A control method and system for a three-in-one electric drive system for electric vehicles

Abstract

This invention relates to the field of electric vehicle control technology, and discloses a control method and system for a three-in-one electric drive system for electric vehicles. The method includes: acquiring vehicle data; identifying driving conditions based on the vehicle data; triggering a control prediction model according to the driving conditions; constructing a multi-objective cost function based on the vehicle data; calculating the target wheel-end torque and the required motor output power; and sending the target wheel-end torque and the required motor output power to the controller to generate control commands, thereby achieving coordinated execution of the three-in-one electric drive system. The method of this invention has a complete structure and a closed-loop control link, making it suitable for dynamic scheduling and control of highly integrated electric drive systems. It possesses comprehensive technical advantages such as strong real-time performance, high energy efficiency, and adaptive thermal management.

Description

Technical Field

[0001] This invention relates to the field of electric vehicle control technology, specifically to a control method and system for a three-in-one electric drive system for electric vehicles. Background Technology

[0002] As the power density of electric vehicles continues to rise, highly integrated three-in-one electric drive systems, which combine motors, electronic controls, and reducers, are gradually replacing separate solutions and becoming the mainstream. However, most existing mass-produced models use fixed Eco / Sport modes or simple lookup table PI control: on the one hand, they cannot identify complex driving scenarios in real time and dynamically adjust torque-power mapping; on the other hand, they only passively limit current when the device temperature reaches a threshold, leading to problems such as high energy consumption, sudden drop in power output, and localized overheating.

[0003] In addition, the electro-thermal-mechanical coupling of the motor, inverter and gearbox exhibits strong nonlinear characteristics across time scales; traditional single time domain or hierarchical control is difficult to meet the driver's instantaneous power demand at the millisecond level, while also proactively constraining temperature rise safety within the second level, and lacks the ability to continuously adapt to vehicle aging and environmental changes. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a control method for a three-in-one electric drive system for electric vehicles, comprising: acquiring vehicle data and identifying driving conditions based on the vehicle data;

[0006] Based on the driving condition trigger control prediction model, a multi-objective cost function is constructed based on vehicle data to calculate the wheel-end target torque and the required motor output power.

[0007] The target torque at the wheel end and the required motor output power are sent to the controller to generate control commands, enabling the coordinated execution of the three-in-one electric drive.

[0008] As a preferred embodiment of the electric vehicle three-in-one electric drive system control method described in this invention, the vehicle data includes data on the power pedal, brake master cylinder, steering wheel, vehicle inertia, wheel speed, road longitudinal slope, battery state of charge, and ambient temperature.

[0009] As a preferred embodiment of the electric vehicle three-in-one electric drive system control method described in this invention, the method for identifying driving conditions includes stacking vehicle data in chronological order to form a three-dimensional tensor. ,Will The input is a three-layer one-dimensional convolution for feature extraction. The convolution output is input into a single-layer gated recurrent unit network. The driving condition probability distribution is obtained through Softmax transformation. The driving condition and confidence level corresponding to the maximum component are output.

[0010] The probability distribution of driving conditions corresponds to six types of driving conditions, including constant speed in the city, stop-and-go driving, strong acceleration, high-speed cruising, uphill and downhill driving.

[0011] As a preferred embodiment of the electric vehicle three-in-one electric drive system control method of the present invention, wherein: the step of triggering the control prediction model according to the driving condition includes using the driving condition, confidence level and vehicle data as inputs to the control prediction model;

[0012] The control prediction model is a dual-time-domain prediction model, which is divided into a fast time domain and a slow time domain. The slow time domain is performed first. At the beginning of the current control cycle, the slow time domain uses a 4-node RC thermal network to make rolling predictions of the temperature evolution in the next 1 second. The highest predicted temperature and the limit temperature are extracted to calculate the temperature rise margin factor, and the energy consumption weight and temperature control weight are dynamically adjusted according to the temperature rise margin factor.

[0013] After obtaining the temperature rise margin factor, the temperature rise margin factor is mapped into wheel end torque, current, motor speed and PWM frequency in the fast time domain, and the optimal solution is obtained for pedal and braking transients.

[0014] The wheel-end torque, current, motor speed, and PWM frequency actually issued in the fast time domain are immediately fed back to the next prediction in the slow time domain, achieving synchronization of information between the two domains.

