Brushless motor energy-saving control adjustment method and system, controller and storage medium

Abstract

The invention relates to the technical field of automation control, and discloses a brushless motor energy-saving control adjustment method and system, a controller and a storage medium, and the method comprises the following steps: S1, collecting the operation state information of a motor: collecting the operation state information, including current, voltage, temperature and load data, of the brushless motor in real time; s2, dynamic modeling and numerical solution: constructing a dynamic model of the brushless motor system based on the acquired running state information, and modeling by adopting a high-order differential equation set; s3, constructing a control cost function; s4, deducing an optimal control strategy; s5, approximately solving a control strategy; s6, adjusting and executing control parameters; and S7, performing feedback and parameter correction. By introducing the low-pass filter, the control change rate constraint and the rolling optimization mechanism, the dynamic closed-loop control effect of more stable control, more accurate feedback and more synchronous response is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of automation control, in particular to a brushless motor energy-saving control adjustment method, system, controller and storage medium. BACKGROUND

[0002] In the automation control scene, accurate and stable state feedback and control instructions are the core of system operation. Whether it is intelligent manufacturing, power regulation or unmanned device navigation, the control system must have fast response and anti-interference ability. Especially in dynamic environment or frequent disturbance, the original sensor signal is prone to fluctuation and noise interference. If it is not processed directly for control decision, the system stability and precision will be greatly reduced.

[0003] In the prior art, model predictive control (MPC) is widely used in multivariable system control scenarios due to its predictability and constraint handling ability. Some improved MPC has realized the optimization of state deviation and control amount, and has good trajectory tracking performance. Some schemes also compress the execution efficiency, which can adapt to small and medium-sized embedded platforms. On the problem of measurement noise, some researches introduce median filtering or moving average algorithm to improve the suppression ability of occasional abnormal points. The advantages of this kind of technology mainly focus on simple optimization structure, high calculation efficiency and strong adaptability to typical linear systems.

[0004] However, the existing method still has some shortcomings. First, the existing technology ignores the influence of the change rate of control input over time on the execution stability of the system. Although the controller converges, the execution instruction often changes suddenly, resulting in frequent switching of the actual device and reducing the service life. In addition, the traditional filtering method performs poorly when facing high-frequency narrow-band noise, and is prone to residual interference signals, affecting state estimation. Some prediction schemes based on fixed cycle execution also have delay accumulation, and the control output often lags behind the system change, forming "prediction that can't keep up". SUMMARY

[0005] In view of the shortcomings of the prior art, the present application provides a brushless motor energy-saving control adjustment method, system, controller and storage medium, which solves the problems of unstable control output, easily disturbed state feedback and reduced control precision caused by prediction lag.

[0006] To achieve the above purpose, the present application realizes the following technical scheme: a brushless motor energy-saving control adjustment method, comprising the following steps: S1, collecting the running state information of the motor: collecting the running state information of the brushless motor in real time, including current, voltage, temperature and load data; S2, dynamic modeling and numerical solution: based on the collected operating state information, a dynamic model of the brushless motor system is constructed, high-order differential equations are used for modeling, and the high-order differential equations are numerically solved by Runge-Kutta fourth-order method; S3, control cost function construction: define the control cost function, the cost function includes state error term, control input term and control smoothness term, and a linear quadratic regulation model is used for construction; S4, optimal control strategy derivation: based on the control cost function, the Euler-Lagrange variational method is used to derive the partial differential equation corresponding to the optimal control strategy; S5, approximate solution of control strategy: the partial differential equation is approximately solved by using deep neural network or support vector regression method to obtain the optimal control input; S6, control parameter adjustment and execution: the control parameters of the brushless motor are adjusted in real time according to the optimal control input, and the control execution is completed; S7, feedback and parameter correction: the feedback control execution result is obtained, and the control model is corrected according to the feedback information.

[0007] Preferably, the real-time acquisition of the operating state information of the brushless motor comprises: a current sensor for acquiring current data of the brushless motor; a voltage sensor for acquiring voltage data of the brushless motor; a temperature sensor for acquiring temperature data of the brushless motor; a torque sensor for acquiring load data of the brushless motor.

[0008] Preferably, the step of constructing the motor dynamic model comprises: establishing a first-order differential equation describing the change of motor current; establishing a second-order thermal balance equation describing the thermal characteristics of the motor; establishing a second-order dynamic equation reflecting the influence of load disturbance on motor speed; numerically solving the above equation set using Runge-Kutta fourth-order method to obtain discrete time state sequence.

