Electric vehicle transmission kickdown control method

Abstract

The application relates to the field of intelligent control, and particularly discloses a downshift control method for an electric vehicle transmission, which acquires system states such as vehicle speed and braking intention in real time, and analyzes the expected braking torque of a driver. Then, an adaptive shift disturbance torque prediction model is introduced, which takes the planned speed and acceleration time sequence in the motor speed regulation process as input, and uses the powerful nonlinear fitting capability of a deep learning model to predict the disturbance torque time sequence curve generated due to speed change during the whole shift. Finally, the driver intention torque and the predicted disturbance torque time sequence curve are fused through feedforward compensation to generate a smooth and accurate motor target torque instruction.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent control, and more particularly, to a control method for downshift of electric vehicle transmission. BACKGROUND

[0002] With the rapid development of new energy vehicle technology, electric vehicles have become an important part of the automotive industry. In order to further improve the power, economy and driving performance of electric vehicles, some electric vehicles begin to use multi-gear transmissions to broaden the high-efficiency working interval of the driving motor and optimize the vehicle energy management. However, the introduction of multi-gear transmission also brings new challenges, especially during the shifting process, the power transmission path will be temporarily interrupted or changed, resulting in fluctuations in driving torque, which will cause longitudinal impact of the vehicle, i.e. the so-called shift shock. Especially in the downshift working condition, for example, when the vehicle decelerates or brakes, the motor speed needs to be quickly raised to match the transmission ratio of the target low gear. If not properly controlled, this process is prone to produce a noticeable jerk, which seriously affects the smoothness and comfort of driving, which is contrary to the quiet and smooth driving experience that electric vehicles should have. Therefore, how to accurately and smoothly control the downshift process of electric vehicle transmission has become a key technical problem to improve the overall quality and market competitiveness of electric vehicles.

[0003] Currently, for the shift control of electric vehicles, the existing technology usually adopts a control strategy based on experience calibration. These methods usually pre-set the motor compensation torque value or fixed torque change rate under specific working conditions, and compensate for the torque interruption during the shifting process through table lookup or simple open-loop control. However, the disadvantages of this method are obvious: first, the actual operating conditions of the vehicle are complex and variable, and the system state (such as battery state of charge, transmission oil temperature, component wear, etc.) is also constantly changing. Fixed calibration data cannot adapt to these changes in real time, resulting in poor control effect under non-calibration conditions and difficulty in ensuring ideal smoothness in the whole life cycle and full working condition range. Second, the traditional control model often oversimplifies the shifting process, ignoring the complex nonlinear dynamic characteristics therein, and thus cannot accurately predict the disturbance torque actually generated during the shifting process, resulting in limited compensation accuracy and difficulty in completely eliminating the shift shock.

[0004] Therefore, an optimized control scheme for downshift of electric vehicle transmission is desired. SUMMARY

[0005] To solve the above technical problems, the present application is proposed. The embodiments of the present application provide a control method for downshift of electric vehicle transmission.

[0006] According to one aspect of the present application, a control method for downshift of electric vehicle transmission is provided, which comprises:

[0007] in response to a downshift instruction, collecting a system state vector, the system state vector comprising vehicle speed, brake pedal position, current gear, battery state of charge, and transmission oil temperature, the downshift instruction comprising a target gear;

[0008] querying a smoothed brake torque from a brake MAP table indexed by vehicle speed and brake pedal position, and converting the smoothed brake torque to a motor-side driver intent torque based on a gear ratio of the current gear;

[0009] performing adaptive shift disturbance torque prediction on the system state vector to obtain a predicted shift disturbance torque time series based on a target speed difference and a predicted shift time;

[0010] performing compensation fusion on the motor-side driver intent torque and the predicted shift disturbance torque time series to obtain a compensated motor target torque time series;

[0011] sending the compensated motor target torque time series to a motor MCU, while a transmission controller executes the downshift instruction.

[0012] In the above electric vehicle transmission downshift control method, converting the smoothed brake torque to a motor-side driver intent torque based on a gear ratio of the current gear comprises: converting the smoothed brake torque to the motor-side driver intent torque based on the gear ratio of the current gear by the following formula:

[0013]

[0014] wherein, is the gear ratio of the current gear, is the smoothed brake torque.

[0015] In the above electric vehicle transmission downshift control method, performing adaptive shift disturbance torque prediction on the system state vector to obtain a predicted shift disturbance torque time series based on a target speed difference and a predicted shift time comprises: calculating a current motor angular velocity based on vehicle speed, a gear ratio of the current gear, and wheel diameter; calculating a target motor angular velocity based on vehicle speed, a gear ratio of the target gear, and wheel diameter; calculating the target speed difference based on the current motor angular velocity and the target motor angular velocity; calculating the predicted shift time based on the target speed difference; inputting the current motor angular velocity, the target motor angular velocity, and the predicted shift time into a trajectory generator to obtain a time series of planned speed and a time series of planned acceleration; and performing adaptive shift disturbance torque prediction based on the time series of planned speed and the time series of planned acceleration to obtain the predicted shift disturbance torque time series.

[0016] In the electric vehicle transmission downshift control method, the predicted shift time is calculated based on the target speed difference, including: based on the current motor angular velocity and the target motor angular velocity, the kinetic energy change required for the speed change is calculated in the following formula:

[0017]

[0018] wherein, is the equivalent rotational inertia, is the target motor angular velocity, is the current motor angular velocity; the kinetic energy change required for the speed change is divided by the average motor power to obtain the predicted shift time.

[0019] In the electric vehicle transmission downshift control method, the adaptive shift disturbance torque prediction is performed based on the time sequence of the planned speed and the time sequence of the planned acceleration to obtain the predicted shift disturbance torque time curve, including: the time sequence of the planned speed and the time sequence of the planned acceleration are sequentially encoded to obtain a planned speed time sequence hidden encoding vector and a planned acceleration time sequence hidden encoding vector; the planned speed time sequence hidden encoding vector and the planned acceleration time sequence hidden encoding vector are interacted with parameter fine-grained features to obtain a planned speed-planned acceleration fine-grained interaction hidden encoding vector; the planned speed-planned acceleration fine-grained interaction hidden encoding vector is decoded to obtain the predicted shift disturbance torque time curve.

[0020] In the electric vehicle transmission downshift control method, the time sequence of the planned speed and the time sequence of the planned acceleration are sequentially encoded to obtain a planned speed time sequence hidden encoding vector and a planned acceleration time sequence hidden encoding vector, including: the time sequence of the planned speed and the time sequence of the planned acceleration are passed through an LSTM-based sequence encoder to obtain the planned speed time sequence hidden encoding vector and the planned acceleration time sequence hidden encoding vector.

