Steady state self-learning method for automatic transmission

Abstract

The invention discloses an automatic transmission steady state self-learning method, and belongs to the technical field of vehicle automatic transmission control. The method comprises the following steps: judging self-learning enabling conditions; recognizing and capturing steady-state working conditions; applying controlled test excitation and acquiring data; calculating and updating parameters; verifying and storing learning results; and learning state management and quit. According to the method, the steady-state driving condition of the vehicle is intelligently captured, the controlled test excitation is applied on the premise that the preset enabling condition is met, the steady-state control parameters are calculated and updated on line through the built-in model based on the excitation response data, and the safety of the learning process is guaranteed through the multi-level enabling judgment and real-time interruption mechanism; the problems that in the prior art, automatic transmission steady-state core parameters lack a systematic reliable online self-learning scheme, the cost is high, the efficiency is low and the gear shifting quality is prone to decline due to parameter drifting due to the fact that the automatic transmission steady-state core parameters depend on off-line calibration are solved, self-adaptive compensation for performance degradation in the whole life cycle of a vehicle is achieved, and the precision and reliability of parameter learning are improved.

Description

Technical Field

[0001] This invention relates to a steady-state self-learning method for automatic transmissions, belonging to the field of vehicle automatic transmission control technology. Background Technology

[0002] The shift quality of an automatic transmission (including key indicators such as shift shock, slip control effect, and clutch engagement speed) is highly dependent on a set of control parameters that have been pre-determined through bench tests and road tests, such as clutch control pressure, torque phase parameters, inertial phase parameters, and system pressure reference.

[0003] However, in actual vehicle use, various factors can cause the preset optimal control parameters to drift, resulting in a decline in shift quality:

[0004] Firstly, component wear and aging, such as wear of clutch friction plates, aging of seals, and changes in hydraulic circuit characteristics;

[0005] Secondly, there are differences in product consistency; even automatic transmissions of the same model may have individual performance differences due to manufacturing tolerances.

[0006] Third, changes in the operating environment and conditions, such as fluctuations in transmission oil temperature, differences in driving altitude, and different driving habits (load control methods).

[0007] Traditional solutions primarily rely on fixed initial calibration data or perform coarse compensation through simple fault diagnosis mechanisms, failing to achieve refined online adaptive adjustments. Although some existing technologies involve self-learning strategies, they are mostly limited to transient learning during gear shifting (such as inertial phase learning). For the core parameters of steady-state operating conditions that determine the basic characteristics of gear shifting (including clutch kiss point, steady-state torque transmission characteristics, system pressure reference, etc.), there is a lack of a systematic, reliable, and safe online self-learning method.

[0008] Currently, to address the aforementioned steady-state parameter drift issue, vehicles typically need to be returned to the factory or recalibrated by professional technicians using specialized diagnostic equipment. This not only incurs high operating costs but is also inefficient, severely impacting the user experience.

[0009] Therefore, there is an urgent need to develop a method that can automatically, accurately, and safely complete the self-learning of steady-state core parameters during normal vehicle operation. Summary of the Invention

[0010] To address the problems existing in the background art, the present invention provides a steady-state self-learning method for automatic transmissions.

[0011] To achieve the above objectives, the present invention adopts the following technical solution: a steady-state self-learning method for an automatic transmission, the method comprising the following steps:

[0012] S1: Self-learning enable condition judgment: The system continuously monitors a set of global enable conditions, and only enters the steady-state self-learning preparation state when all conditions are met at the same time.

[0013] The global enable conditions include:

[0014] Condition 1: The transmission fluid temperature is within the optimal learning range;

[0015] Condition 2: Neither EMS nor TCU reported any fault codes;

[0016] Condition 3: The vehicle's mileage or operating time reaches the preset self-learning trigger cycle;

[0017] Condition 4: The vehicle battery voltage is within the normal operating range;

[0018] Condition 5: The vehicle is in a non-aggressive driving mode.

[0019] S2: Steady-state condition identification and capture: When all global enable conditions in S1 are met, the system begins to continuously identify steady-state conditions that meet the preset conditions, marks them as learning windows, and starts the learning preparation process.

[0020] The steady-state operating condition must simultaneously meet the following conditions:

[0021] The vehicle is in a stable cruising speed state, a slow acceleration state or slow deceleration state where the absolute value of acceleration is lower than a preset threshold, a stable engine torque output state where the torque fluctuation rate within the past preset time period is lower than a set threshold, and the transmission is not in the process of shifting gears and the current gear is fixed, and the duration of all the above states exceeds the preset time period.