[0009] Preferably, the step of defining the control cost function comprises: using a linear quadratic regulation model to construct the control cost function; The cost function includes: state error square weighted term; control input square weighted term; control change rate (derivative) square weighted term; The weight matrix is obtained by empirical parameter adjustment or genetic algorithm optimization.

[0010] Preferably, the step of deriving the control strategy by Euler-Lagrange variational method comprises: Taking the control cost function as the Lagrangian function form; Applying variation to the control input and state function within the time interval; Solving the corresponding Euler-Lagrange equation to obtain the partial differential control strategy expression; The derivation process is implemented in MATLAB using a symbolic derivative engine.

[0011] Preferably, the step of using a neural network or kernel space method for approximate solution comprises: Building a deep neural network model, with the input being the current state of the system and the output being the optimal control input; Or build a support vector regression model, using historical state-control pairs as training samples; The neural network is trained using the Adam optimizer, and the loss function is the residual of the cost function; Cross-validation is used during model training to avoid overfitting.

[0012] Preferably, the step of executing the feedback control result and dynamically adjusting the control model comprises: Compare the current control input with the obtained state feedback; Based on the error size and direction, use the online gradient descent algorithm to correct the modeling parameters in real time; Each update keeps a historical model snapshot and records it in the storage unit for rollback or training auxiliary network.

[0013] The present application also provides a brushless motor energy-saving control adjustment system, comprising: A state acquisition module for real-time acquisition of the operating state information of the brushless motor, including current, voltage, temperature and load data, with an acquisition frequency greater than 100Hz; A modeling and optimization module for constructing a motor dynamic model based on the acquired state information using the Runge-Kutta fourth-order method, defining a control cost function using linear quadratic regulation method, and deriving a control strategy using Euler-Lagrange variational method; A control output and execution module for generating optimal control input using DNN or SVR method and outputting control signals to the motor driver; A feedback learning and updating module for updating key parameters in the optimization model using online gradient descent algorithm based on feedback data to achieve closed-loop adaptive control.

[0014] The present application also includes a brushless motor energy-saving controller, the controller comprising: An input interface unit is configured to receive current, voltage, temperature and load data uploaded by the state acquisition module; A processing unit is configured to perform the following operations: A Runge-Kutta fourth-order method is used to establish a state space model of the brushless motor; A control cost function is constructed according to LQR control theory; The control strategy is solved based on Euler-Lagrange variational method; A deep neural network or a support vector regression model is used to fit the control input; An output interface unit is configured to send the control input generated by the processing unit to the motor actuator; A storage unit is configured to store state data, cost function model, network training weight, feedback information and parameter update history, for online learning and updating.

[0015] The application also includes a storage medium having a computer program stored thereon, which, when executed, runs the above-mentioned brushless motor energy-saving control adjustment method.

[0016] The application provides a brushless motor energy-saving control adjustment method, system, controller and storage medium. 1、The application introduces a first-order low-pass filter in the signal processing link, uses an exponential weighting method to denoise the original measurement data, and effectively suppresses high-frequency interference. Compared with the method in the prior art which does not perform preprocessing or only uses a mean sliding window, the filter structure achieves a better balance between response speed and filtering accuracy, and solves the problem of unstable state estimation.

[0017] 2、The application uses an improved MPC cost function design containing a control rate constraint term, which is equivalent to introducing a "smooth force" into the control input, and enhances the continuity and physical feasibility of system dynamic execution. Unlike the traditional MPC scheme which only considers the state and input deviation, this method alleviates the problem of frequent wear and tear of the actuator caused by the jump of the control instruction in the actual system.

[0018] 3、By increasing the weight matrix to quantize and punish the control input rate, the application introduces a fine energy consumption management mechanism in the execution layer. The traditional scheme usually ignores this factor, resulting in a sharp fluctuation of the control output and a significant power consumption peak. After introducing this design, the system runs more energy-efficiently and the power load is more uniform.