[0021] In the aforementioned electric vehicle transmission downshift control method, the planned speed temporal implicit coding vector and the planned acceleration temporal implicit coding vector are subjected to dense interaction of parameters with fine-grained features to obtain a planned speed-planned acceleration fine-grained interactive implicit coding vector. This includes: performing local feature analysis and dimensional unification on the planned speed temporal implicit coding vector to obtain the sequence distribution of the planned speed local temporal granular coding vector; inserting the planned acceleration temporal implicit coding vector into different positions of the sequence distribution of the planned speed local temporal granular coding vector, and then inputting it into an interactive encoder based on a bidirectional LSTM model to obtain a set of planned speed-planned acceleration temporal cross-position interactive coding feature vectors; and dynamically aggregating the set of planned speed-planned acceleration temporal cross-position interactive coding feature vectors to obtain the planned speed-planned acceleration fine-grained interactive implicit coding vector.

[0022] In the above-mentioned electric vehicle transmission downshift control method, the time-series curves of the driver's intended torque at the motor end and the predicted shift disturbance torque are compensated and fused to obtain the compensated target torque time-series curve of the motor. This includes: adding the driver's intended torque at the motor end and the predicted shift disturbance torque at each time point in the time-series curve of the predicted shift disturbance torque to obtain the compensated target torque time-series curve of the motor.

[0023] Compared with existing technologies, this application provides a downshift control method for electric vehicle transmissions. It analyzes the driver's desired braking torque by real-time acquisition of system states such as vehicle speed and braking intention. Then, an adaptive shift disturbance torque prediction model is introduced. This model takes the planned speed and acceleration time series during motor speed regulation as input and utilizes the powerful nonlinear fitting capability of deep learning models to predict the time-series curve of disturbance torque caused by speed changes throughout the shifting process. Finally, the driver's intended torque and the predicted disturbance torque time-series curve are fused with feedforward compensation to generate a smooth and accurate target torque command for the motor. In this way, torque fluctuations during downshifting are actively and accurately offset, effectively solving the problems of low control accuracy and poor adaptability of traditional methods, and improving the smoothness of downshifting and driving comfort of electric vehicles under various operating conditions. Attached Figure Description

[0024] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0025] Figure 1Flowchart for downshift control method for electric vehicle transmission according to embodiments of the present application;

[0026] Figure 2 Flowchart for adaptive shift disturbance torque prediction for system state vector to obtain predicted shift disturbance torque time series based on target speed difference and predicted shift time for downshift control method for electric vehicle transmission according to embodiments of the present application;

[0027] Figure 3 Flowchart for adaptive shift disturbance torque prediction based on time series of planned speed and time series of planned acceleration to obtain predicted shift disturbance torque time series for downshift control method for electric vehicle transmission according to embodiments of the present application;

[0028] Figure 4 Flowchart for parameter fine-grained feature dense interaction of planned speed time series implicit encoding vector and planned acceleration time series implicit encoding vector to obtain planned speed-planned acceleration fine-grained interaction implicit encoding vector for downshift control method for electric vehicle transmission according to embodiments of the present application. DETAILED DESCRIPTION

[0029] Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part but not all of the embodiments of the present application. It should be understood that the present application is not limited to the described embodiments.

[0030] As shown in the present application and claims, unless the context clearly indicates otherwise, the words "one", "an", "a", and / or "the" do not mean "only one", but can include a plurality or "one or more" unless the context clearly indicates otherwise. Generally, the terms "comprise", "comprising", "include", "including" and the like are intended to indicate a detailed inclusion, but do not exclude other non-identified steps or elements.

[0031] Although the present application makes various references to certain modules in the system according to embodiments of the present application, however, any number of different modules can be used and run on the user terminal and / or server. The modules are merely illustrative, and different aspects of the system and method can use different modules.

[0032] Flowcharts are used in the present application to illustrate the operations performed by the system according to embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in sequence. Instead, various steps can be processed in reverse order or simultaneously, as needed. Other operations can also be added to these processes, or one or more steps can be removed from these processes.

[0033] Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part but not all of the embodiments of the present application, and the present application can be implemented in many different forms. Therefore, the attached drawings should be considered as one of the embodiments of the present application, but not a limitation of the present application.

[0034] The existing downshift control strategy for electric vehicles mostly relies on fixed experience calibration, which is difficult to adapt to complex and variable actual working conditions, resulting in the technical problems of shift shock and poor driving experience. In the technical solution of the present application, a downshift control scheme based on dynamic prediction and adaptive compensation is proposed, i.e. an electric vehicle transmission downshift control method. After receiving the downshift instruction, the method first collects real-time states such as vehicle speed and brake pedal position, queries and calculates the smooth brake torque that can accurately reflect the driver's deceleration intention from the brake MAP table. At the same time, in order to actively respond to the power system disturbance inevitably produced in the shift process, the core is to build a precise disturbance torque prediction process: it calculates the target speed difference that the motor needs to adjust according to the transmission ratio of the current gear and the target gear, and predicts the shift time required to complete the speed regulation based on this. Then, input these parameters into the trajectory generator to plan the ideal speed and acceleration change curve of the motor within the entire shift time. These time series containing rich dynamic information will be sent to an adaptive prediction model based on deep learning, which can accurately output a predicted shift disturbance torque time sequence curve through the fine-grained feature interaction and decoding of the speed and acceleration sequence. Finally, the smooth brake torque representing the driver's intention is accurately compensated and fused with the predicted disturbance torque time sequence curve to generate the final compensated motor target torque time sequence curve, which is sent to the motor controller for execution. In this way, the downshift disturbance can be accurately offset in a forward-looking and whole-process manner, effectively solving the problems of poor adaptability and low control precision of traditional methods, and achieving comfortable downshift experience in various working conditions.

[0035] Figure 1 The flowchart of the electric vehicle transmission downshift control method according to the embodiments of the present application. As shown in FIG. 1, the downshift control method according to the embodiments of the present application mainly includes the following steps: Figure 1As shown, the electric vehicle transmission downshift control method according to the embodiments of the present application comprises: S100, in response to a downshift instruction, collecting a system state vector, the system state vector comprising vehicle speed, brake pedal position, current gear, battery state of charge and transmission oil temperature, the downshift instruction comprising a target gear; S200, querying a smooth brake torque from a brake MAP table with vehicle speed and brake pedal position as indexes, and converting the smooth brake torque into a motor end driver intent torque based on the transmission ratio of the current gear; S300, based on a target speed difference and a predicted shift time, performing adaptive shift disturbance torque prediction on the system state vector to obtain a predicted shift disturbance torque time sequence curve; S400, compensating and fusing the motor end driver intent torque and the predicted shift disturbance torque time sequence curve to obtain a compensated motor target torque time sequence curve; S500, sending the compensated motor target torque time sequence curve to the motor MCU, while the transmission controller executes the downshift instruction.