[0015] As a preferred embodiment of the control method for a three-in-one electric drive system for electric vehicles described in this invention, the slow time domain includes abstracting the main heat sources and heat dissipation paths of the three-in-one electric drive into four equivalent thermal nodes, namely the motor stator, inverter module, reducer gearbox, and radiator coupled to the environment. Based on vehicle data, the motor stator temperature, inverter module temperature, reducer gear oil temperature, and radiator outlet temperature are read to form a four-node temperature vector. ;

[0016] Calculate the four-node heat output based on the control commands issued in the previous fast domain cycle:

[0017] ;

[0018] in, This represents the column vector of four-node heating power at time k; Indicates the stator copper loss of the motor; Indicates inverter switching losses; This represents the mechanical friction and oil shear loss of the reducer; Indicates matrix transpose;

[0019] Treating each node as an isothermal block with heat capacity, and heat transfer between nodes only through conduction or forced convection, an equivalent resistive-capacitive network is established for the four nodes:

[0020] ;

[0021] in, Indicates the heat capacity of node i; Represents the instantaneous temperature of node i; express Regarding time The derivative; Indicates the node index; This represents the instantaneous temperature of node j; This represents the equivalent thermal resistance between node i and node j; This represents the self-heating power of node i;

[0022] Using forward Euler discretization with a fixed step size Δt = 20 ms, at time... Discretization yields:

[0023] ;

[0024] in, This represents the four-node temperature vector predicted up to the next time step; Represents a four-node temperature column vector at time k; Indicates the discrete sampling step size; Represents the diagonal matrix of heat capacity; Represents the four-node global admittance matrix; Represents the environmental coupling column vector; Indicates ambient temperature; This represents the k-th discrete time.

[0025] Take the maximum predicted value among the four node temperatures. Calculate the temperature rise margin factor:

[0026] ;

[0027] in, Indicates the temperature rise margin factor; Indicates the set limit temperature; This represents the maximum predicted value among the four node temperatures;

[0028] The weights of the cost function in the multi-objective cost function are adjusted in real time using the temperature rise margin factor, and the upper limits of wheel end torque, current and PWM frequency are reduced proportionally to establish a soft constraint upper bound table.

[0029] A multi-objective cost function is constructed using a weighted quadratic performance index, and the temperature rise margin factor, cost function weights, and soft constraint upper bound table are optimized and output.

[0030] As a preferred embodiment of the electric vehicle three-in-one electric drive system control method described in this invention, the fast time domain includes obtaining the time based on vehicle data. A small-signal linearized electromechanical coupling model is established based on the vehicle linear velocity, motor mechanical speed, and wheel-end output torque executed in the previous control cycle.

[0031] Based on the temperature rise margin factor and soft constraint upper bound table of the slow time domain output, a quadratic cost function is constructed. The weights in the quadratic cost function are taken from the weights of the cost function of the slow time domain output. The first step control quantity is obtained through the solver.

[0032] The initial control input is superimposed onto the current operating point to form the target current and target gear. Then, combined with the field weakening criterion, a d-axis current command is generated, which is then calculated into a voltage command via the current loop. The target PWM frequency is set as follows:

[0033] ;

[0034] in, Indicates the target PWM frequency; This indicates taking the minimum value; This indicates the upper limit of the PWM frequency for slow-time domain output; Indicates the reference switching frequency calibrated for the vehicle;

[0035] Calculate the target wheel-end torque and the required motor output power, and then issue the execution order.

[0036] A control system for a three-in-one electric drive system of an electric vehicle using any of the methods described in this invention, wherein: an identification module acquires vehicle data and identifies driving conditions based on the vehicle data;

[0037] The calculation module, based on the driving condition triggering control prediction model, constructs a multi-objective cost function based on vehicle data to calculate the target torque at the wheel end and the required motor output power;

[0038] The control module sends the target torque at the wheel end and the required motor output power to the controller, generates control commands, and realizes the coordinated execution of the three-in-one electric drive.