[0019] 4. This invention couples rolling optimization with a dynamic feedback mechanism, applying only the first control command of the current moment in each cycle, thus achieving real-time closed-loop coupling of control and prediction. Traditional MPC systems suffer from feedback delay and prediction error accumulation. With this scheme, control decisions and system responses are more closely synchronized, improving overall steady-state maintenance and disturbance response capabilities. Attached Figure Description

[0020] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a system structure diagram of the present invention; Figure 3 This is an architecture diagram of the controller of the present invention; Figure 4 This is a control diagram of the storage medium of the present invention. Detailed Implementation

[0021] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Please see the appendix Figure 1 This invention provides a brushless motor energy-saving control adjustment method, system, controller, and storage medium, including the following steps: S1. Collect motor operating status information: Collect brushless motor operating status information in real time, including current, voltage, temperature and load data; As the starting point of the entire control process, S1's technical implementation involves not only the acquisition of basic operating parameters but also strict requirements for details such as data processing, structural integration, and time synchronization. The quality of state acquisition directly affects the accuracy and stability of subsequent modeling and control strategies. Therefore, each sub-module has been designed in depth in the implementation details, supplemented by mathematical expressions and system compensation mechanisms to ensure data accuracy and system robustness.

[0023] In this step, the system needs to acquire multiple key physical variables of the motor in real time during dynamic operation to form a state vector for modeling. The acquisition accuracy and update frequency of this state information must meet the constraints of a continuous control system to ensure that the model equations can be established and solved based on high-quality input data.

[0024] In this embodiment, the operating status acquisition module is deployed between the motor controller and the drive axle. The acquired operating status data includes, but is not limited to, current, voltage, stator temperature, and output load torque. All status data are connected to the main control chip through a high-precision ADC interface, and after filtering, calibration, and synchronization processing, are uniformly output to the modeling module.

[0025] Typically, Hall effect current sensors are used to measure the instantaneous value of the current in a three-phase motor. The sampled three-phase current signal is denoted as I. a (t), I b (t), I c The equivalent current value I(t) is obtained by the following transformation: Among them: I a (t), I b (t), I c (t) represents the instantaneous value of the three-phase current, in amperes (A); I(t) represents the current effective equivalent current value of the motor, in amperes (A); Used to convert the peak root mean square value to the equivalent value.

[0026] The effective value approximation method used here can be used for subsequent heat loss estimation and model state variable definition.

[0027] As an alternative, in some embodiments, the third phase can be calculated using only the two-phase current (based on the balance assumption) to reduce the number of sensors required, but the calculation error needs to be compensated.

[0028] The system measures the bus voltage, which is then input to the ADC channel via a resistor divider. Due to rapid voltage fluctuations, the system incorporates a second-order low-pass filter, as follows: Where: H(s) represents the filter transfer function; ω c ζ is the cutoff frequency of the filter, in radians per second (rad / s); ζ is the damping coefficient, representing the trade-off between filter stability and response speed; s is the complex frequency domain variable (Laplace transform parameter), in radians per second (rad / s).

[0029] In a specific implementation, ω c =2π×30, which limits the voltage measurement signal frequency to within 30Hz, suitable for slow-changing load conditions.

[0030] Temperature acquisition points are deployed on the surface of the stator housing or directly attached to the pins of the main drive MOS transistor. The temperature signal is obtained by forming a bridge circuit between the thermistor (NTC) and the operational amplifier.

[0031] In some implementations, the system combines temperature measurement results with thermal model correction coefficients to improve subsequent thermal balance modeling and thermal protection strategies.

[0032] It is worth mentioning that, to prevent sampling bias caused by the hysteresis of the NTC thermal response, this system adds a first-order prediction model to the temperature signal, in the following form: in: The predicted temperature is used for subsequent temperature rise modeling; T(t)1 is the currently collected temperature in degrees Celsius (°C); τ is the prediction time constant, used to adjust the prediction speed and steady-state stability. The rate of temperature change is approximately obtained using the finite difference method.

[0033] This strategy effectively mitigates the impact of thermal inertia on real-time performance.

[0034] Load data is acquired via a miniature strain gauge mounted on the output shaft. The output voltage is sampled, differentially amplified, and filtered before being converted into the actual torque value L(t)1, in Newton-meters (N·m). The torque-voltage relationship is obtained through calibration and is typically approximated as linear. L(t)1=k torque ·V strain (t); Where: k torque V represents the sensor sensitivity coefficient, measured in N·m / V. strain (t) represents the actual output voltage of the strain gauge.

