the deep belief network and online prediction of the deep belief network. The model predictive control stage includes building a state-space model and solving for the optimal control quantity through rolling optimization.

[0015] Furthermore, the offline training steps for the deep belief network are as follows:

[0016] Data preparation: Collect typical operating condition data, the parameter range of which is speed. Load ,temperature The typical working condition data is divided into a training set and a validation set.

[0017] Pre-training: The contrastive divergence algorithm is used to pre-train the three-layer Restricted Boltzmann Machine sequentially, initialize the weights and biases of the Restricted Boltzmann Machine, and use the hidden layer of the previous Restricted Boltzmann Machine as the visible layer of the next Restricted Boltzmann Machine.

[0018] Backpropagation fine-tuning: Using the mean square error between the predicted and actual temperatures as the loss function, the parameters of each deep belief network are fine-tuned through the backpropagation algorithm until the mean square error is reached. .

[0019] Furthermore, the online prediction process of the deep belief network is as follows: the feature vectors collected and processed in real time are input into the offline trained deep belief network model, and the bearing average temperature prediction sequence for the next 5 sampling periods is output.

[0020] Furthermore, the steps of the model predictive control stage are as follows:

[0021] Constructing a state-space model:

[0022]

[0023] Where x(k+1) is the eigenvector of the next period k+1. Let k be the eigenvector of the current period. This is the control quantity for the current period k. Let A be the bearing temperature change rate in the current period k, B be the state transition matrix, C be the control input matrix, and D be the direct transfer matrix.

[0024] Define the objective function and constraints:

[0025]

[0026] Where J is the objective function of the model predictive control stage. For prediction in the time domain; To control the time domain; Is the future number Predicted output for each cycle; It controls the increment; This is the preset temperature threshold for the bearing temperature;

[0027] The constraint conditions are set according to different bearing cooling modes. The control quantity of k in the previous cycle under active forced lubrication cooling mode is... satisfy Control quantity of k in the first cycle under passive splash lubrication and cooling mode satisfy ;

[0028] Solving for the optimal control input: The interior-point method is used to solve the constrained quadratic programming problem, obtaining the optimal control sequence. The first element of the optimal control sequence is then extracted as the control input for the current period k. .

[0029] Furthermore, the triggering condition for the bearing cooling mode is:

[0030] Active forced lubrication and cooling mode triggering: within two consecutive cycles. or , The average temperature is n, and the bearing speed is n; the maximum oil pump speed is n. The minimum sustaining speed is In active forced lubrication and cooling mode, the control quantity of k in the current cycle. Linear mapping to oil pump speed:

[0031] ;

[0032] Passive splash lubrication cooling mode trigger: within two consecutive cycles and ; The average temperature is given by , and n is the bearing speed; under passive splash lubrication cooling mode, the control quantity k in the current cycle is... Linear mapping to splash disk rotation speed:

[0033] ;

[0034] The initial state and default maintenance state of the temperature control system for electric drive axle bearings in new energy vehicles are set to "passive splash lubrication cooling mode". It will switch to active forced lubrication cooling mode only when the triggering conditions of active forced lubrication cooling mode are met; once the conditions are no longer met, it will return to the default passive splash lubrication cooling mode.

[0035] Furthermore, when the adaptive deep belief network-model predictive controller detects oil pump jamming, it automatically switches to passive splash lubrication and cooling mode and triggers a fault alarm;

[0036] When the adaptive deep belief network-model predictive controller detects splash disk jamming, it automatically switches to active forced lubrication and cooling mode and triggers a fault alarm.

[0037] When both cooling actuators fail, an audible and visual alarm is issued through the instrument panel of the new energy vehicle, indicating "Critical failure of the cooling system, please stop the vehicle immediately and safely." The motor is then requested to reduce its power via the CAN bus, and fault data is recorded.

[0038] Furthermore, the specific process of real-time feedback and strategy optimization is as follows:

[0039] When the temperature control error Or three consecutive cycles of temperature fluctuation At the same time, the average temperature of the bearing and control quantity When the temperature rises, a protective full-speed cooling mode is activated. This mode maximizes cooling intensity by operating the cooling actuators at their rated maximum speed, prioritizing the prevention of temperature increases. Under this condition, the deep belief network parameter updates and state-space model iteration processes are temporarily suspended. Temperature control error Furthermore, once the fluctuation meets the steady-state criteria, the protective full-speed cooling mode is exited, and the adaptive deep belief network-model predictive controller re-solves for the control quantity.