[0022] S3: Applying controlled test stimuli and collecting data: During the learning window, the TCU actively applies test stimuli for different learning objectives and simultaneously collects key response data before and after the application of the stimuli;

[0023] The specific incentive methods for different learning objectives are as follows:

[0024] If the learning target is the clutch kiss point, then while keeping the torque transmitted in the current gear unchanged, the control pressure of the target clutch is increased in a stepwise manner to obtain the pressure increment, while keeping the working state of other clutches unchanged.

[0025] If the learning objective is the system pressure reference, then adjust the duty cycle of the main oil pressure solenoid valve and observe the following characteristics of the actual system pressure.

[0026] The data collected includes: changes in the input and output shaft speeds, changes in turbine speed, actual pressure of the target clutch, actual transmission ratio, and engine output torque.

[0027] S4: Parameter Calculation and Update: Substitute the response data collected in S3 into the physical model or statistical model preset in the TCU for calculation to obtain the corresponding new steady-state control parameter values. Compare the calculated new parameter values ​​with the original calibration values ​​stored in the TCU, and generate smooth updated parameter values ​​through filtering.

[0028] The calculations of the physical or statistical model include:

[0029] For clutch kiss point learning, by analyzing the applied pressure increment and observing whether slippage occurs at the input and output shaft speeds, the instantaneous pressure point at which the clutch begins to transmit torque is calculated, i.e., the real-time kiss point.

[0030] For system pressure reference learning, the system pressure reference correction value is calculated by analyzing the deviation between the actual system pressure and the target pressure and the response speed after the duty cycle of the main oil pressure solenoid valve.

[0031] S5: Learning Result Verification and Storage: After a single parameter update, the system enters a short-term verification phase. In subsequent gear shifting operations involving the updated parameters, the TCU continuously monitors the gear shifting quality indicators. If the gear shifting quality is within the expected range, the learning is confirmed to be effective, and the verified effective parameter values ​​are written into the TCU.

[0032] S6: Learning State Management and Exit: The entire self-learning process adopts a state machine management mechanism. In any stage of S2-S5, if the system detects that the safety conditions or steady-state conditions are not met, the current self-learning process will be interrupted immediately, all temporarily collected data will be cleared, and the normal driving mode will be restored. The system will continue to monitor the operating conditions and wait for the next learning window to restart the learning process.

[0033] Compared with the prior art, the beneficial effects of the present invention are:

[0034] This invention intelligently captures the steady-state driving conditions of a vehicle and applies controlled test stimuli under preset enabling conditions. Based on the stimuli response data, it calculates and updates steady-state control parameters online through a built-in model. Furthermore, it ensures the safety of the learning process through multi-level enabling judgment and real-time interruption mechanisms. This effectively solves the problems in existing technologies, such as the lack of a reliable online self-learning scheme for the core steady-state parameters of automatic transmissions, the reliance on offline calibration leading to high costs and low efficiency, and the easy degradation of shift quality due to parameter drift. It achieves adaptive compensation for performance degradation throughout the vehicle's entire life cycle, improves the accuracy and reliability of parameter learning, and the learning process does not interfere with the driver's normal driving. At the same time, it significantly reduces customer complaints and the need for return calibration, lowers the total life cycle maintenance cost, and enhances product reputation and brand value. Detailed Implementation

[0035] The technical solutions of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0036] An automatic transmission steady-state self-learning method intelligently captures steady-state driving conditions that meet preset criteria by real-time monitoring of vehicle operating status. Under these conditions, it actively applies precise and controllable test stimuli and, based on the stimuli response data, uses a built-in algorithm model to calculate and update steady-state control parameters online. This allows for online adaptive calibration or correction of shift control parameters to compensate for performance degradation and differences throughout the vehicle's lifespan. The method includes the following steps:

[0037] S1: Self-learning enable condition judgment: The system (the steady-state self-learning module built into the TCU) continuously monitors a set of global enable conditions, and only when all conditions are met at the same time is the system allowed to enter the steady-state self-learning preparation state.