[0039] The beneficial effects of this invention are as follows: This invention constructs a dual-time-domain model predictive control framework that integrates multimodal real-time perception, rapid operating condition identification, and temperature rise margin regulation, achieving coordinated optimization of torque output and electro-thermal constraints under different driving conditions. The slow-time-domain prediction module, combined with a thermal network model, assesses the temperature rise trend of key nodes in advance and dynamically generates soft constraints and weighted scheduling factors; the fast-time-domain optimization module tracks driving needs at high frequency and generates the optimal current and shifting strategy within feasibility, achieving millisecond-level response and parallel control of thermal safety. The method of this invention has a complete structure and a closed-loop control link, making it suitable for dynamic scheduling control of highly integrated electric drive systems. It has comprehensive technical advantages such as strong real-time performance, high energy efficiency, and adaptive thermal management. Attached Figure Description

[0040] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 This is an overall flowchart of a control method for a three-in-one electric drive system for electric vehicles, provided as an embodiment of the present invention. Detailed Implementation

[0042] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0043] Example 1, referring to Figure 1 As an embodiment of the present invention, a control method for a three-in-one electric drive system for electric vehicles is provided, comprising:

[0044] S1: Acquire vehicle data and identify driving conditions based on the vehicle data.

[0045] Furthermore, multimodal real-time data acquisition is performed within the vehicle's central controller, synchronously acquiring 22 raw signals at a fixed 10ms cycle, including those related to the power pedal, brake master cylinder, steering wheel, vehicle inertia, wheel speed, road longitudinal slope, battery state of charge, and ambient temperature. These signals are provided by Hall effect displacement sensors, piezoelectric pressure transmitters, magnetic encoders, a six-axis IMU, wheel speed sensors, a GNSS barometric pressure fusion module, the battery management system, and external temperature sensors. The signals are uniformly triggered via TTCAN, LIN, and SENT vehicle buses and time-stamped using the IEEE-1588 protocol.

[0046] Specifically, the 22 raw signals are: power pedal opening, power pedal opening rate of change, brake master cylinder pressure, brake master cylinder pressure rate of change, steering wheel angle, steering wheel angular velocity, vehicle longitudinal acceleration, vehicle lateral acceleration, vehicle yaw rate, vehicle linear velocity, front left wheel speed, front right wheel speed, rear left wheel speed, rear right wheel speed, road longitudinal slope angle, battery state of charge (SOC), battery pack total current, battery pack total voltage, maximum cell temperature, ambient temperature, vehicle roll rate, and vehicle pitch rate.

[0047] The acquired signals are first processed locally using second-order Butterworth denoising and five-point differential derivative calculations. Then, they are normalized according to the range constant in the EEPROM to form a standardized vector. To ensure data integrity, this invention employs a dual-channel redundancy design and implements CRC-16 verification for key signals, providing immediate compensation and fault marking for missing or out-of-bounds sampling. Finally, the processed vector is written to a shared memory area for subsequent condition identification steps, while the original data is retained in a circular buffer for 2 seconds to support algorithm self-learning and fault tracing.

[0048] Furthermore, within a 10ms control cycle, the Vehicle Central Controller (VCU) automatically reads the most recent 1-second (50 frames) normalized data vector sequence from the circular buffer, stacks them into a three-dimensional tensor in chronological order, and inputs it into a pre-built lightweight convolutional-recurrent hybrid neural network model. This model first extracts local feature changes of each signal on a short timescale using a three-layer one-dimensional convolutional network, then uses a gated recurrent unit (GRU) to aggregate and model the global temporal dependencies across signals within 1 second, ultimately outputting a log-probability vector of length six. This vector is then transformed using Softmax to obtain probability distributions representing six driving conditions (urban constant speed, stop-and-go, strong acceleration, high-speed cruising, uphill and downhill). The GRU model can use a publicly available gated recurrent unit model; its parameters, after quantization and pruning, can achieve real-time inference within an automotive-grade MCU with less than 3kB of storage space, without requiring modification to the GRU structure itself.

[0049] It should be noted that the six driving conditions can be mapped to core segments of current mainstream driving cycles (WLTP, NEDC, US standard FTP-75, etc.), accounting for the vast majority of the vehicle's historical statistical distribution. Discretizing complex continuous driving conditions into six typical scenarios can significantly reduce the complexity and computational load of the driving condition identification model, meeting the real-time inference requirements of automotive-grade MCUs. Through the above classification, this invention can dynamically adjust the weights and constraints in the model predictive control cost function based on differences in four dimensions: power demand curve, energy flow, thermal load, and mechanical transmission characteristics. This enables optimal efficiency, rapid response, and thermally safe electric drive cooperative control in various complex scenarios.