[0036] Specifically, in step S100, in response to a downshift instruction, a system state vector is collected, the system state vector comprising vehicle speed, brake pedal position, current gear, battery state of charge and transmission oil temperature, the downshift instruction comprising a target gear. It should be understood that the shift process is a complex dynamic process closely coupled with multiple vehicle subsystems, and the control effect is affected by a variety of time-varying factors. Therefore, in response to the downshift instruction, the system state vector is collected, which can comprehensively and in real time grasp the accurate working state of the vehicle at the moment of downshift, provide necessary high-dimensional input information for subsequent accurate torque prediction and compensation control, and ensure that the control strategy can adapt to the current specific physical environment and operating conditions, thereby achieving adaptive and high-precision control effect. By collecting these key state parameters such as vehicle speed, brake pedal position, current gear, battery state of charge and transmission oil temperature, control inaccuracy caused by fixed parameters can be avoided, and the robustness and accuracy of the control model can be improved, ultimately achieving smooth and impact-free downshift.

[0037] More specifically, in one specific example of the present application, first, the vehicle control unit (VCU) as the central hub of the system, continuously monitors the driver's operation intention and the vehicle's running status. When the VCU judges that the current working condition meets the downshift condition (for example, based on the shift strategy map of vehicle speed and accelerator pedal opening) or receives the driver's forced downshift instruction through the shift paddle, etc., it will generate a downshift instruction containing the target gear. Second, in response to this downshift instruction, the VCU will immediately trigger a data acquisition task. It broadcasts data requests to each relevant subsystem controller through the controller area network (CAN bus) or directly reads the shared information on the bus. Specifically, it will obtain the accurate vehicle speed signal from the anti-lock braking system or vehicle stability system controller; obtain the driver's brake pedal position from the brake pedal sensor to judge the strength of the braking intention; obtain the current actual gear information from the transmission controller and the transmission oil temperature sensor reading, because the oil temperature will affect the response of the hydraulic system and the friction characteristics of the clutch; at the same time, obtain the current state of charge of the battery from the battery management system, because the state of charge of the battery will affect the maximum available power and torque response capability of the motor. The bit, state of charge of the battery and transmission oil temperature will be integrated into a structured data packet, i.e. the system state vector. This vector is passed to the subsequent torque calculation and prediction module as a timestamp-accurate snapshot. That is, this system state vector can be directly used as a multi-dimensional input to feed a pre-trained deep neural network model, which can learn and infer the complex relationship required for subsequent control from these raw inputs, thereby starting the entire adaptive downshift control process.

[0038] Specifically, in step S200, the smooth braking torque is queried from the braking MAP table indexed by the vehicle speed and the brake pedal position, and the smooth braking torque is converted into the motor end driver intention torque based on the transmission ratio of the current gear. It should be understood that the driver's braking intention is usually expressed by the depression depth of the brake pedal, and the actual deceleration effect of the vehicle is not only affected by the pedal depth, but also closely related to dynamic factors such as the current vehicle speed. In order to convert the driver's vague braking intention into precise and controllable motor torque instructions and ensure the smoothness of the vehicle response during braking, a standardized and optimized mapping mechanism is needed. Therefore, in the technical solution of the present application, first, the driver's brake pedal input and real-time vehicle speed information are converted into a smooth and driver-desired vehicle braking torque, which represents the vehicle's deceleration demand at the wheel end, by consulting the pre-calibrated braking MAP table. Second, since the electric motor is a direct torque generator or absorber, and the driver's intention is reflected at the wheel end, the braking torque demand at the wheel end must be accurately calculated back to the motor end according to the transmission ratio of the current gear, so as to obtain the motor end driver intention torque. This can provide a highly consistent baseline torque instruction with the driver's intention for subsequent motor torque control, ensuring that the vehicle response during downshift braking meets the driver's expectations and avoids a jarring feeling, significantly improving the driving comfort.

[0039] More specifically, in the embodiment of the present application, the process is usually completed by the vehicle controller or the transmission controller. A two-dimensional braking MAP table is stored in the system, which is generated through a large number of tests and fine calibration, taking the vehicle speed and brake pedal position as input, and outputting the corresponding smooth braking torque, and using linear interpolation and other methods to process non-calibration points to ensure continuity. Then, the real-time transmission ratio data of the current gear is used to convert the smooth braking torque at the wheel end into the driver intention torque at the motor end through simple mathematical operations (i.e., the motor end torque is equal to the wheel end torque divided by the transmission ratio). Specifically, in the embodiment of the present application, the smooth braking torque is converted into the motor end driver intention torque based on the transmission ratio of the current gear, including: based on the transmission ratio of the current gear, the smooth braking torque is converted into the motor end driver intention torque by the following formula:

[0040]

[0041] wherein, is the transmission ratio of the current gear, is the smooth braking torque.

[0042] Specifically, in step S300, the system state vector is adaptively predicted for a shift disturbance torque based on a target speed difference and a predicted shift time to obtain a predicted shift disturbance torque time curve. It should be understood that a conventional control method cannot accurately predict and quantify the dynamic disturbance caused by the rapid speed regulation of the motor during the shift process. When downshifting, the motor speed needs to jump from the current value to the new speed matching the target low gear in a very short time. This sharp change in speed will inevitably cause significant inertial torque fluctuations, i.e., shift disturbance torque, which is the core physical source of vehicle impact and jerk. The size and shape of this disturbance is not fixed, it is closely related to the speed difference to be crossed, the allowed shift time and the current system state (such as the clutch response affected by the oil temperature, the motor capacity affected by the battery state, etc.). Therefore, in the technical solution of the present application, the system state vector is adaptively predicted for a shift disturbance torque based on a target speed difference and a predicted shift time. It no longer regards the disturbance as a simple error that can be compensated for afterwards, but as a predictable dynamic process related to multiple variables. By comprehensively considering the target speed difference and the predicted shift time that determine the core characteristics of the disturbance, and combining the state vector (vehicle speed, battery state of charge, transmission oil temperature, etc.) reflecting the current real physical characteristics of the system, a prediction model that can dynamically adapt to the current working condition is constructed. The goal of this model is to output a time curve that accurately describes how the disturbance torque changes over time during the entire shift period, rather than a single compensation value. The controller can instruct the motor to output a compensation torque of the same size and opposite direction according to this curve, thereby canceling the disturbance at the source. This ultimately achieves fundamental suppression of shift impact, regardless of vehicle speed, braking intensity, or vehicle operating state, ensuring a smooth downshift process like a flowing cloud, improving driving comfort and overall vehicle quality.

[0043] Figure 2 A flowchart of the adaptive shift disturbance torque prediction of the system state vector based on a target speed difference and a predicted shift time for the electric vehicle transmission downshift control method according to the embodiments of the present application. As shown in FIG. 4, the system state vector is adaptively predicted for a shift disturbance torque based on a target speed difference and a predicted shift time to obtain a predicted shift disturbance torque time curve. Figure 2As shown, step S300 includes: S310, calculating a current motor angular velocity based on the vehicle speed, the transmission ratio of the current gear, and the wheel diameter; S320, calculating a target motor angular velocity based on the vehicle speed, the transmission ratio of the target gear, and the wheel diameter; S330, calculating a target speed difference based on the current motor angular velocity and the target motor angular velocity; S340, calculating the predicted shift time based on the target speed difference; S350, inputting the current motor angular velocity, the target motor angular velocity, and the predicted shift time into a trajectory generator to obtain a time sequence of a planned speed and a time sequence of a planned acceleration; and S360, performing adaptive shift disturbance torque prediction based on the time sequence of the planned speed and the time sequence of the planned acceleration to obtain the predicted shift disturbance torque time sequence curve.