[0038] The global enable conditions include:

[0039] Condition 1: The transmission oil temperature is within the optimal learning range, specifically 60℃-110℃;

[0040] Condition 2: Neither the EMS (Engine Control System) nor the TCU (Transmission Control Unit) reported any fault codes related to torque control, speed monitoring, or clutch control;

[0041] Condition 3: The vehicle's mileage or operating time reaches the preset self-learning trigger cycle;

[0042] Condition 4: The vehicle battery voltage is within the normal operating range to ensure stable solenoid valve operation;

[0043] Condition 5: The vehicle is in a non-aggressive driving mode, such as Normal Mode or ECO Mode, excluding Sport Mode, Manual Mode, and Traction Mode.

[0044] S2: Steady-state condition identification and capture: When all global enable conditions in S1 are met, the system begins to continuously identify steady-state conditions that meet the preset conditions. Once a condition that meets all conditions is detected, it is marked as a learning window and the learning preparation process is started.

[0045] The steady-state operating condition must simultaneously meet the following conditions:

[0046] The vehicle is in a stable speed cruising state (such as high-speed cruising), a slow acceleration state or slow deceleration state where the absolute value of acceleration is lower than a preset threshold, a stable engine torque output state where the torque fluctuation rate within a preset time period is lower than a set threshold, and the transmission is not in the process of shifting gears and the current gear is fixed, and the duration of all the above states exceeds a preset time period (such as 10 seconds).

[0047] S3: Applying controlled test stimuli and data acquisition: During the learning window, the TCU actively applies micro-amplitude, precise and controllable test stimuli for different learning objectives, while ensuring that the driver does not notice them. Simultaneously, it acquires key response data before and after the application of the stimuli at high speed (the acquisition period is preset according to the parameter accuracy requirements, such as 10ms).

[0048] The specific incentive methods for different learning objectives are as follows:

[0049] If the learning target is the clutch kiss point, then while keeping the torque transmitted in the current gear unchanged, the control pressure of the target clutch is increased in a stepwise manner to obtain the pressure increment, while keeping the working state of other clutches unchanged.

[0050] If the learning objective is the system pressure reference, then fine-tune the duty cycle of the main oil pressure solenoid valve and observe the following characteristics of the actual system pressure.

[0051] The data collected includes: changes in the input and output shaft speeds, changes in turbine speed, actual pressure of the target clutch, actual transmission ratio, and engine output torque.

[0052] S4: Parameter Calculation and Update: Substitute the response data collected in S3 into the physical or statistical model preset in the TCU for calculation to obtain the corresponding new steady-state control parameter values. Compare the calculated new parameter values ​​with the original calibration values ​​stored in the TCU, and generate smooth updated parameter values ​​by using first-order low-pass filtering or moving average filtering to avoid noise interference from single learning.

[0053] The calculations of the physical or statistical model include:

[0054] For clutch kiss point learning, after analyzing the applied pressure increment, observe whether there is slight slippage in the input and output shaft speeds (manifested as a slight deviation of the actual transmission ratio from the theoretical transmission ratio), and accurately calculate the instantaneous pressure point at which the current clutch begins to transmit torque, i.e., the real-time kiss point.

[0055] For system pressure reference learning, the system pressure reference correction value is calculated by analyzing the deviation between the actual system pressure and the target pressure and the response speed after the duty cycle of the main oil pressure solenoid valve is fine-tuned.

[0056] S5: Learning Result Verification and Storage: After a single parameter update, the system enters a short-term verification phase. In subsequent shifting operations involving the updated parameters, the TCU continuously monitors shifting quality indicators (such as shifting impact and slip friction). If the shifting quality remains or improves within the expected range, the learning is confirmed to be effective. The verified effective parameter values ​​are written into the TCU through the non-volatile memory interface to permanently replace the old parameters, serving as the basis for all subsequent shifting control parameters.

[0057] S6: Learning State Management and Exit: The entire self-learning process employs a state machine management mechanism. At any stage from S2 to S5, if the system detects a situation that does not meet safety or steady-state conditions (such as the driver suddenly accelerating, braking, requesting a gear shift, or the vehicle's operating state exceeding a preset range), it immediately interrupts the current self-learning process, clears all temporarily collected data, and restores normal driving mode to ensure driving safety. The system continuously monitors the operating conditions and waits for the next learning window to restart the learning process.

[0058] Example 1: Clutch Kiss Point Steady-State Self-Learning

[0059] 1. Prerequisites and Definitions

[0060] Target clutch: The C1 shift clutch commonly used in automatic transmissions is selected as the learning target;

[0061] Kiss Point: Defined as the hydraulic control valve command current or pressure value at which the clutch just begins to transmit a small torque (enough to overcome idle drag torque and generate a measurable speed slip) at the current transmission oil temperature. This value is a core reference parameter based on all shift curves (such as torque phase fill curves) of the C1 clutch.