[0040] When the temperature control error Furthermore, the fluctuations are stable, and the active forced lubrication and cooling mode reduces the oil pump speed to a minimum. The passive splash lubrication cooling mode reduces the splash disc speed to [a certain value]. .

[0041] This invention also provides a temperature control system for electric drive axle bearings in new energy vehicles, used to execute the aforementioned temperature control method for electric drive axle bearings in new energy vehicles, comprising:

[0042] The temperature acquisition module is used to acquire temperature signals at the inner ring, outer ring, and rolling elements of the bearing, and to calculate the average temperature and the rate of temperature change, while constructing a feature vector.

[0043] The adaptive deep belief network-model predictive controller module is used to receive the feature vector of the temperature acquisition module, perform temperature prediction through the deep belief network, perform rolling optimization and output control quantity in combination with model predictive control, and perform strategy optimization after real-time feedback.

[0044] The dual-mode cooling execution module is used to receive the control input from the adaptive deep belief network-model prediction controller module, switch between active forced lubrication or passive splash lubrication cooling mode according to the bearing operating state, and realize the linear mapping from the control input to the speed of the actuator. It also has fault switching and alarm functions.

[0045] The feedback optimization module is used to periodically collect the actual output of the cooling actuator and the real-time temperature of the bearing, calculate the temperature control error, and trigger the corresponding control quantity update or energy-saving speed reduction strategy.

[0046] The beneficial effects of this invention are:

[0047] This invention integrates adaptive deep belief networks and model predictive control, achieving rapid response, low overshoot, and high-precision control of bearing temperature through dual-mode cooling. The dual-mode cooling automatically selects the cooling method based on bearing temperature and rotational speed, and automatically reduces speed when the temperature stabilizes, significantly reducing energy consumption. Comprehensive fault self-diagnosis and a protective full-speed cooling mode ensure the reliability of the temperature control system for electric drive axle bearings in new energy vehicles, while periodic feedback optimization allows the adaptive deep belief network-model predictive controller to continuously adapt to changing operating conditions. This effectively improves the accuracy, energy efficiency, and reliability of temperature control for electric drive axle bearings in new energy vehicles. Attached Figure Description

[0048] Figure 1 This is a flowchart of a method for controlling the temperature of bearings in the electric drive axle of new energy vehicles. Detailed Implementation

[0049] To facilitate understanding and implementation of the present invention by those skilled in the art, the present invention will be further described in detail below with reference to the accompanying drawings.

[0050] like Figure 1 As shown, a method for temperature control of electric drive axle bearings in new energy vehicles includes the following steps:

[0051] Step S1: Temperature sensors are installed on the inner ring, outer ring, and rolling elements of the electric drive axle bearing of the new energy vehicle to collect temperature signals. The temperature signals are preprocessed and the average temperature and temperature change rate are calculated. A feature vector is constructed based on the average temperature, temperature change rate, bearing speed, and bearing load.

[0052] Step S2: Design an adaptive deep belief network-model prediction controller (adaptive DBN-MPC controller), input the feature vector into the adaptive deep belief network-model prediction controller, and output the control quantity;

[0053] Step S3: The adaptive deep belief network-model predictive controller selects the active forced lubrication cooling mode or the passive splash lubrication cooling mode according to the output control quantity and the bearing's operating state, and linearly maps the control quantity to the corresponding cooling actuator action;

[0054] Step S4: Adaptive deep belief network-model predictive controller periodically collects the actual output of the cooling actuator and the real-time temperature of the bearing, and performs real-time feedback and strategy optimization.

[0055] In this embodiment, PT100 platinum resistance temperature sensors are installed on the inner ring, outer ring, and rolling elements of the high-speed bearing of the electric drive axle of the new energy vehicle to collect the real-time temperature of the bearing's inner ring. outer ring temperature of bearing Temperature of rolling elements The Kalman filter algorithm is used to eliminate noise from the temperature sensor, and the average temperature is calculated. The calculation formula is:

[0056] ;

[0057] Calculate the rate of temperature change :

[0058] ;

[0059] In the formula:

[0060] The average temperature for the current sampling period;

[0061] For the previous sampling period ( The average temperature of the period;

[0062] This represents the sampling time interval.