[0050] To ensure the reliability of the recognition results, this invention preferably uses the maximum probability value as the operating condition confidence level. When the confidence level is not lower than 0.50, the corresponding operating condition label is directly output along with the confidence level. If the confidence level is lower than 0.50, it temporarily reverts to the default "standard operating condition" label and fixes the confidence level at 0.50 to maintain the continuity and stability of the control strategy. The inference latency of the entire recognition process is strictly limited to within 3ms. The recognition result, along with the timestamp, is encapsulated into a structure and written to the VCU shared memory area for asynchronous reading by the next model prediction control module within 1ms. At the same time, the original tensor and recognition result are retained for 2s to support subsequent incremental learning and fault tracing.

[0051] S2: Based on the driving condition trigger control prediction model, a multi-objective cost function is constructed based on vehicle data to calculate the target torque at the wheel end and the required motor output power.

[0052] Furthermore, the vehicle's central controller first aggregates all key state information in a single control cycle to form a complete input vector. This vector not only includes the driving condition labels and their confidence levels from the previous step, but also integrates vehicle speed, motor speed, reference torque obtained from pedal mapping, road longitudinal slope, battery state of charge and charging / discharging current, as well as ambient temperature and the real-time temperatures of the three major heat sources of the electric drive (motor stator, inverter power module, and reducer gearbox).

[0053] Based on the complete input vector, this invention constructs a dual-time-domain model predictive control framework to meet the dual requirements of instantaneous dynamic response and second-level thermal-energy look-ahead. The framework consists of two levels of predictors, one in the fast domain and one in the slow domain. They operate sequentially in the same control loop, feeding each other's data, and ultimately output a set of coherent and consistent target control quantities.

[0054] The slow-domain predictor is activated first. This predictor uses a 1-second (50-step) rolling window with a step size of 20ms. Internally, it embeds a simplified automotive-grade four-node RC linear thermal network to predict the temperature evolution trajectory of the motor, inverter, gearbox, and radiator within the next second. During the calculation, the slow-domain predictor maps the gear position, PWM frequency, and current-torque commands from the fast-domain predictor to the node heating power. After prediction, the controller extracts the peak node temperature and calculates the temperature rise margin factor together with the device's extreme temperature and the ambient temperature. This factor, ranging from 0 to 1, visually represents the size of the heat dissipation margin in the next second: a margin close to 0 indicates that the temperature will reach its peak, requiring immediate power reduction; a margin close to 1 indicates extremely low thermal risk, allowing for a more relaxed power output. The heat dissipation margin refers to the difference between the maximum heat power that can be removed and the actual heat dissipation power under the current environmental and cooling conditions. Subsequently, the slow domain injects the temperature rise margin factor into its multi-objective cost function, continuously schedules the energy consumption weight, temperature control weight, and soft current limiting threshold, and simultaneously provides the gear switching and PWM frequency limiting strategy for the next cycle. At this point, the slow domain has completed a proactive balance of the "energy-temperature rise" coupling contradiction and pushes the calculated temperature rise margin factor, updated weights, and upper limit parameters to the fast domain in real time.

[0055] Specifically, the main heat sources and heat dissipation paths of the three-in-one electric drive are abstracted into four equivalent thermal nodes: the motor stator, the inverter module, the gearbox, and the radiator coupled to the environment. Based on vehicle data, the motor stator temperature, inverter module temperature, gearbox oil temperature, and radiator outlet temperature are read to form a four-node temperature vector. .

[0056] Calculate the four-node heat output based on the control commands issued in the previous fast domain cycle:

[0057] ;

[0058] in, This represents the column vector of four-node heating power at time k; Indicates the stator copper loss of the motor; Indicates inverter switching losses; This represents the mechanical friction and oil shear loss of the reducer; This represents the matrix transpose. There are already mature methods for calculating motor copper losses, inverter switching losses, and gearbox losses, which will not be discussed in detail in this invention. They can be calculated using Joule's law and the approximate linear relationship between the switching energy and frequency of IGBT / SiC devices.

[0059] Treating each node as an isothermal block with heat capacity, and heat transfer between nodes only through conduction or forced convection, an equivalent resistive-capacitive network is established for the four nodes:

[0060] ;

[0061] in, Indicates the heat capacity of node i; Represents the instantaneous temperature of node i; express Regarding time The derivative; Indicates the node index; This represents the instantaneous temperature of node j; This represents the equivalent thermal resistance between node i and node j; This represents the self-heating power of node i. The self-heating of the heat sink node is zero, while convective heat dissipation can be equivalent to the thermal resistance between the heat sink and the ambient temperature.