[0044] Specifically, in steps S310 and S320, the current motor angular velocity is calculated based on the vehicle speed, the transmission ratio of the current gear, and the wheel diameter, and the target motor angular velocity is calculated based on the vehicle speed, the transmission ratio of the target gear, and the wheel diameter. It can be understood that the motor is the core actuator of the powertrain system, and its speed is a key state quantity for control. The macro motion state (vehicle speed) of the vehicle and the micro rotation state of the motor are strictly related through the transmission system (transmission and wheels). In order to accurately regulate the speed of the motor, it is necessary to accurately know the starting speed of the motor before shifting and the target speed of the motor after shifting. Therefore, in the technical solution of the present application, the current motor angular velocity is calculated based on the vehicle speed, the transmission ratio of the current gear, and the wheel diameter, and the target motor angular velocity is calculated based on the vehicle speed, the transmission ratio of the target gear, and the wheel diameter. Macroscopic speed information at the vehicle level, which is easy to measure, is accurately converted into microcosmic angular velocity information at the motor level, which is directly used for control, aiming to quantify the specific speed regulation task required by the shift action for the motor, i.e., to determine the starting point (current motor angular velocity) and the end point (target motor angular velocity) of the motor speed adjustment. This provides the most basic and core input data for subsequent calculation of the target speed difference, the predicted shift time, and the planned speed regulation trajectory, ensures that the entire prediction and control logic is based on an accurate physical model, and is the logical starting point for achieving accurate disturbance torque prediction.

[0045] More specifically, in one specific example of the present application, first, a real-time vehicle speed signal from the CAN bus is received, which is usually provided by the ABS / ESP system. At the same time, it obtains the current gear ratio from its own storage or the TCU and the target gear ratio contained in the downshift instruction. The wheel diameter, as a fixed physical parameter, is usually preset in the controller memory. Second, two independent calculations are performed. The first calculation is the current motor angular velocity, which converts the current vehicle speed (usually in km / h) into the angular velocity of the wheel (rad / s), and then multiplies the main reduction ratio and the gear ratio of the current gear, so as to obtain the current motor angular velocity strictly synchronized with the current vehicle speed. The second calculation is the target motor angular velocity, which uses the same current vehicle speed, but multiplies the main reduction ratio and the gear ratio of the target gear, to calculate the target angular velocity that the motor must reach in order to match the new gear after the shift is completed, while the vehicle speed remains unchanged. Finally, the calculated current motor angular velocity and target motor angular velocity, as two key values, will be output to the subsequent control algorithm. In a more integrated control system, this calculation logic can also be encapsulated inside a larger neural network model as an initial link in the model processing flow. The network can directly learn and infer these two angular velocity values from the vehicle speed and gear information, thereby serving a more complex end-to-end control strategy.

[0046] Specifically, in step S330, the target speed difference is calculated based on the current motor angular velocity and the target motor angular velocity. It should be understood that the core physical task of the downshift process is to drive the motor to complete an accurate speed adjustment, and the target speed difference is a direct quantification of the magnitude of this speed adjustment task. It connects the two isolated speed state points (current angular velocity and target angular velocity) calculated in the previous step to form a clear and directional speed adjustment requirement. It converts the abstract shift action into a specific and measurable physical quantity, i.e., the angular velocity variation that the motor needs to compensate. This target speed difference not only defines the distance of the speed adjustment, but also implies the direction of the speed adjustment (for downshift, it is the increase of the speed). It provides the most core input parameter for all subsequent energy and time-related calculations, and is the bridge connecting state perception and dynamic prediction. Specifically, the target motor angular velocity is subtracted from the current motor angular velocity, and the result is the target speed difference. The calculated target speed difference is stored as a key intermediate variable and immediately called by subsequent algorithm modules, such as for calculating the kinetic energy change required for the speed change or directly input into the model for predicting the shift time.

[0047] Specifically, in step S340, the predicted shift time is calculated based on the target speed difference. It should be understood that shift smoothness depends not only on how much speed needs to be adjusted (target speed difference), but also on how long it takes to do so. Time is the core dimension that determines the rate of energy change (i.e. power) and acceleration, and directly affects the magnitude and duration of shift shock. A fixed shift time that does not take into account the actual working conditions cannot adapt to changing system states. For example, when the battery is fully charged and the motor is in optimal performance, the speed regulation can be completed faster; otherwise, it needs more time. Therefore, in order to achieve truly adaptive control, the optimal or most likely time required to complete a specific speed adjustment under current conditions must be dynamically predicted.

[0048] More specifically, in the embodiments of the present application, the predicted shift time is calculated based on the target speed difference, including: first, based on the current motor angular velocity and the target motor angular velocity, the kinetic energy change required for speed change is calculated as follows:

[0049]

[0050] wherein, is the equivalent moment of inertia, is the target motor angular velocity, is the current motor angular velocity; then, the kinetic energy change required for speed change is divided by the average motor power to obtain the predicted shift time.

[0051] Specifically, in step S350, the current motor angular velocity, the target motor angular velocity and the predicted shift time are input into the trajectory generator to obtain the time sequence of the planned speed and the time sequence of the planned acceleration. It should be understood that the shift disturbance torque is not generated at once at the shift moment, but is a dynamic quantity that changes continuously throughout the speed regulation process, and its specific form is closely related to the process of motor speed change, not just the result. Simply knowing the starting point, end point and total duration is not enough to depict the details of this process. In order to accurately predict this dynamic disturbance torque, the ideal motion state of the motor at each moment during speed regulation must first be planned, i.e. how its speed and acceleration change over time. Therefore, the task of the trajectory generator is to generate an optimal motor angular velocity change curve under given time and speed constraints. This curve not only ensures that the motor accurately reaches the target speed within the specified time, but also ensures that its change process is smooth, avoiding sudden changes in speed or acceleration, because any sudden change will be physically translated into shock. By taking the derivative of this planned speed curve, a corresponding acceleration curve can be obtained simultaneously. Ultimately, this step aims to output two time sequences accurate to the millisecond level, i.e. the planned speed sequence and the planned acceleration sequence, which together completely define the ideal dynamic behavior of the motor during shifting.