[0035] In scenarios involving frequent motor starts and stops or sudden load changes, the module's sampling frequency can be increased to 1kHz to handle transient characteristics. Under normal circumstances, a sampling frequency of 500Hz is sufficient for the vast majority of operating conditions.

[0036] To ensure consistent timing of data across all channels, this system employs a unified scheduling clock to trigger sampling and introduces a timestamp mechanism to ensure the consistency of the input state vector x(t).

[0037] In one possible implementation, each data stream caches the state of the current and previous two sampling points into a circular buffer: {x(t),x(t-Δt),x(t-2Δt)}; Where: Δt is the time interval between adjacent sampling periods, in seconds; x(t-Δt) is the state vector of the previous sampling period; x(t-2Δt) is the state vector of the sampling period before that; and x(t) is the current system state vector.

[0038] The cache depth is at least 3, which can be used for first-order difference calculation, slope fitting, or for subsequent sliding window prediction training of neural network models.

[0039] As a safeguard for system robustness, the system incorporates upper and lower limit judgments and sampling anomaly detection mechanisms for collected data. When it is detected that continuous sampling results remain statically unchanged, data jumps, or is lost, a fault-tolerant strategy will be triggered.

[0040] For example, if more than two channels return invalid values ​​(such as ADC saturation or voltage value sudden change) within any consecutive sampling period, the system will automatically trigger soft degradation control logic, suspend the output of the high-level control module, and only maintain the operation of the underlying steady-state control logic.

[0041] S1 not only achieves accurate perception of the core operating status of the brushless motor, but also provides high-quality real-time data support for subsequent modeling and optimization, effectively ensuring the operational stability and model repeatability of the overall energy-saving control system.

[0042] S2. Dynamic Modeling and Numerical Solution: Based on the collected operating state information, a dynamic model of the brushless motor system is constructed. This modeling employs a system of high-order differential equations, which are then numerically solved using the Runge-Kutta fourth-order method. S2, to achieve optimal derivation of the subsequent control strategy, requires state modeling of the motor itself, i.e., establishing the dynamic relationships between state variables. This modeling must not only conform to the physical operating laws of the motor but also possess numerical solvability and structural adaptability for control optimization and neural network training.

[0043] In this embodiment, the modeling approach adopts a "domain-specific modeling + joint solution" strategy. That is, the system is divided into three categories: electrical subsystem, thermal subsystem, and mechanical subsystem, and then uniformly transformed into a state-space representation, and discretely solved using numerical methods.

[0044] Differential equation model: in: R is the rate of change of current, representing the instantaneous rate of change of the current I(t) in the stator winding at time t; I(t) is the current, representing the instantaneous current flowing through the stator winding of the motor at time t; R is the resistance, representing the equivalent resistance of the stator winding of the motor; L is the inductance, representing the equivalent inductance of the stator winding of the motor; V(t) is the external voltage, representing the external input voltage applied across the stator winding of the motor at time t. This is a proportional term, representing the rate of current decay caused by resistance R; This is the voltage-dependent term, representing the direct contribution of the external input voltage to the change in winding current.

[0045] Its essence is a first-order linear differential equation, which reflects the causal response relationship between current and voltage, and constitutes the basic electrical equations for modeling.

[0046] The aforementioned thermal model adopts a second-order differential form with constant coefficients: Where: T(t) is the temperature of the motor stator or power device (such as a MOSFET) at time t, in degrees Celsius (°C); This is the first time derivative of temperature, i.e., the rate of temperature change, with units of °C / s; This is the second time derivative of temperature, i.e., the temperature acceleration, with units of °C / s². 2 α is the thermal damping coefficient, representing the degree of inertia in the temperature response, with units of 1 / s; β is the thermal restitution coefficient, reflecting the rate at which the system temperature recovers to the ambient temperature, with units of 1 / s. 2 ;T env Ambient temperature, a reference value for the actual external ambient temperature, in °C; η is the Joule heat conversion efficiency coefficient, representing the efficiency of converting current loss into heat energy, in °C / A. 2 I(t) represents the current effective equivalent current value of the motor, in amperes (A).

[0047] In this embodiment, a nonlinear thermal decay function and an air-cooled heat dissipation model are further introduced to improve the temperature modeling accuracy in high-power motor scenarios.