[0062] To achieve millisecond-level rolling prediction on automotive-grade MCUs, forward Euler discretization with a fixed step size Δt = 20 ms is used. Discretization yields:

[0063] ;

[0064] in, This represents the four-node temperature vector predicted up to the next time step; Represents a four-node temperature column vector at time k; Indicates the discrete sampling step size; Represents the diagonal matrix of heat capacity; Represents the four-node global admittance matrix; Represents the environmental coupling column vector; Indicates ambient temperature; Let k represent the k-th discrete time. By pre-combining the constant terms, it can be written in standard linear discrete state-space form:

[0065] ;

[0066] in, Represents the state transition matrix; Represents the heat generation power input matrix; This represents the ambient temperature input matrix. Since A, B, and E depend only on... , The constants Δt and Δt can be calculated offline during the calibration phase and then fixed as lookup table constants. During runtime, each step only requires one 4×4 matrix-vector multiplication operation (approximately 40 MAC instructions), which can complete a single-step recursion in less than 0.05ms. After multiple iterations, the temperature trajectory for the next 1 second can be obtained, from which the predicted peak temperature is extracted to calculate the temperature rise margin factor λ.

[0067] To express the remaining heat dissipation margin before the limiting temperature using a single scalar, this invention normalizes the predicted peak value using the device's ultimate safe temperature (including the automotive manufacturer's safety margin) as the upper bound and the current ambient temperature as the lower bound, and takes the maximum predicted value among the four node temperatures. Calculate the temperature rise margin factor:

[0068] ;

[0069] in, Indicates the temperature rise margin factor; Indicates the set limit temperature; This represents the maximum predicted value among the four-node temperatures. =1 indicates that there is almost no temperature rise or pressure within the next 1 second; when When the temperature approaches 0, it indicates that the temperature is about to reach its peak, and the power output needs to be reduced immediately. Thermal safety information can be transmitted using a single fixed-point number, which greatly reduces the amount of data exchange and model dimensionality between the fast and slow domains.

[0070] The temperature rise margin factor is injected in real time into the slow-time-domain cost function and constraint upper limit, enabling continuous adjustment of energy consumption, temperature control, and power balance with the thermal margin. The thermal margin refers to the remaining space between the current or predicted temperature and the allowable limit temperature. The controller employs a linear scheduling law:

[0071] ;

[0072] in, Indicates energy consumption weight; Indicates the initial energy consumption weight; Indicates temperature control weight; This represents the initial temperature control weight. The larger the size, the more important energy consumption becomes. Larger dimensions place greater emphasis on temperature control. As thermal margin decreases, energy consumption becomes less important, while temperature control becomes more crucial. Simultaneously, the physical upper bounds of wheel-end torque, current, and PWM frequency are also softly contracted by the same factor, establishing a soft-constraint upper bound table. When the current is reduced, the upper limit is reduced proportionally to ensure that the power is automatically reduced when the temperature approaches the limit without any abrupt hard current limiting.

[0073] After reconstructing the weights and soft constraints, a multi-objective cost function is constructed in the slow time domain using a weighted quadratic performance index:

[0074] ;

[0075] in, This represents the total cost function of slow-time-domain rolling predictive control; Indicates the prediction step size index; , , , The weighting coefficients of the cost function are scheduled online with the temperature rise margin factor. Indicates the first Stepper motor outputs mechanical power; Indicates the first Predict wheel end torque step by step; The reference torque representing the synchronization step size; Indicates the first Predict the stator temperature of the motor step by step; Indicates the safe temperature threshold for the motor; This represents the difference between the predicted torques of two consecutive steps. The constraint set consists of the upper bound of the soft-contracted constraints and the hard temperature constraints. The hard temperature constraints are derived from component specifications and vehicle reliability calibration, and are fixed in the controller in the form of calibration tables.

[0076] Since the thermal network has been explicitly iterated in the previous step, the optimization problem only involves linearized dynamics and a convex quadratic objective. The controller adopts the gradient-projection iterative (GP-MPC) algorithm: using the optimal solution of the previous cycle as the initial value, it can converge to the current local optimum within 0.3ms in 6-8 iterations, and output the total control quantity. It should be noted that this total control quantity does not directly drive the actuator. It is equivalent to a slow, global scheduling decision—informing the fast domain in advance of the available thermal margin, torque upper bound, gear tendency, and weight coefficients for the current cycle, so that the fast domain always falls within the safe and feasible region when refining control at the millisecond level, and remains consistent with the thermal management objective.