[0063] Constructing feature vectors :

[0064] ;

[0065] in, For bearing speed, The bearing load is obtained indirectly by the electric drive axle torque sensor.

[0066] The design of the Adaptive Deep Belief Network-Model Prediction Controller (Adaptive DBN-MPC Controller) is divided into deep belief network ( The system consists of two phases: the Deep Belief Network (DBN) phase and the Model Predictive Control (MPC) phase. The Deep Belief Network phase includes offline training and online prediction of the Deep Belief Network, achieving high-precision temperature prediction through DBN. The MPC phase involves building a state-space model and solving for the optimal control variable through rolling optimization. The DBN phase is responsible for extracting operating condition features and building a temperature prediction model, providing five future temperature sequences for the MPC phase, and is divided into offline training and online prediction phases.

[0067] The offline training phase of DBN includes the following steps:

[0068] Data equipment: Collects 100,000 sets of typical operating condition data (speed) Load ,temperature ), divided into training set ( ) and validation set ( This is used to verify the generalization ability of DBN.

[0069] Pre-training: Contrastive divergence ( The algorithm initializes a 3-layer RBM (Restricted Boltzmann Machine); by comparing the divergence ( The algorithm pre-trains the 3-layer RBM sequentially (using the training set), initializing the network weights and biases; using the mean square error (MSE) between the predicted and actual bearing temperatures as the loss function, it fine-tunes the network parameters of each DBN through backpropagation (BP) until... Among them, the contrast divergence ( Algorithm: Used for layer-by-layer pre-training of a 3-layer RBM, solving the training convergence problem caused by random initial parameters in DBN, and measuring the deviation between predicted and actual temperatures. Different feature vectors constructed from 100,000 sets of working condition data are used as inputs to the visible layer of the RBM; based on the currently randomly initialized RBM weights or biases, the activation probability of hidden layer neurons is calculated and the hidden layer state is sampled; the hidden layer state is used to inversely calculate the reconstructed data of the visible layer, and the difference between the original input and the reconstructed data is compared; based on the difference, the weights and biases of this layer of the RBM are adjusted, and the 3-layer RBM is trained sequentially from bottom to top (the previous hidden layer becomes the next visible layer), completing the initialization.

[0070] Backpropagation (BP) algorithm: After pre-training, the network parameters of DBN need to be further fine-tuned to ensure that the temperature prediction accuracy meets the requirements. The core is to optimize based on the mean square error (MSE) as the loss function.

[0071] ;

[0072] Where N is the number of samples, Let be the predicted average temperature of the electric drive axle bearing under the i-th typical operating condition sample. The measured average temperature of the electric drive axle bearing under the i-th typical operating condition sample is denoted as .

[0073] Require That is, the average deviation between the predicted temperature and the actual temperature. The fine-tuning steps include: Forward propagation: Inputting the training set data into the pre-trained DBN, and calculating the output layer by layer to obtain the predicted temperature;

[0074] Error backpropagation: The global error MSE (based on the sum of squares of the deviations between the predicted bearing average temperature and the measured bearing average temperature of all samples in the 100,000 typical working condition training set, i.e. mean square error MSE) is backpropagated from the top layer of DBN to the bottom layer, and the weights and biases of all layers are corrected according to the gradient descent rule.

[0075] Iteration Termination: Repeat the forward propagation and error backpropagation process until the MSE meets the requirements, ensuring that the DBN has high-precision temperature prediction capabilities.

[0076] Online prediction during the DBN phase:

[0077] After offline training is completed, DBN, as a component of the DBN-MPC controller, predicts the bearing temperature trend in real time, providing input for subsequent MPC rolling optimization.

[0078] The feature vector of the current sampling period, collected and processed in real time (k is the current sampling period) Input the trained DBN; obtain the temperature sequence:

[0079] ;

[0080] in, The average temperature of the first future period k is based on the current period. The average temperature of the second future period k is based on the current period. The temperature series provides the MPC with the average temperature of the 5th future cycle k based on the current cycle. It is a key basis for the MPC to achieve "rolling optimization" (advance prediction and precise control).