[0062] Steady-state operating condition selection: Select the operating condition where the vehicle is cruising stably with a small throttle opening in a high gear (such as 6th gear). At this time, the C1 clutch is fully engaged, the engine torque output is stable, the vehicle has a large inertia, and the sensitivity to micro-disturbances is low, which can ensure the safety and accuracy of the learning process.

[0063] 2. Detailed Implementation Steps

[0064] S1: Self-learning enablement condition judgment

[0065] The TCU's steady-state self-learning module runs continuously after each vehicle power-on, checking the following global enable flags in real time:

[0066] The transmission fluid temperature sensor reading is within the range of [65℃, 105℃] (this range is the optimal range for hydraulic fluid characteristics and clutch friction characteristics).

[0067] Neither the EMS nor the TCU itself reported any fault codes related to torque control, speed monitoring, or clutch control.

[0068] Since successfully learning the C1 clutch kiss point last time, the vehicle has accumulated more than 500 kilometers;

[0069] The vehicle battery voltage is >12.2V and <15.5V to ensure stable solenoid valve drive voltage;

[0070] The driving mode selector is set to NORMAL or ECO mode, excluding SPORT, MANUAL, and TRACTION modes.

[0071] The steady-state self-learning module transitions from the waiting state to the ready state only when all of the above flags are true.

[0072] S2: Steady-state condition identification and capture

[0073] In the ready state, the steady-state self-learning module continuously monitors the following vehicle operation signals to identify the learning window for 6-gear steady-state cruise control:

[0074] The current gear is 6th gear;

[0075] The engine torque is within the range of [30Nm, 80Nm], and the standard deviation of torque over the past 5 seconds is <5Nm;

[0076] The absolute value of vehicle acceleration is <0.1 m / s².

[0077] Abrupt changes in brake and accelerator pedal operation (peder operation change rate is lower than preset threshold).

[0078] All of the above states last for 15 seconds.

[0079] Once all the above conditions are met, the steady-state self-learning module enters the "window activation" state and starts an 8-second learning time window.

[0080] S3: Apply controlled test stimuli and collect data

[0081] Baseline data acquisition (first 2 seconds): Within the first 2 seconds after the learning time window starts, the steady-state self-learning module continuously records the current control current of the target clutch C1. Turbine speed Output shaft speed And calculate the theoretical transmission ratio. (This value should be close to the theoretical gear ratio of 6th gear).

[0082] Applying excitation: At the second second of the learning time window, the TCU sends a precise current step command to the control solenoid valve of clutch C1 to test the current. ,in A calibrated, minute current increment (e.g., 20mA) corresponds to an increase of approximately 0.15 bar in clutch control pressure. This current increment is precisely designed to cause only a slight increase in the clamping force between the clutch friction plates, resulting in a detectable over-clamping effect in the fully engaged state, but the resulting additional torque... Much less than the current engine output torque The vehicle's acceleration changes very little, and the driver is unaware of it.

[0083] Excitation response data acquisition: when the test current is applied Within the next 4 seconds, the following data is collected at high speed with a period of 10ms: actual transmission ratio ,in For real-time turbine speed, The output shaft speed, C1 clutch pressure sensor reading (if the TCU is equipped with this sensor), engine real-time torque, and output shaft real-time speed are displayed in real time.

[0084] S4: Parameter Calculation and Update

[0085] Data analysis: After the excitation is applied, due to the over-engagement of clutch C1, its torque transmission capacity is momentarily slightly higher than the actual demand of the engine, resulting in a slight braking load on the input shaft (turbo), which manifests as an actual transmission ratio. Negative glitch appears within 100-300ms (deviation from theoretical transmission ratio) (Approximately 0.5%-1%), which was subsequently restored to a stable value due to the engine EMS torque closed-loop control adjustment;

[0086] Model calculation: First, identify the start and end times of the transmission ratio glitch, and calculate the actual transmission ratio within that time period. Compared with the theoretical transmission ratio integral error (This integral error is a small manifestation of slip friction work); based on the clutch hydraulic-torque characteristic model preset in the TCU, the integral error... The current clutch pressure exceeds the pressure value of its actual kiss point. There is a fixed mapping relationship, which can be achieved by looking up a table or simplifying a formula. ,in To determine the pressure corresponding to the current kiss point, a preset calibration coefficient is used. Then, based on the current-pressure characteristics (IP curve) of the solenoid valve, the pressure is... Convert to the corresponding current command ;

[0087] Filtering Update: To avoid noise interference from a single learning iteration, a first-order low-pass filter algorithm is used to smooth the parameters. The filtering formula is as follows: ,in The old kiss point current value stored in NVM. These are the filter coefficients (ranging from 0.3 to 0.6, representing the level of confidence in the new data being learned); if this is the first time learning, then... Use the factory calibration values.