[0077] The optimization results are immediately transmitted via the real-time bus: target wheel-end torque, target motor power, gear shifting command, and target PWM switching frequency. Within the same control cycle, the temperature rise margin factor, the updated weighted reassembly, and the soft constraint upper bound table are written to shared memory for direct access by the fast domain in the next 10ms loop; while the current, gear, and frequency actually executed by the fast domain are transmitted back to the slow domain for the next round of heat generation power estimation, achieving closed-loop consistency of thermal-electrical-mechanical information. This cycle repeats continuously, allowing the vehicle to respond to the driver's power demands in milliseconds while maintaining forward-looking control of the temperature rise risk and energy consumption in the next second, fundamentally solving the technical challenge of the incompatibility between power, energy consumption, and temperature control in traditional fixed strategies.

[0078] Furthermore, the fast domain predictor immediately takes over the control law solution task upon obtaining the latest temperature rise margin factor. The fast domain uses a 200ms (10-step) window, still rolling in 20ms steps, but its internal dynamics retain only a linearized motor-vehicle transmission model, ignoring the slow thermal network, thus significantly reducing the state dimension. The controller first softly contracts the allowable torque, current, and PWM frequency upper bounds according to the temperature rise margin factor: the smaller the margin, the more the upper bound contracts, ensuring that instantaneous commands are always thermally safe. Subsequently, the fast domain constructs a quadratic cost function centered on accurate reference torque tracking and torque change rate suppression, and calls a fast quadratic programming solver based on hard real-time multiplier iteration. Within less than 0.6ms, it obtains the target wheel-end torque, target motor power, gear shifting command, and target PWM switching frequency value to be executed in the current control cycle. This result is immediately sent to the actuators such as the motor controller, inverter, and two-speed reducer, driving the vehicle to output power as planned.

[0079] Specifically, based on vehicle data, the time is obtained. A small-signal linearized electromechanical coupling model is established based on the vehicle linear speed, motor mechanical speed, and wheel-end output torque executed in the previous control cycle.

[0080] Based on the temperature rise margin factor of the slow time domain output and the soft constraint upper bound table, a quadratic cost function is constructed, expressed as:

[0081] ;

[0082] in, This represents the total scalar value of the fast time-domain objective function; This represents the weighting coefficient for instantaneous torque error; Indicates the weight of the torque change rate; Indicates the first Predict the wheel-end torque step by step. It should be noted that... The torque error weights are represented in both the fast and slow time domain cost functions, but in the slow time domain they represent the torque error weights within the control cycle, while in the fast time domain they represent the instantaneous torque error weights. β controls the torque error weights, and η limits the torque rate of change. Both are automatically scaled by the weight vector provided in the slow time domain to ensure a smoother output when the thermal margin is insufficient.

[0083] Since both the system model and the objective function are linear quadratic forms, the controller employs a pre-compiled real-time fast quadratic programming (RT-FQP) solver. RT-FQP uses the previous optimal solution as a warm start in each cycle, requiring only 3-5 iterations to satisfy the KKT conditions; for 10×2 control variables, the fixed-point arithmetic computation is less than 400 MAC, and the actual measured solution time is 0.55ms, fully integrated into the 10ms control cycle.

[0084] The initial control input is obtained through the solver and superimposed onto the current operating point to form the target current and target gear. Then, combined with the field weakening criterion, a d-axis current command is generated, which is then calculated into a voltage command via the current loop. The target PWM frequency is set as follows:

[0085] ;

[0086] in, Indicates the target PWM frequency; This indicates taking the minimum value; This indicates the upper limit of the PWM frequency for slow-time domain output; This indicates the reference switching frequency calibrated for the vehicle.

[0087] In the fast time-domain control cycle, after receiving current commands from the d-axis and q-axis, the driver first adjusts the inverter through a closed-loop current loop (i.e., traditional field-oriented control, FOC) to ensure that the three-phase stator current follows the target value in real time. Simultaneously, the controller, using the known number of motor pole pairs, excitation flux constant, and current q-axis current feedback, calculates the intermediate quantity—the electromagnetic torque on the motor side—in real-time in the software, following the standard torque conversion rules for permanent magnet synchronous motors. This electromagnetic torque is then combined with the gear ratio and efficiency of the reducer to convert it into wheel-end output torque, which is recorded as the target wheel-end torque for this cycle.