[0048] As an extended modeling approach, a wind-cooling function term can be added to the right-hand side of this equation to represent the change in environmental heat exchange capacity with motor speed: Q cool (t)=γ·ω(t) n ; Among them: Q cool (t) represents the air-cooled heat dissipation power in watts (W); γ represents the air-cooled heat dissipation efficiency coefficient in W·s. n / rad; ω(t) is the angular velocity of the motor, in radians per second (rad / s); n is the power parameter (empirical coefficient) of the effect of air cooling, which is a dimensionless constant.

[0049] After substituting it into the thermal model correction term, the complete form is updated as follows: Where: T(t) is the temperature of the motor stator or power device (such as a MOSFET) at time t, in degrees Celsius (°C); This is the first time derivative of temperature, i.e., the rate of temperature change, with units of °C / s; This is the second time derivative of temperature, i.e., the temperature acceleration, with units of °C / s². 2 α is the thermal damping coefficient, representing the degree of inertia in the temperature response, with units of 1 / s; β is the thermal restitution coefficient, reflecting the rate at which the system temperature recovers to the ambient temperature, with units of 1 / s. 2;T env Ambient temperature, a reference value for the actual external ambient temperature, in °C; η is the Joule heat conversion efficiency coefficient, representing the efficiency of converting current loss into heat energy, in °C / A. 2 I(t) represents the current effective equivalent current value of the motor, in amperes (A); γ is the air-cooling heat dissipation coefficient, representing the degree of influence of angular velocity on heat dissipation; ω(t) is the angular velocity of the motor rotor, in radians per second (rad / s); n is the power parameter (empirical coefficient) of the air-cooling effect, which is a dimensionless constant.

[0050] This thermodynamic model allows for the dynamic calculation of internal temperature rise trends using real-time measured current and rotational speed values, and enables power adjustment in conjunction with subsequent cost functions.

[0051] In the mechanical model section, the previously defined second-order rotational dynamics equations are as follows: Where J is the equivalent moment of inertia of the motor and load, reflecting its inertial resistance to angular acceleration, and the unit is kilogram-meter. 2 (kg·m 2 ); Angular acceleration, which is the second derivative of angular velocity, is expressed in rad / s. 2 B is the viscous friction coefficient of the system, with units of Newton-meter-second (N·m·s); This is the first derivative of the angular velocity, i.e., angular acceleration, with units of rad / s. 2 ;k T is the linear proportionality coefficient between current and output torque, with units of Newton-meter per ampere (N·m / A); I(t) represents the current effective equivalent current value of the motor, with units of amperes (A); L(t) is the load torque, the resistance torque applied to the motor shaft by the load or external mechanism, with units of Newton-meter (N·m).

[0052] To further enhance the model's dynamic responsiveness, this embodiment models the load disturbance as a time-varying function and introduces a disturbance term δ. L (t), rewritten in the following form: Where: δ L (t) represents the disturbance load term, which indicates sudden disturbance torque or non-modeled external force term that is not easily measured in the system, and the unit is Newton-meter (N·m).

[0053] In addition, in some high-speed control scenarios, a nonlinear drag term C·ω of angular velocity can also be introduced. 2 (t) represents the high-speed resistance of the rotating system, and is therefore corrected as follows: Where: C is the nonlinear aerodynamic drag coefficient, reflecting the square drag effect generated by the system under high-speed rotation, with units of N·m·s. 2 / rad 2 ;ω 2 (t) represents the square of the angular velocity, used to characterize the property that air resistance or fluid resistance increases with the square of the velocity, and its unit is (rad / s). 2 .

[0054] To enable model solving and subsequent controller embedding and deployment, this embodiment unifies the above differential equations into a state-space representation: Define the state variables as follows: Where: x(t) is the five-dimensional state vector of the system, which includes current, temperature, rate of temperature change, angular velocity and angular acceleration; u(t) is the control input, i.e. the bus terminal voltage, in volts.

[0055] The system dynamics are uniformly represented as: Where: x(t) is the five-dimensional state vector of the system, which includes current, temperature, rate of temperature change, angular velocity and angular acceleration; is the first-order time derivative of the state vector; f(·) represents the nonlinear state transition function of the system, derived from the above differential model; u(t) is the control input, i.e., the bus terminal voltage, in volts; t is the current time variable, in seconds.

[0056] To facilitate numerical solution and discretized control, and to achieve time discretization of the state equations, this embodiment adopts the fourth-order Runge-Kutta method, the discretized expression of which is detailed above and will not be repeated here.