[0088] S5: Learning Result Validation and Storage

[0089] Short-term verification: During this vehicle driving cycle, if a gear shift involving the C1 clutch occurs subsequently (e.g., downshifting from 6th to 5th gear), the TCU will use the updated... As the starting point for the C1 clutch torque phase control, the learning and verification unit monitors the impact index (acceleration change rate) of the shift in real time. If the impact is within the preset good range, the learning and verification is marked as successful.

[0090] Storage: At the end of this vehicle ignition cycle, if the learning verification is successful, the steady-state self-learning module will store the data via the NVM interface. Write the specified storage location to the TCU's EEPROM (non-volatile memory) to permanently overwrite the old value, and simultaneously reset the learning mileage counter of clutch C1.

[0091] S6: Learning Status Management and Exit

[0092] During any of the S2-S5 phases, if the steady-state self-learning module detects any of the following interruption conditions, it will immediately execute the learning interruption procedure: the driver depresses the brake pedal (brake pedal travel > 5%), the throttle opening changes abruptly (rate of change > 500% / s), a shift request is received (manual or automatic), the engine torque changes suddenly by > 50 Nm, or the learning time window expires. During the interruption, the system clears all temporarily collected data and calculation results, seamlessly reverting to normal driving control mode without affecting driver operation; it will then continue to monitor the operating conditions and wait for the next learning window to restart learning.

[0093] Example 2: System Pressure Reference Steady-State Self-Learning

[0094] 1. Prerequisites and Definitions

[0095] Learning objective: To compensate for the system pressure reference offset caused by oil pump wear and valve body characteristic drift, and to ensure the accuracy of main oil pressure control;

[0096] Steady-state operating condition selection: Select the operating condition where the vehicle is in neutral (N) or park (P), the engine is idling, and the transmission oil temperature is stable. Under this condition, the system load is stable and the response to pressure fine adjustments is more accurate.

[0097] 2. Detailed Implementation Steps

[0098] S1: Self-learning enablement condition judgment

[0099] The global enable conditions include: transmission oil temperature at [60℃, 110℃], no related fault codes in EMS and TCU, vehicle cumulative running time reaching the system pressure reference learning trigger cycle, normal battery voltage, and vehicle in non-aggressive driving mode. When all conditions are met simultaneously, the vehicle enters the ready state.

[0100] S2: Steady-state condition identification and capture

[0101] Steady-state operating conditions: The vehicle is in N or P gear, the engine speed is stable in the idle range (e.g., 650-750 rpm), the transmission oil temperature fluctuates by less than 5℃ / minute, and there is no operation of the accelerator pedal or brake pedal. If the above conditions last for more than 10 seconds, it is marked as the learning window period.

[0102] S3: Apply controlled test stimuli and collect data

[0103] Reference data acquisition: Record the reference duty cycle of the current main oil pressure solenoid valve. Actual system pressure (Data collected by the main oil pressure sensor);

[0104] Apply excitation: Send a precision duty cycle adjustment command to the main hydraulic solenoid valve. ,in This is a preset small increment, such as 2%;

[0105] Response data acquisition: With a period of 10ms, the actual system pressure, solenoid valve duty cycle feedback value, and engine idle speed are collected within 5 seconds after the excitation is applied, and the pressure set-up time and pressure following deviation are recorded.

[0106] S4: Parameter Calculation and Update

[0107] Based on the collected excitation-response data, the following characteristics and deviations of the system pressure are analyzed, the correction value of the mapping relationship between the duty cycle of the main oil pressure solenoid valve and the actual pressure is calculated, and combined with the preset pressure control model, the updated system pressure reference parameters are obtained. After filtering, the final updated value is generated.