[0088] Estimating motor power is more straightforward. The inverter's DC bus port measures the bus voltage and current within the same control cycle. Multiplying these two values ​​and performing a short-time moving average yields the electric drive input power. Since the fast time-domain window only covers 200 milliseconds, the motor mechanical power and DC-side power are approximately conserved within this short timescale. Therefore, the controller uses this power as the target motor power for the current cycle.

[0089] After obtaining the target wheel-end torque and the target motor power, the controller writes the current gear command and the target PWM switching frequency into the execution buffer. At the end of the control cycle, the hardware layer directly sends the data to the inverter and motor drive unit.

[0090] S3: Sends the target torque at the wheel end and the required motor output power to the controller to generate control commands and realize the coordinated execution of the three-in-one electric drive.

[0091] After the fast time-domain control cycle completes its calculations, the central controller has generated the target wheel-end torque, target motor power, target current command, inverter PWM target switching frequency, and the next gear command. The three-in-one electric drive system then collaboratively completes the execution in the following order at the end of the same cycle.

[0092] To ensure internal thermal balance of the three-in-one unit, the temperature samples of the motor, inverter, and reducer are sent back to the central controller at 20ms cycles during the execution phase. If the temperature of any node shows a tendency to exceed the threshold, the controller immediately calls the soft constraint upper limit table to further reduce the upper limit of current, upper limit of torque, and PWM frequency for the next cycle, and increases the pump speed or fan speed on the heat sink side to suppress thermal shock before it occurs.

[0093] All execution results—including actual current, activated gear positions, measured inverter switching frequency, and latest temperatures at the three nodes—are written to the shared storage area in the form of event frames. This provides accurate initial values ​​for the next round of rapid time-domain prediction and also enters the slow time-domain heat generation power estimation module, ensuring complete consistency between the thermal model and physical execution. This process repeats continuously, with the motor, electronic control system, and reducer maintaining synchronized operation on a millisecond timescale. This satisfies the driver's immediate power requirements while continuously keeping the system operating within safe and efficient temperature ranges, achieving the overall real-time, adaptive, and electro-thermal-mechanical integrated control objective of this invention.

[0094] Example 2, in an exemplary embodiment, also provides a three-in-one electric drive system control system for electric vehicles, including an identification module for acquiring vehicle data and identifying driving conditions based on the vehicle data.

[0095] The calculation module triggers the control prediction model based on driving conditions, constructs a multi-objective cost function based on vehicle data, and calculates the target torque at the wheel end and the required motor output power.

[0096] The control module sends the target torque at the wheel end and the required motor output power to the controller, generates control commands, and realizes the coordinated execution of the three-in-one electric drive.

[0097] If the above functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0098] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0099] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0100] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0101] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A control method for a three-in-one electric drive system for electric vehicles, characterized in that, include: Acquire vehicle data and identify driving conditions based on the vehicle data; Based on the driving condition trigger control prediction model, a multi-objective cost function is constructed based on vehicle data to calculate the wheel-end target torque and the required motor output power. The target torque at the wheel end and the required motor output power are sent to the controller to generate control commands, enabling the coordinated execution of the three-in-one electric drive. The vehicle data includes data on the power pedal, brake master cylinder, steering wheel, vehicle inertia, wheel speed, road longitudinal slope, battery state of charge, and ambient temperature. The identification of driving conditions includes stacking vehicle data in chronological order to form a three-dimensional tensor. ,Will The input is a three-layer one-dimensional convolution for feature extraction. The convolution output is input into a single-layer gated recurrent unit network. The driving condition probability distribution is obtained through Softmax transformation. The driving condition and confidence level corresponding to the maximum component are output. The probability distribution of driving conditions corresponds to six types of driving conditions, including urban constant speed, stop-and-go driving, strong acceleration, high-speed cruising, uphill and downhill driving. The method of triggering the control prediction model based on driving conditions includes using driving conditions, confidence levels, and vehicle data as inputs to the control prediction model. The control prediction model is a dual-time-domain prediction model, which is divided into a fast time domain and a slow time domain. The slow time domain is performed first. At the beginning of the current control cycle, the slow time domain uses a 4-node RC thermal network to make rolling predictions of the temperature evolution in the next 1 second. The highest predicted temperature and the limit temperature are extracted to calculate the temperature rise margin factor, and the energy consumption weight and temperature control weight are dynamically adjusted according to the temperature rise margin factor. After obtaining the temperature rise margin factor, the temperature rise margin factor is mapped into wheel end torque, current, motor speed and PWM frequency in the fast time domain. At the same time, the optimal solution is obtained for pedal and brake transients, and control commands are issued. The wheel-end torque, current, motor speed, and PWM frequency actually issued in the fast time domain are immediately fed back to the next prediction in the slow time domain, achieving synchronization of information between the two domains.