[0081] The specific process of constructing the state-space model in the MPC phase is as follows:

[0082] MPC, based on DBN, predicts temperature sequences and outputs adjustment signals for the temperature control system of electric drive axle bearings in new energy vehicles through the logic of "building a state-space model → setting objective functions and constraints → solving for the optimal control quantity", ensuring accurate temperature control and smooth actuator operation.

[0083] The adaptive DBN-MPC controller utilizes the future temperature prediction sequence provided by the deep belief network (DBN) to construct a state-space model and solves for the optimal control quantity through rolling optimization.

[0084] The nonlinear model identified by DBN is locally linearized to obtain the state-space model:

[0085] ;

[0086] in, For the next cycle eigenvectors (next cycle state vectors);

[0087] This is the feature vector (state vector) of the current period k, which contains the core operating condition features that affect the bearing temperature.

[0088] The control quantity (control input) for the current cycle k represents the adjustment signal of the cooling system (which will be linearly mapped to the speed setpoint of the oil pump or splash plate later).

[0089] The bearing temperature change rate for the current period k directly reflects the temperature control accuracy.

[0090] ;

[0091] The average temperature of the current period k. Preset a temperature threshold for the bearing temperature;

[0092] A is the state transition matrix, describing the state variables of the current period. State variables for the next period The impact;

[0093] B is the control input matrix, describing the control quantity for the current period k. For the next cycle state The regulatory effect;

[0094] C is the output matrix, describing the current state. bearing temperature change rate Mapping relationship;

[0095] D is the direct transfer matrix, which describes the direct impact of the control output.

[0096] MPC parameter initialization: In the offline phase, based on DBN training data (100,000 sets of typical working condition data), the initial values ​​of the state transition matrix A, control input matrix B, output matrix C, and direct transfer matrix D are initially determined through system identification methods.

[0097] MPC Online Update: During the feedback optimization phase, when temperature control error is detected... Or the output error of the cooling actuator At that time, the state space model's state transition matrix A, control input matrix B, output matrix C, and direct transfer matrix D will be updated to ensure that the model matches the actual operating conditions.

[0098] The specific process of solving the control variable in the rolling optimization stage of MPC is as follows:

[0099] Constructing and solving the objective function: Substituting the predicted temperature sequence from the deep belief network (DBN), the preset bearing temperature setpoint, relevant parameters of current and future operating conditions (average bearing temperature, bearing speed, bearing load), and parameters of the state-space model into the objective function and constraints, the quadratic programming problem is solved using the interior-point method to obtain the control quantity k for the current period. .

[0100] Assume the MPC prediction time domain is 5 prediction steps (corresponding to the temperature sequence of the next 5 periods output in the online prediction stage), the control time domain is consistent with the prediction time domain (N=5), and the objective function is:

[0101] ;

[0102] in: The objective function for the Model Predictive Control (MPC) phase;

[0103] These are weighting coefficients that balance the output tracking error and the control rate of change.

[0104] ; For prediction in the time domain (corresponding to the 5-step temperature sequence output by DBN);

[0105] ; To ensure the control time domain is consistent with the prediction time domain and that the optimization covers the entire prediction period, this optimization requires calculating the control quantity sequence for the next 5 periods. To ensure that the control strategy is fully adapted to future operating conditions, including This is the control quantity for the first cycle in the future. (This is the control quantity for the fourth future cycle).

[0106] Is the future number The predicted output (average temperature) for each step is based on the current period's k-state vector. and control sequence calculation;

[0107] The first term of the objective function: This represents minimizing the deviation between the predicted temperature and the preset temperature threshold of the bearing, ensuring temperature control accuracy (target). ;

[0108] Second item: ,in It is a control increment used to smooth control actions (avoiding abrupt changes in the actions of the cooling actuator). It is a weighting coefficient that balances the output tracking error and the control rate of change. The larger the size, the more emphasis is placed on the smoothness of the actuator; The smaller the value, the more emphasis is placed on temperature control accuracy, and dynamic adjustments are needed based on actual operating conditions (such as prioritizing accuracy during high-speed heavy-load conditions and prioritizing energy saving during low-speed steady-state conditions).