[0108] S5: Learning Result Validation and Storage

[0109] During subsequent gear shifting in vehicle operation, the shifting time (related to the hydraulic oil filling speed) is monitored. If the shifting time is within a preset reasonable range, the verification is successful, and the updated system pressure reference parameters are written to the EEPROM for permanent storage.

[0110] S6: Learning Status Management and Exit

[0111] Similar to Example 1, if a change in operating conditions or a safety risk is detected, learning is immediately interrupted, normal control is restored, and the system waits for the next learning window.

[0112] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, it is intended that all variations falling within the meaning and scope of the equivalents of the claims be included within the present invention.

[0113] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A steady-state self-learning method for an automatic transmission, characterized in that: The method includes the following steps: S1: Self-learning enable condition judgment: The system continuously monitors a set of global enable conditions, and only enters the steady-state self-learning preparation state when all conditions are met at the same time. S2: Steady-state condition identification and capture: When all global enable conditions in S1 are met, the system begins to continuously identify steady-state conditions that meet the preset conditions, marks them as learning windows, and starts the learning preparation process. S3: Applying controlled test stimuli and collecting data: During the learning window, the TCU actively applies test stimuli for different learning objectives and simultaneously collects key response data before and after the application of the stimuli; S4: Parameter Calculation and Update: Substitute the response data collected in S3 into the physical model or statistical model preset in the TCU for calculation to obtain the corresponding new steady-state control parameter values. Compare the calculated new parameter values ​​with the original calibration values ​​stored in the TCU, and generate smooth updated parameter values ​​through filtering. S5: Learning Result Verification and Storage: After a single parameter update, the system enters a short-term verification phase. In subsequent gear shifting operations involving the updated parameters, the TCU continuously monitors the gear shifting quality indicators. If the gear shifting quality is within the expected range, the learning is confirmed to be effective, and the verified effective parameter values ​​are written into the TCU. S6: Learning State Management and Exit: The entire self-learning process adopts a state machine management mechanism. In any stage of S2-S5, if the system detects that the safety conditions or steady-state conditions are not met, the current self-learning process will be interrupted immediately, all temporarily collected data will be cleared, and the normal driving mode will be restored. The system will continue to monitor the operating conditions and wait for the next learning window to restart the learning process.

2. The automatic transmission steady-state self-learning method according to claim 1, characterized in that: The global enable conditions mentioned in S1 include: Condition 1: The transmission fluid temperature is within the optimal learning range; Condition 2: Neither EMS nor TCU reported any fault codes; Condition 3: The vehicle's mileage or operating time reaches the preset self-learning trigger cycle; Condition 4: The vehicle battery voltage is within the normal operating range; Condition 5: The vehicle is in a non-aggressive driving mode.

3. The automatic transmission steady-state self-learning method according to claim 2, characterized in that: The steady-state operating condition described in S2 must simultaneously meet the following conditions: The vehicle is in a stable cruising speed state, a slow acceleration state or slow deceleration state where the absolute value of acceleration is lower than a preset threshold, a stable engine torque output state where the torque fluctuation rate within the past preset time period is lower than a set threshold, and the transmission is not in the process of shifting gears and the current gear is fixed, and the duration of all the above states exceeds the preset time period.

4. The automatic transmission steady-state self-learning method according to claim 3, characterized in that: The specific incentive methods for different learning objectives described in S3 are as follows: If the learning target is the clutch kiss point, then while keeping the torque transmitted in the current gear unchanged, the control pressure of the target clutch is increased in a stepwise manner to obtain the pressure increment, while keeping the working state of other clutches unchanged. If the learning objective is the system pressure reference, then adjust the duty cycle of the main oil pressure solenoid valve and observe the following characteristics of the actual system pressure.

5. The automatic transmission steady-state self-learning method according to claim 4, characterized in that: The data collected in S3 includes: changes in the speed of the input and output shafts, changes in the turbine speed, the actual pressure of the target clutch, the actual transmission ratio, and the engine output torque.

6. The automatic transmission steady-state self-learning method according to claim 5, characterized in that: The calculations of the physical or statistical model described in S4 include: For clutch kiss point learning, by analyzing the applied pressure increment and observing whether slippage occurs at the input and output shaft speeds, the instantaneous pressure point at which the clutch begins to transmit torque is calculated, i.e., the real-time kiss point. For system pressure reference learning, the system pressure reference correction value is calculated by analyzing the deviation between the actual system pressure and the target pressure and the response speed after the duty cycle of the main oil pressure solenoid valve.