[0052] More specifically, in one specific example of the present application, first, the trajectory generator receives three key inputs: the current motor angular velocity calculated in the previous step, the target motor angular velocity, and the predicted shift time. Second, a trajectory planning algorithm is employed inside the module, such as S-Curve or polynomial interpolation (e.g. quintic polynomial) algorithm. Take S-Curve as an example, it can guarantee continuous change of acceleration, thus minimizing jerk, which is suitable for scenarios that require smooth transition. The algorithm will solve a set of function equations describing the change of velocity over time, according to the three boundary conditions of the input. Finally, the module samples the velocity function at a pre-set time step (e.g. every 10 milliseconds) over the entire interval from 0 to the predicted shift time, generating a time series of planned velocity. Meanwhile, the first derivative of the velocity function is taken to get the acceleration function, and it is also sampled at the same time step, generating a time series of planned acceleration. These two sequences are then packed and output for use by the subsequent prediction model. In a preferred embodiment, a specially trained generative neural network (such as GAN or VAE) can also play the role of trajectory generator, which can learn and generate more optimized and smooth motion trajectories under specific working conditions.

[0053] Specifically, in step S360, adaptive shift disturbance torque prediction is performed based on the time sequence of the planned speed and the time sequence of the planned acceleration to obtain the predicted shift disturbance torque time curve. It should be understood that the generation mechanism of the shift disturbance torque is extremely complex and highly nonlinear. It is not only directly related to the inertia change of the motor, but also affected by the coupling of various time-varying factors such as the elasticity, damping, clearance and oil characteristics of the transmission chain. Traditional analytical methods based on physical models are difficult to accurately capture all these complex dynamics, and simple lookup tables or linear compensation cannot adapt to changing conditions. Therefore, a more powerful method is needed to learn and simulate this complex mapping relationship from the motor motion state to the system disturbance torque. Based on this, in the technical solution of the present application, adaptive shift disturbance torque prediction is further performed based on the time sequence of the planned speed and the time sequence of the planned acceleration, which can construct a high-precision, adaptive disturbance torque predictor. By learning a large amount of actual or simulation data, an end-to-end relationship between input (planned speed and acceleration sequence) and output (actual generated disturbance torque sequence) is directly established. The input planned speed and acceleration sequence is because they most directly drive the dynamic response of the system and contain the most critical time sequence information required for disturbance prediction. A time sequence curve is generated that is synchronized with the planned trajectory and can accurately reproduce the size and shape of the disturbance torque in the real physical process. By using deep neural networks to perform deep feature extraction and interactive analysis of the speed and acceleration sequence, the model can capture the subtle, nonlinear, and cross-time dependencies between the two and between them and the final disturbance torque. This means that the predicted disturbance torque time curve not only approaches the real value in value, but also highly coincides with the real disturbance in shape and phase. This high-fidelity prediction provides a benchmark for subsequent compensation control, so that the compensation torque can accurately and real-time offset every instant disturbance, thereby fundamentally eliminating the shift shock, and achieving consistent and extreme smooth downshift experience regardless of the complex driving scenarios or the aging state of the vehicle.

[0054] Figure 3 A flowchart for adaptive shift disturbance torque prediction based on the time sequence of the planned speed and the time sequence of the planned acceleration for the electric vehicle transmission downshift control method according to the embodiments of the present application. As Figure 3As shown, according to the electric vehicle transmission downshift control method of the embodiment of the application, step S360 comprises: S361, sequentially encoding the time sequence of the planned speed and the time sequence of the planned acceleration to obtain a planned speed time sequence implicit encoding vector and a planned acceleration time sequence implicit encoding vector; S362, performing parameter fine-grained feature dense interaction on the planned speed time sequence implicit encoding vector and the planned acceleration time sequence implicit encoding vector to obtain a planned speed-planned acceleration fine-grained interaction implicit encoding vector; and S363, performing feature decoding on the planned speed-planned acceleration fine-grained interaction implicit encoding vector to obtain the predicted shift disturbance torque time sequence curve.

[0055] Specifically, in step S361, the time sequence of the planned speed and the time sequence of the planned acceleration are sequentially encoded to obtain a planned speed time sequence implicit encoding vector and a planned acceleration time sequence implicit encoding vector. It can be understood that simple planned speed and planned acceleration instantaneous values or statistical features cannot capture the dynamic evolution process thereof. The planned speed and acceleration sequence itself is a language containing rich dynamic information, and LSTM (Long Short-Term Memory Network) as a neural network structure specially designed for processing and understanding such time sequence language can effectively remember the long-term dependence in the sequence and understand how the state at each time point evolves from the previous state. Therefore, in the embodiment of the application, the time sequence of the planned speed and the time sequence of the planned acceleration are input into an LSTM-based sequence encoder to obtain the planned speed time sequence implicit encoding vector and the planned acceleration time sequence implicit encoding vector. The time sequence of the planned speed and the time sequence of the planned acceleration are input, dimensioned and condensed into feature semantics capable of representing the complete dynamic state characteristics of each. This process is not a simple information compression, but a deep feature extraction process. The planned speed time sequence implicit encoding vector captures the macro trend and key inflection point of the motor speed from the start to the target throughout the change process; and the planned acceleration time sequence implicit encoding vector more finely depicts the rate and rhythm of the torque change. Through the LSTM encoder, the model can learn the smoothness of the speed, the peak value of the acceleration, and the cooperativity of the changes of the two, and other high-order features within the entire time window, which are crucial for accurately predicting the nonlinear disturbance torque.

[0056] Specifically, in step S362, the planning speed time sequence implicit encoding vector and the planning acceleration time sequence implicit encoding vector are interacted in a parameter fine-grained feature dense manner to obtain a planning speed-planning acceleration fine-grained interaction implicit encoding vector. It should be understood that, since the shift disturbance torque is not determined by a single factor of speed or acceleration, but is caused by the complex and nonlinear coupling effect between the two. The planning speed and the planning acceleration time sequence implicit encoding vectors, although each condenses the dynamic information of the source time sequence, the interaction relationship between them, for example, the change rate of acceleration when the speed reaches a certain peak, or the length of the speed plateau when the acceleration is zero, is the key to revealing the source of system disturbance. Traditional feature fusion methods, such as simple splicing or element-level operation, can only juxtapose information and cannot deeply explore such structural dependence, so a more thorough and fine interaction paradigm is needed. Based on this, in the technical solution of the present application, the planning speed time sequence implicit encoding vector and the planning acceleration time sequence implicit encoding vector are further interacted in a parameter fine-grained feature dense manner.

[0057] In particular, this parameter fine-grained feature dense interaction method discards the mode of shallowly fusing the planning speed time sequence implicit encoding vector and the planning acceleration time sequence implicit encoding vector as homogeneous information, but aims to answer a deeper question: what kind of systematic disturbance effect will be produced when a specific acceleration dynamic acts on a specific stage of the speed evolution process. In this feature interaction process, first, the planning speed time sequence implicit encoding vector with richer information is structured and deconstructed, which is deconstructed from a static point to a dynamic sequence composed of local granularity encoding. Then, the planning acceleration time sequence implicit encoding vector is systematically and densely embedded into all possible positions of the speed sequence, thereby creating a complete scenario set for the speed-acceleration interaction. By deeply encoding each scenario and finally intelligently aggregating the most critical speed-acceleration time sequence interaction mode, a unique planning speed-planning acceleration fine-grained interaction implicit encoding vector with a deep understanding of the complex relationship between the two is generated. This enables the predicted shift disturbance torque time sequence curve to accurately reproduce the complex nonlinear dynamics caused by the joint action of speed and acceleration, improving the accuracy and depth of disturbance torque prediction, and thus providing the most solid algorithm support for realizing truly full-working-condition and high-fidelity gearless downshift.