[0109] In this embodiment, the core of the constraint is "ensuring that the control quantity is within the capability range of the cooling actuator and adapts to the dual-mode cooling logic," which is divided into two categories:

[0110] Active forced lubrication and cooling mode: or n represents the bearing speed;

[0111] The control quantity must be mapped to the oil pump speed, and the following mathematical expression must be satisfied:

[0112] ;

[0113] ;

[0114] Corresponding oil pump speed This prevents the oil pump from running dry at low speeds or operating beyond its rated speed, which could cause damage.

[0115] Passive splash lubrication cooling mode: and n represents the bearing speed;

[0116] The control quantity mapped to the splash plate rotation speed must satisfy the following mathematical expression:

[0117] ;

[0118] ;

[0119] Corresponding splash disk speed It balances cooling performance with mechanical wear, and cooling performance with energy-saving requirements under low-load conditions.

[0120] The objective function and constraints together constitute a constrained quadratic programming problem, which needs to be solved using the interior-point method to obtain the optimal control sequence.

[0121] ;

[0122] Extracting control input: optimal solution The first element is the control quantity of the current period k. .

[0123] The adaptive DBN-MPC controller selects either the active forced lubrication cooling mode or the passive splash lubrication cooling mode based on the output control quantity, and linearly maps the control quantity to the corresponding cooling actuator action. The specific triggering conditions are as follows:

[0124] Active forced lubrication and cooling trigger: or (meeting any of the conditions), and for two consecutive sampling periods ( Maintain this state. Represents the bearing speed. Maximum oil pump speed. Minimum holding speed Active mode: Control quantity (Typically 50-100) Linear mapping to oil pump speed:

[0125] ;

[0126] Passive splash lubrication cooling trigger: and (Both conditions are met), and this state is maintained for two consecutive sampling periods (0.4s). n represents the bearing speed. In passive mode: control quantity... (Typically 0-100) Linearly mapped to oil pump speed:

[0127] ;

[0128] If the adaptive deep belief network-model predictive controller detects oil pump sticking (error between actual speed and commanded speed) (And for two consecutive cycles), it automatically switches to passive splash lubrication and cooling mode and triggers a fault alarm;

[0129] If the adaptive deep belief network-model predictive controller detects splash disk jamming (actual rotation speed error) It automatically switches to active forced lubrication and cooling mode and triggers a fault alarm;

[0130] If both actuators fail, an audible and visual alarm will be issued via the instrument panel, indicating "Critical cooling system malfunction, please stop the vehicle immediately and safely." The motor will be requested to reduce power / limp mode via the CAN bus, and fault data (temperature, speed, fault time, etc.) will be recorded.

[0131] In this embodiment, during active forced lubrication and cooling, the adaptive DBN-MPC controller collects the actual speed of the oil pump (within acceptable error limits). ) and average temperature ;

[0132] During passive splash lubrication and cooling, the adaptive DBN-MPC controller acquires the actual rotational speed of the splash disc (within acceptable error range). ), average temperature ;

[0133] Adaptive DBN-MPC controller with Calculate the temperature control error periodically:

[0134] ;

[0135] when Or three consecutive cycles of temperature fluctuation And if the average temperature of the bearing and control quantity When the temperature rises, a protective full-speed cooling mode will be activated. This mode maximizes cooling intensity by running the cooling actuators (oil pump or splash plate) at their rated maximum speed, prioritizing the prevention of temperature increases and ensuring bearing temperature control safety. Under this condition, DBN parameter updates and state-space model iterations are temporarily suspended to avoid impacting emergency cooling effects on computational resources and transient fluctuations in control quantities caused by parameter optimization. The cooling process will continue until the temperature control error converges to... Furthermore, once the fluctuations stabilize, the system exits the protective full-speed cooling mode. The adaptive deep belief network-model predictive controller then resolves the control input and linearly maps the control input to the corresponding cooling actuator action, achieving a rapid cooling effect.

[0136] when And the fluctuations are stable (active forced lubrication and cooling mode fluctuates for 5 consecutive cycles). Passive splash lubrication cooling mode fluctuates continuously for 5 cycles. When the active forced lubrication and cooling mode is activated, the oil pump speed is reduced to [a certain value]. The passive splash lubrication cooling mode reduces the splash disc speed to [a certain value]. Reduce ineffective energy consumption.