[0057] In one possible implementation, if the system has modeling errors or significant external disturbance fluctuations, a "model correction term" can be further introduced, such as: x n+1 =x n +Δt1·f(x n ,u n )+w n ; Where: x n x is the state vector at the current discrete time point n; n+1 The state prediction value at the next discrete time point n+1; Δt1 is the discrete time step, in seconds, typically 5ms or 10ms; f(·) is the state transition function, obtained by discretization based on the continuous-time state differential model; u n This is the control input for the current time point, in volts; w nThe state perturbation vector represents the system error or modeling bias, and is generally assumed to be Gaussian white noise.

[0058] S2 not only constructs a complete electro-thermal-mechanical coupled system model of a brushless motor, but also achieves an inclusive expression of engineering uncertainties through structural extension and disturbance term design. The model has a clear mathematical form and physically measurable parameters, which facilitates numerical implementation using existing computing platforms, and provides a sufficient data and structural foundation for the subsequent construction and learning of optimal control strategies.

[0059] S3. Construction of Control Cost Function: Define the control cost function, which includes state error term, control input term and control smoothness term, and construct it using a linear quadratic adjustment model; S3 further constructs a cost function for optimized control to ensure a reasonable balance between energy consumption, system stability, and control smoothness in actual operation. The construction of the cost function needs to fully consider the complexity of the motor system, especially under conditions of load disturbance, temperature rise limitations, and changes in heat dissipation conditions. The optimization objective not only includes the traditional minimization of energy consumption but also needs to take into account the robustness of the control strategy.

[0060] Generally, to enable controllers to have strong generalization ability and practical adaptability, the cost function is usually in the form of a weighted quadratic form. The advantage of this form is that it is easy to solve using analytical or numerical optimization methods, while providing an accurate quantitative description of the state deviation and control input amplitude of multivariable systems.

[0061] In one possible implementation, this embodiment employs a control cost function of the following form: Where: J(u) is the control cost function, corresponding to the cumulative loss value over the entire control time interval, in dimensionless units; t0 is the start time; t f The termination time is in seconds; x(t) is the current system state vector; u(t) is the control input vector, which in this embodiment usually refers to the bus voltage V(t), in volts; Q is the state variable penalty matrix, used to measure the penalty intensity when the state variable deviates from the target value. It is a symmetric positive definite matrix, and its unit is the reciprocal of the square of the variable; R1 is the control input penalty matrix, used to limit the amplitude of the control input and prevent extreme control outputs. It is a symmetric positive definite matrix; λ is the smoothness penalty coefficient, used to suppress drastic changes in the control signal; The time derivative of the control input represents the rate of change of the control command, and its unit is volts per second (V / s).

[0062] In practical applications, the configuration of the weight matrices Q and R1 usually needs to be determined through experimental data or simulation analysis. Alternatively, the initial state can take the following form: Q=diag{q1,q2,q3,q4,q5}R1=diag{r1}; Where: q1 is the penalty weight for current deviation, in units of 1 / A. 2 q2 represents the penalty weight for temperature deviation, in units of 1 / ℃. 2 q3 represents the penalty weight for the rate of temperature change, with units of 1 / (°C / s). 2 q4 represents the penalty weight for angular velocity deviation, in units of 1 / (rad / s). 2 q5 is the penalty weight for angular acceleration, in units of 1 / (rad / s). 2 ) 2 r1 is the penalty weight for controlling the input amplitude, ensuring that the control command is within a reasonable range, with units of 1 / V. 2 .

[0063] The rate of change of the control input can be quite drastic under certain operating conditions, especially under sudden load changes or large external disturbances. Therefore, this embodiment introduces a smoothness penalty term: ||·|| is the Euclidean norm of the vector. In this formula, since the control input is a one-dimensional signal V(t), the norm degenerates into the square form of a scalar; u(t) is the control input vector of the system at time t; The derivative of the control input with time represents the rate of change of the control input signal; V(t) is the control voltage signal at time t, in volts. The square of the voltage change rate is a quantitative indicator representing the drastic fluctuations in the control input, with units of (V / s)².

[0064] This measure suppresses drastic changes in the control signal by adjusting the parameter λ. Under different operating conditions, a larger λ value suppresses rapid changes in the control signal, while a smaller λ value allows for more flexible control strategies.