2. The control method for a three-in-one electric drive system for electric vehicles as described in claim 1, characterized in that: The slow time domain includes abstracting the main heat sources and heat dissipation paths of the three-in-one electric drive into four equivalent thermal nodes: the motor stator, the inverter module, the gearbox, and the radiator coupled to the environment. Based on vehicle data, the motor stator temperature, inverter module temperature, gearbox oil temperature, and radiator outlet temperature are read to form a four-node temperature vector. ; Calculate the four-node heat output based on the control commands issued in the previous fast domain cycle: ； in, This represents the column vector of four-node heating power at time k; Indicates the stator copper loss of the motor; Indicates inverter switching losses; This represents the mechanical friction and oil shear loss of the reducer; Indicates matrix transpose; Treating each node as an isothermal block with heat capacity, and heat transfer between nodes only through conduction or forced convection, an equivalent resistive-capacitive network is established for the four nodes: ； in, Indicates the heat capacity of node i; Represents the instantaneous temperature of node i; express Regarding time The derivative; Indicates the node index; This represents the instantaneous temperature of node j; This represents the equivalent thermal resistance between node i and node j; This represents the self-heating power of node i; Using forward Euler discretization with a fixed step size Δt = 20 ms, at time... Discretization yields: ； in, This represents the four-node temperature vector predicted up to the next time step; Represents a four-node temperature column vector at time k; Indicates the discrete sampling step size; Represents the diagonal matrix of heat capacity; Represents the four-node global admittance matrix; Represents the environmental coupling column vector; Indicates ambient temperature; This represents the k-th discrete time. Take the maximum predicted value among the four node temperatures. Calculate the temperature rise margin factor: ； in, Indicates the temperature rise margin factor; Indicates the set limit temperature; This represents the maximum predicted value among the four node temperatures; The weights of the cost function in the multi-objective cost function are adjusted in real time using the temperature rise margin factor, and the upper limits of wheel end torque, current and PWM frequency are reduced proportionally to establish a soft constraint upper bound table. A multi-objective cost function is constructed using a weighted quadratic performance index, and the temperature rise margin factor, cost function weights, and soft constraint upper bound table are optimized and output.

3. The control method for a three-in-one electric drive system for electric vehicles as described in claim 2, characterized in that: The fast time domain includes obtaining the time based on vehicle data. A small-signal linearized electromechanical coupling model is established based on the vehicle linear velocity, motor mechanical speed, and wheel-end output torque executed in the previous control cycle. Based on the temperature rise margin factor and soft constraint upper bound table of the slow time domain output, a quadratic cost function is constructed. The weights in the quadratic cost function are taken from the weights of the cost function of the slow time domain output. The first step control quantity is obtained through the solver. The initial control input is superimposed onto the current operating point to form the target current and target gear. Then, combined with the field weakening criterion, a d-axis current command is generated, which is then calculated into a voltage command via the current loop. The target PWM frequency is set as follows: ； in, Indicates the target PWM frequency; This indicates taking the minimum value; This indicates the upper limit of the PWM frequency for slow-time domain output; Indicates the reference switching frequency calibrated for the vehicle; Calculate the target wheel-end torque and the required motor output power, and then issue the execution order.

4. A control system for a three-in-one electric drive system for an electric vehicle, applied to the control method for a three-in-one electric drive system for an electric vehicle according to any one of claims 1-3, characterized in that, include, The identification module acquires vehicle data and identifies driving conditions based on the vehicle data. The calculation module, based on the driving condition triggering control prediction model, constructs a multi-objective cost function based on vehicle data to calculate the target torque at the wheel end and the required motor output power; The control module sends the target torque at the wheel end and the required motor output power to the controller, generates control commands, and realizes the coordinated execution of the three-in-one electric drive.

Images