[0137] Another embodiment of the present invention provides a temperature control system for electric drive axle bearings of new energy vehicles, used to execute the aforementioned temperature control method for electric drive axle bearings of new energy vehicles, comprising:

[0138] The temperature acquisition module is used to acquire temperature signals at the inner ring, outer ring, and rolling elements of the bearing, and to calculate the average temperature and the rate of temperature change, while constructing a feature vector.

[0139] The adaptive deep belief network-model predictive controller module is used to receive the feature vector of the temperature acquisition module, perform temperature prediction through the deep belief network, perform rolling optimization and output control quantity in combination with model predictive control, and perform strategy optimization after real-time feedback.

[0140] The dual-mode cooling execution module is used to receive the control input from the adaptive deep belief network-model prediction controller module, switch between active forced lubrication or passive splash lubrication cooling mode according to the bearing operating state, and realize the linear mapping from the control input to the speed of the actuator. It also has fault switching and alarm functions.

[0141] The feedback optimization module is used to periodically collect the actual output of the cooling actuator and the real-time temperature of the bearing, calculate the temperature control error, and trigger the corresponding control quantity update or energy-saving speed reduction strategy.

[0142] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for temperature control of electric drive axle bearings in new energy vehicles, characterized in that, Includes the following steps: Step S1: Temperature sensors are installed on the inner ring, outer ring, and rolling elements of the electric drive axle bearing of the new energy vehicle to collect temperature signals. The temperature signals are preprocessed and the average temperature and temperature change rate are calculated. A feature vector is constructed based on the average temperature, temperature change rate, bearing speed, and bearing load. Step S2: Design an adaptive deep belief network-model prediction controller. Input the feature vector into the adaptive deep belief network-model prediction controller and output a control quantity. The adaptive deep belief network-model prediction controller consists of two stages: a deep belief network and model prediction control. The deep belief network stage includes offline training and online prediction of the deep belief network. The model prediction control stage includes: constructing a state-space model and setting the objective function and constraints. ； Where J is the objective function of the model predictive control stage. For prediction in the time domain; To control the time domain; Is the future number Predicted output for each cycle; It controls the increment; This is the preset temperature threshold for the bearing; λ is the weighting coefficient. The optimal control quantity is obtained through rolling optimization: the interior-point method is used to solve the constrained quadratic programming problem, the optimal control sequence is obtained, and the first element of the optimal control sequence is extracted as the control quantity for the current period k. ; Step S3: The adaptive deep belief network-model predictive controller selects the active forced lubrication cooling mode or the passive splash lubrication cooling mode according to the output control quantity and the bearing's operating state, and linearly maps the control quantity to the corresponding cooling actuator action; Step S4: Adaptive deep belief network-model predictive controller periodically collects the actual output of the cooling actuator and the real-time temperature of the bearing, calculates the temperature control error, and performs real-time feedback and strategy optimization.

2. The method for controlling the temperature of electric drive axle bearings in new energy vehicles according to claim 1, characterized in that, The offline training steps for the deep belief network are as follows: Data preparation: Collect typical operating condition data, the parameter range of which is speed. Load ,temperature Typical working condition data are divided into training set and validation set; Pre-training: The contrastive divergence algorithm is used to pre-train the three-layer Restricted Boltzmann Machine sequentially, initialize the weights and biases of the Restricted Boltzmann Machine, and use the hidden layer of the previous Restricted Boltzmann Machine as the visible layer of the next Restricted Boltzmann Machine. Backpropagation fine-tuning: Using the mean square error between the predicted and actual temperatures as the loss function, the parameters of each deep belief network are fine-tuned through the backpropagation algorithm until the mean square error is reached. .

3. The method for controlling the temperature of electric drive axle bearings in new energy vehicles according to claim 1, characterized in that, The online prediction process of the deep belief network is as follows: the feature vectors collected and processed in real time are input into the offline trained deep belief network model, and the bearing average temperature prediction sequence for the next 5 sampling periods is output.