[0058] Figure 4 A flowchart of the parameter fine-grained feature dense interaction of the planning speed time sequence implicit encoding vector and the planning acceleration time sequence implicit encoding vector to obtain a planning speed-planning acceleration fine-grained interaction implicit encoding vector according to the electric vehicle transmission downshift control method of the embodiments of the present application. As shown in FIG. 4, the parameter fine-grained feature dense interaction of the planning speed time sequence implicit encoding vector and the planning acceleration time sequence implicit encoding vector to obtain a planning speed-planning acceleration fine-grained interaction implicit encoding vector according to the electric vehicle transmission downshift control method of the embodiments of the present application includes the following steps. Figure 4As shown, according to the electric vehicle transmission downshift control method of the embodiment of the application, step S362 includes: S3621, performing local feature analysis dimension unification on the planning speed time sequence implicit coding vector to obtain a sequence distribution of planning speed local time sequence granularity coding vectors; S3622, inserting the planning acceleration time sequence implicit coding vector into different positions of the sequence distribution of planning speed local time sequence granularity coding vectors respectively, and then inputting the same into a bidirectional LSTM model-based interaction encoder to obtain a set of planning speed-planning acceleration time sequence cross-position interaction coding feature vectors; and S3623, performing dynamic aggregation on the set of planning speed-planning acceleration time sequence cross-position interaction coding feature vectors to obtain the planning speed-planning acceleration fine-grained interaction implicit coding vector.

[0059] Specifically, in step S3621, the planning speed time sequence implicit coding vector is subjected to local feature analysis dimension unification to obtain a sequence distribution of planning speed local time sequence granularity coding vectors, which is expressed by the following formula:

[0060]

[0061] wherein, represents the planning speed time sequence implicit coding vector, is a convolution weight matrix, is a convolution bias vector, is a sliding window size, represents the position of a feature value in the sequence, is an activation function, represents a planning speed local time sequence semantic embedding coding vector, is a bilinear interpolation operation, is a target dimension, is the number of feature values in the sequence, represents the sequence distribution of planning speed local time sequence granularity coding vectors.

[0062] It should be appreciated that the feature dense interaction process requires a substrate with internal structure and position information, while the original planning speed time sequence implicit encoding vector, although rich in information, is essentially a static, integral vector that lacks internal sequential structure for probe embedding. In order to achieve exhaustive exploration of the interaction between speed and acceleration, it is necessary to first deconstruct the speed feature as the interaction subject from an indivisible point into a dynamic sequence composed of multiple ordered parts, thereby giving it internal position and order concepts that can be explored. Therefore, in the technical solution of the present application, local feature analysis dimensional unification is performed on the planning speed time sequence implicit encoding vector, aiming to anatomize the macroscopic, implicit encoding vector representing the speed dynamics during the entire gear shifting, and explicitly separate its internal, finer-grained components (e.g., acceleration segment features in the initial speed regulation period, constant speed segment features in the middle period, and deceleration stabilization segment features in the final period). Through local feature analysis technology, the network first identifies and extracts these local speed time sequence dynamic segments implicitly contained in the overall vector. Subsequently, in order to organize these local information with possibly different dimensions or forms into a regular form suitable for subsequent sequence model processing, linear interpolation and other dimensional unification methods are used to ensure that each local encoding vector has the same length. Finally, an ordered sequence composed of multiple equal-length encoding vectors, i.e., the sequence distribution of planning speed local time sequence granularity encoding vectors, is generated.

[0063] Specifically, in step S3622, after inserting the planning acceleration time sequence implicit encoding vector into different positions of the sequence distribution of planning speed local time sequence granularity encoding vectors, respectively, it is input into the interaction encoder based on the bidirectional LSTM model to obtain a set of planning speed-planning acceleration time sequence cross-position interaction encoding feature vectors, which can be represented by the following formula:

[0064]

[0065] wherein represent the 1st, 2nd, 3rd, and 4th planning speed local time sequence granularity encoding vectors in the sequence distribution of planning speed local time sequence granularity encoding vectors, respectively, represents the planning acceleration time sequence implicit encoding vector, represents the sequence distribution of planning speed-planning acceleration local time sequence granularity insertion encoding vectors, represents the bidirectional LSTM model, represents the set of planning speed-planning acceleration time sequence cross-position interaction encoding feature vectors.

[0066] ​​​It should be appreciated that the generation of the shift disturbance torque is not simply the superposition of the speed and acceleration, but a complex result of their interaction and mutual modulation at a certain time node. In order to understand this coupling relationship, it is necessary to go beyond macro correlation analysis and delve into micro structural dependence. Therefore, by inserting the planning acceleration time sequence implicit encoding vector into different positions of the sequence distribution of the planning speed local time sequence granularity encoding vector, the dynamic information of the planning acceleration can be regarded as an information probe, which is systematically injected into each key stage of the planning speed evolution process to observe and quantify the global semantic changes it triggers. Each new sequence formed after inserting the planning acceleration time sequence implicit encoding vector into a specific position of the planning speed local time sequence granularity encoding vector sequence represents a unique interaction hypothesis, raising a specific question: what systemic consequences will occur when the motor acts on this specific stage of the speed evolution process with such an acceleration dynamic? Inputting these mixed sequences representing different interaction scenarios one by one into the interaction encoder based on the bidirectional LSTM model is to conduct in-depth and comprehensive context modeling for each hypothesis. The adoption of bidirectional LSTM is crucial because it can capture long-range dependencies in both directions before and after the insertion point, accurately understanding the global dynamic changes triggered by the acceleration probe at different positions, rather than just local effects. This process breaks down a complex, elusive planning speed-planning acceleration time sequence nonlinear coupling problem into a series of analyzable and comparable interaction scenarios, and generates high-quality feature representations for each scenario. This provides perfect input for the subsequent attention mechanism to dynamically identify and amplify the most critical interaction patterns, which is a decisive step to achieve ultra-high precision prediction of shift disturbance torque.

[0067] Specifically, in step S3623, the set of planning speed-planning acceleration time sequence cross-position interaction encoding feature vectors is dynamically aggregated to obtain the planning speed-planning acceleration fine-grained interaction implicit encoding vector, which is represented by the following formula:

[0068]

[0069] wherein, is the mean vector of the set of planning speed-planning acceleration time sequence cross-position interaction encoding feature vectors, represents vector concatenation, is the contribution weight matrix, represents the planning speed-planning acceleration time sequence contribution degree, represents the planning speed-planning acceleration fine-grained interaction implicit encoding vector.

[0070] It should be appreciated that the set of planning speed-planning acceleration fine-grained interaction implicit encoding vectors, although informative, has a non-uniform value distribution. That is, not all interactions of acceleration acting on different positions of speed have equal impact on the final shift disturbance torque. Interactions at certain positions can be the key driving factor of the impact, while interactions at other positions can have little effect. If a simple average or concatenation operation is used to process this set, the most critical interaction signals will inevitably be diluted, resulting in important information being drowned in a large amount of secondary information, so that an accurate and effective final representation cannot be formed. Therefore, by introducing a dynamic aggregation strategy such as an attention mechanism, the contribution or importance score of each interaction configuration (i.e., each planning speed-planning acceleration time sequence position interaction encoding feature vector) can be adaptively learned and evaluated. It no longer treats all planning speed-planning acceleration time sequence interactions equally, but according to the current working conditions and the data itself, gives higher weights to those interactions that best reveal the core coupling relationship between speed and acceleration, while suppressing irrelevant or redundant interaction signals. Finally, by weighted summing the set of feature vectors, the network can intelligently and dynamically synthesize a unique, highly refined final encoding vector, i.e., a planning speed-planning acceleration fine-grained interaction implicit encoding vector. This highly condensed and highly insightful encoding vector provides the highest quality, highest signal-to-noise ratio input to the subsequent decoder, enabling it to predict the complete time series curve of the shift disturbance torque, thereby ensuring that the electric vehicle can achieve a silky downshift experience in any complex scenario.

[0071] Specifically, in step S363, the planning speed-planning acceleration fine-grained interaction implicit encoding vector is feature-decoded to obtain the predicted shift disturbance torque time series curve. It should be appreciated that the planning speed-planning acceleration fine-grained interaction implicit encoding vector, although containing deep insights into the disturbance generation mechanism, is not the final control target. The control system needs a specific, executable, time-varying torque compensation instruction, i.e., an accurate disturbance torque time series curve. Therefore, a decoding process is necessary to restore and expand the abstract, high-dimensional planning speed-planning acceleration fine-grained interaction implicit encoding vector into a specific, time-distributed torque time series. Through decoding, an instruction curve can be generated that specifies how much torque needs to be compensated at each time in the next few hundred milliseconds. This process aims to ensure that the predicted torque curve is not only accurate in total amount, but also highly consistent with the real physical disturbance in shape, peak value, and phase.

[0072] More specifically, in one specific example of the present application, firstly, the decoder module receives the unique and most critical input: the planning speed-planning acceleration fine-grained interaction-implicit encoding vector generated by the pre-sequence dynamic aggregation step. Secondly, the decoder usually adopts a recurrent neural network structure inside, such as LSTM or GRU, which echoes the encoder. It takes the input context vector as the initial hidden state, and then generates the output sequence in an autoregressive manner within a preset time step (consistent with the time step of the planning trajectory). At the first time step, it generates the first torque prediction value according to the context vector; at the second time step, it considers both the context vector and the torque value generated in the last step to generate the current torque prediction value, and so on, until the complete torque sequence covering the entire prediction shift time is generated. Finally, this sequence composed of a series of torque prediction values is the final output - the predicted shift disturbance torque time series curve. In the specific implementation process, the entire encoding-interaction-decoding architecture can be an end-to-end trained overall neural network, and the decoder part can be one or more LSTM layers plus a fully connected layer (also known as a time-distributed fully connected layer), which is responsible for mapping the hidden state of each time step to a specific torque value.

[0073] Specifically, in step S400, the motor-end driver intention torque and the predicted shift disturbance torque time series curve are compensated and fused to obtain a compensated motor target torque time series curve. It should be understood that there are two kinds of torque demands in the shift process, which are of completely different natures but act on the power system at the same time: one is the intention torque derived from the driver's operation, which represents the normal driving power of the vehicle; the other is the unexpected disturbance torque generated by mechanical actions such as motor speed regulation and clutch switching. If not handled, the motor controller can only respond to the driver's intention and cannot foresee and counteract the upcoming internal disturbance, which will inevitably result in shift impact and driving discomfort. Therefore, in order to actively and prospectively eliminate the disturbance in its infancy before it occurs, in the technical solution of the present application, the motor-end driver intention torque and the predicted shift disturbance torque time series curve are further compensated and fused to generate a new, unified final motor control instruction that meets the driver's intention and actively offsets the internal disturbance. By compensating and fusing the predicted disturbance torque time series curve with the driver's intention torque, a compensation component equal in size and opposite in direction to the disturbance torque is reserved in advance in the actual output torque of the motor. In this way, when the physical disturbance actually occurs, the compensation component output by the motor can exactly offset it, so that the final torque transmitted to the wheel end can smoothly and uninterruptedly follow the original intention of the driver.

[0074] More specifically, in the embodiments of the present application, the motor end driver intention torque and the predicted shift disturbance torque time curve are compensated and fused to obtain a compensated motor target torque time curve, including: adding the predicted shift disturbance torque at each time point in the motor end driver intention torque and the predicted shift disturbance torque time curve to obtain the compensated motor target torque time curve.

[0075] Specifically, in step S500, the compensated motor target torque time curve is sent to the motor MCU, and the transmission controller executes the downshift instruction. It should be understood that the shift process is essentially a highly synchronized and precise process between the power source (motor) and the transmission mechanism (transmission). The motor is responsible for accurately adjusting its speed during shifting and actively suppressing disturbances, while the transmission is responsible for switching gears at the mechanical level at the right time. Both must be perfectly coordinated in time, and any disconnection of either party will cause an impact. Therefore, in order to achieve precise timing synchronization between power and transmission, in the technical solution of the present application, the compensated motor target torque time curve is further sent to the motor MCU, and the transmission controller executes the downshift instruction, aiming to issue two parallel but independent control instructions to the respective execution units under the same time reference. Sending the compensated target torque time curve to the motor MCU is to instruct the motor to abandon the conventional torque output mode that only responds to the driver's intention, and instead strictly follow a carefully designed complex torque trajectory that includes feedforward compensation, thereby actively managing power output at the source. At the same time, the downshift instruction is issued to the transmission controller, which is to trigger physical shift actions such as clutch disengagement and engagement, synchronizer engagement, etc. The synchronous issuance of these two instructions ensures that the soft regulation of the motor and the hard execution of the transmission are in step.

[0076] In summary, the electric vehicle transmission downshift control method according to the embodiments of the present application is illustrated, which collects system states such as vehicle speed and braking intention in real time, and analyzes the driver's expected braking torque. Then, an adaptive shift disturbance torque prediction model is introduced, which takes the planned speed and acceleration time series in the motor speed regulation process as input, and uses the powerful nonlinear fitting capability of the deep learning model to predict the disturbance torque time curve caused by speed change during the entire shift. Finally, the driver intention torque and the predicted disturbance torque time curve are feedforward compensated and fused to generate a smooth and accurate motor target torque instruction. In this way, the torque fluctuation during downshift is actively and accurately offset, effectively solving the problems of low control accuracy and poor adaptability of traditional methods, and improving the downshift smoothness and ride comfort of electric vehicles in various working conditions.