4. The method for controlling the temperature of electric drive axle bearings in new energy vehicles according to claim 1, characterized in that, The state-space model is as follows: ； in, For the next cycle eigenvectors, Let k be the eigenvector of the current period. This is the control quantity for the current period k. Let A be the bearing temperature change rate in the current period k, B be the state transition matrix, C be the control input matrix, and D be the direct transfer matrix. The constraint conditions are set according to different bearing cooling modes. The control quantity of k in the previous cycle under active forced lubrication cooling mode is... satisfy Control quantity of k in the first cycle under passive splash lubrication and cooling mode satisfy .

5. The method for temperature control of electric drive axle bearings in new energy vehicles according to claim 1, characterized in that, The triggering condition for bearing cooling mode is: Active forced lubrication and cooling mode triggering: within two consecutive cycles. or , The average temperature is n, and the bearing speed is n; the maximum oil pump speed is n. The minimum sustaining speed is ; In active forced lubrication and cooling mode, the control quantity of k in the current cycle. Linear mapping to oil pump speed: ； Passive splash lubrication cooling mode trigger: within two consecutive cycles and ; The average temperature is given by , and n is the bearing speed; under passive splash lubrication cooling mode, the control quantity k in the current cycle is... Linear mapping to splash disk rotation speed: ； The initial state and default maintenance state of the temperature control system for electric drive axle bearings in new energy vehicles are set to passive splash lubrication cooling mode. It will switch to active forced lubrication cooling mode only when the triggering conditions for active forced lubrication cooling mode are met; once the conditions are no longer met, it will return to the default passive splash lubrication cooling mode.

6. The method for temperature control of electric drive axle bearings in new energy vehicles according to claim 1, characterized in that, When the adaptive deep belief network-model predictive controller detects oil pump jamming, it automatically switches to passive splash lubrication and cooling mode and triggers a fault alarm. When the adaptive deep belief network-model predictive controller detects splash disk jamming, it automatically switches to active forced lubrication and cooling mode and triggers a fault alarm. When both cooling actuators fail, an audible and visual alarm will be issued through the instrument panel of the new energy vehicle, indicating a serious malfunction in the cooling system. The vehicle should be stopped immediately and safely. The alarm will then request the motor to reduce its power via the CAN bus and record the fault data.

7. The method for temperature control of electric drive axle bearings in new energy vehicles according to claim 1, characterized in that, The specific process of real-time feedback and strategy optimization is as follows: When the temperature control error Or three consecutive cycles of temperature fluctuation At the same time, the average temperature of the bearing and control quantity When the temperature rises, a protective full-speed cooling mode is activated. This mode maximizes cooling intensity by operating the cooling actuators at their rated maximum speed, prioritizing the prevention of temperature increases. Under this condition, the deep belief network parameter updates and state-space model iteration processes are temporarily suspended. Temperature control error Furthermore, once the fluctuation meets the steady-state criteria, the protective full-speed cooling mode is exited, and the adaptive deep belief network-model predictive controller re-solves for the control quantity. When the temperature control error Furthermore, the fluctuations are stable, and the active forced lubrication and cooling mode reduces the oil pump speed to a minimum. The passive splash lubrication cooling mode reduces the splash disc speed to [a certain value]. .

8. A temperature control system for electric drive axle bearings of new energy vehicles, used to execute the temperature control method for electric drive axle bearings of new energy vehicles according to any one of claims 1-7, characterized in that, include: The temperature acquisition module is used to acquire temperature signals at the inner ring, outer ring, and rolling elements of the bearing, and to calculate the average temperature and the rate of temperature change, while constructing a feature vector. The adaptive deep belief network-model predictive controller module is used to receive the feature vector of the temperature acquisition module, perform temperature prediction through the deep belief network, perform rolling optimization and output control quantity in combination with model predictive control, and perform strategy optimization after real-time feedback. The dual-mode cooling execution module is used to receive the control input from the adaptive deep belief network-model prediction controller module, switch between active forced lubrication or passive splash lubrication cooling mode according to the bearing operating state, and realize the linear mapping from the control input to the speed of the actuator. It also has fault switching and alarm functions. The feedback optimization module is used to periodically collect the actual output of the cooling actuator and the real-time temperature of the bearing, calculate the temperature control error, and trigger the corresponding control quantity update or energy-saving speed reduction strategy.

Images