[0077] As described above, the data collaboration management platform consistent with embodiments of the present application can be implemented in various wireless terminals, such as a server having electric vehicle transmission kickdown control algorithm, etc. In one possible implementation, the data collaboration management platform consistent with embodiments of the present application can be integrated into a wireless terminal as a software module and / or a hardware module. For example, the data collaboration management platform can be a software module in the operating system of the wireless terminal, or can be an application developed for the wireless terminal; of course, the data collaboration management platform can also be one of the many hardware modules of the wireless terminal.

[0078] Embodiments of the present disclosure have been described above, with the understanding that these descriptions are exemplary only, and are not intended to be exhaustive or to limit the scope of the embodiments disclosed. Many modifications and variations are possible in light of the above teachings. It is contemplated that the use of the terms "including," "comprising," "having," "containing," or "including" and variations thereof, are intended to be equivalent to the term "consisting of" or "consisting of." The use of the term "about" is intended to allow for variations, such as would be expected by one of ordinary skill in the art, such as experimental error, measurement error, manufacturing tolerances, and the like. The use of the term "or" is intended to mean "and / or," unless otherwise indicated. The use of the term "based on" is intended to mean "based, at least in part, on" unless otherwise indicated. The use of the term "one" is intended to mean "one or more but not including more than one," unless otherwise indicated. The use of the term "another" is intended to mean "one or more but not including more than one," unless otherwise indicated. The use of the term "some" is intended to mean "one or more but not including more than one," unless otherwise indicated. The use of the term "plurality" is intended to mean "one or more than one," unless otherwise indicated. The use of the term "another" is intended to mean "one or more but not including more than one," unless otherwise indicated. The use of the term "some" is intended to mean "one or more but not including more than one," unless otherwise indicated. The use of the term "plurality" is intended to mean "one or more than one," unless otherwise indicated. The use of the term "another" is intended to mean "one or more but not including more than one," unless otherwise indicated. The use of the term "some" is intended to mean "one or more but not including more than one," unless otherwise indicated. The use of the term "plurality" is

Claims

1. A method of electric vehicle transmission kickdown control, the method comprising: The method comprises: in response to a downshift instruction, collecting a system state vector, the system state vector comprising vehicle speed, brake pedal position, current gear, battery state of charge, and transmission oil temperature, the downshift instruction comprising a target gear; querying a smooth brake torque from a brake MAP table indexed by vehicle speed and brake pedal position, and converting the smooth brake torque into a motor-end driver-intent torque based on a gear ratio of the current gear; based on a target speed difference and a predicted shift time, performing adaptive shift disturbance torque prediction on the system state vector to obtain a predicted shift disturbance torque time series curve; compensating and fusing the motor-end driver-intent torque and the predicted shift disturbance torque time series curve to obtain a compensated motor target torque time series curve; sending the compensated motor target torque time series curve to a motor MCU, while a transmission controller executes the downshift instruction; based on a target speed difference and a predicted shift time, performing adaptive shift disturbance torque prediction on the system state vector to obtain a predicted shift disturbance torque time series curve, comprising: calculating a current motor angular velocity based on vehicle speed, a gear ratio of the current gear, and wheel diameter; calculating a target motor angular velocity based on vehicle speed, a gear ratio of the target gear, and wheel diameter; calculating a target speed difference based on the current motor angular velocity and the target motor angular velocity; calculating the predicted shift time based on the target speed difference; inputting the current motor angular velocity, the target motor angular velocity, and the predicted shift time into a trajectory generator to obtain a time series of planned speed and a time series of planned acceleration; based on the time series of planned speed and the time series of planned acceleration, performing adaptive shift disturbance torque prediction to obtain the predicted shift disturbance torque time series curve; based on the time series of planned speed and the time series of planned acceleration, performing adaptive shift disturbance torque prediction to obtain the predicted shift disturbance torque time series curve, comprising: sequence-encoding the time series of planned speed and the time series of planned acceleration to obtain a planned speed time series implicit encoding vector and a planned acceleration time series implicit encoding vector; performing parameter fine-grained feature dense interaction on the planned speed time series implicit encoding vector and the planned acceleration time series implicit encoding vector to obtain a planned speed-planned acceleration fine-grained interaction implicit encoding vector; performing feature decoding on the planned speed-planned acceleration fine-grained interaction implicit encoding vector to obtain the predicted shift disturbance torque time series curve.

2. The electric vehicle transmission kickdown control method of claim 1, wherein, based on a gear ratio of the current gear, converting the smooth brake torque into a motor-end driver-intent torque, comprising: based on the gear ratio of the current gear, converting the smooth brake torque into the motor-end driver-intent torque by the following formula: ； wherein the transmission ratio of the current gear stage, is the smooth braking torque.

3. The electric vehicle transmission kickdown control method of claim 2, wherein, based on the target speed difference, calculating the predicted shift time, comprising: based on the current motor angular velocity and the target motor angular velocity, calculating a kinetic energy change amount required for speed change by the following formula: ； wherein, is the equivalent moment of inertia, is the target motor angular velocity, is the current motor angular velocity; dividing the kinetic energy change amount required for speed change by motor average power to obtain the predicted shift time.

4. The electric vehicle transmission kickdown control method of claim 3, wherein, The time sequence of the planning speed and the time sequence of the planning acceleration are sequentially coded to obtain a planning speed time sequence implicit coding vector and a planning acceleration time sequence implicit coding vector, including: passing the time sequence of the planning speed and the time sequence of the planning acceleration through an LSTM-based sequence encoder to obtain the planning speed time sequence implicit coding vector and the planning acceleration time sequence implicit coding vector.

5. The electric vehicle transmission kickdown control method of claim 4, wherein, The planning speed time sequence implicit coding vector and the planning acceleration time sequence implicit coding vector are subjected to parameter fine-grained feature dense interaction to obtain a planning speed-planning acceleration fine-grained interaction implicit coding vector, including: The planning speed time sequence implicit coding vector is subjected to local feature analysis type dimension unification to obtain a sequence distribution of planning speed local time sequence granularity coding vectors; The planning acceleration time sequence implicit coding vector is inserted into different positions of the sequence distribution of the planning speed local time sequence granularity coding vectors respectively, and then input into a bidirectional LSTM model-based interaction encoder to obtain a set of planning speed-planning acceleration time sequence cross-position interaction coding feature vectors; The set of planning speed-planning acceleration time sequence cross-position interaction coding feature vectors are subjected to dynamic aggregation to obtain the planning speed-planning acceleration fine-grained interaction implicit coding vector.

6. The electric vehicle transmission kickdown control method of claim 1, wherein, The motor-end driver intention torque and the predicted shift disturbance torque time sequence curve are compensated and fused to obtain a compensated motor target torque time sequence curve, including: The predicted shift disturbance torque at each time point in the motor-end driver intention torque and the predicted shift disturbance torque time sequence curve is added to obtain the compensated motor target torque time sequence curve.

Images