[0065] In digital control systems, it is often necessary to convert the continuous-time cost function into a discrete form for solution using numerical optimization methods. One possible implementation is the discretized form as follows: Where: x k Let u be the state vector at time k. k The input is the control input at time k; N is the total number of control steps; u k+1 -u kTo control the amount of change in the input, it is used to approximate the derivative term in continuous time.

[0066] By using this form of cost function, this embodiment can not only effectively limit the system's operating state and control input, but also improve the stability of the control process and the system's lifespan.

[0067] S4. Derivation of the optimal control strategy: Based on the control cost function, the partial differential equations corresponding to the optimal control strategy are derived using the Euler-Lagrange variational method. In step S3 above, the system's performance indicators such as energy consumption, stability, and control smoothness have been quantified by constructing a reasonable cost function. Based on this, the core objective of step S4 is to perform optimization based on the cost function to determine the optimal control input sequence, ensuring that the motor system's performance meets expected requirements during actual operation.

[0068] Generally, directly solving for the optimal control input analytically is quite difficult for nonlinear, time-varying, and constrained motor systems. Therefore, this embodiment employs a numerical optimization method, combining state-space equations, cost functions, and constraints to construct a complete optimization problem. Within this optimization framework, the system iteratively finds the minimum value of the cost function while satisfying state and input constraints, thereby achieving optimal control.

[0069] As one possible approach, the optimization problem is typically solved within a prediction time window, with control inputs adjusted in real-time through rolling optimization. This method is particularly suitable for scenarios with external disturbances, uncertainties in system parameters, or changes in environmental conditions. By dynamically adjusting the state and input within the prediction window, the robustness and adaptability of the system can be effectively improved.

[0070] Specifically, the optimization problem in this embodiment can be expressed in the following form: The constraints include: State equation constraints: Input constraints: u min ≤u(t)≤u max ; State variable constraints: x min ≤x(t)≤x max ; Where: J(u) is the control cost function, corresponding to the cumulative loss value over the entire control time interval, in dimensionless units; t0 is the start time; t fThe termination time is in seconds; x(t) is the current system state vector; u(t) is the control input vector, which in this embodiment usually refers to the bus voltage V(t), in volts; Q is the state variable penalty matrix, used to measure the penalty intensity when the state variable deviates from the target value. It is a symmetric positive definite matrix, and its unit is the reciprocal of the square of the variable; R1 is the control input penalty matrix, used to limit the control input amplitude and prevent extreme control outputs. It is a symmetric positive definite matrix, and its unit is 1 / V2; λ is the smoothness penalty coefficient, used to suppress drastic changes in the control signal; The time derivative of the control input represents the rate of change of the control command, measured in volts per second (V / s); f(x(t),u(t),t) is the state equation, describing the dynamic evolution of the system, including factors such as system inertia, damping, and input signal; u min u max To control the upper and lower limits of the input, ensuring that the control signal remains within the capability range of the physical actuator. The unit is volts; x min x max These are the upper and lower limits of the state variable, used to ensure that the system state remains within a safe range. The unit depends on the actual physical quantity of the state variable.

[0071] In one possible implementation, this embodiment employs a direct optimization method for solving the problem. The direct optimization method parameterizes the control input, transforming the continuous optimization problem into a finite-dimensional nonlinear programming problem. The optimizer iteratively calculates and gradually reduces the value of the cost function until a preset convergence condition is met.

[0072] Generally, the direct optimization method includes the following steps: Initial condition settings: Initialize optimization variables based on the current state variables and environment information.

[0073] State prediction: Predicting future states using the system's state equations.

[0074] Cost function calculation: Calculate the value of the cost function based on the current predicted state and input.

[0075] Gradient calculation: The gradient of the cost function with respect to the control input is calculated using numerical methods.

[0076] Iterative update: Adjust the control input according to the gradient direction and perform optimization update.

[0077] Alternatively, this embodiment can also employ a model predictive control (MPC) approach. Within the MPC framework, each optimization only calculates the optimal input sequence for a future time period. The optimizer continuously updates the predictive model and cost function based on real-time state data, making the control strategy more robust.

[0078] Specifically, the following steps are performed within each sampling period: State measurement and estimation: Acquiring current system state information through sensors; Optimization solution: Solve for the current optimal control sequence within the prediction time window; Input execution: Apply control input only for the first step; Status update: The system status changes according to the actual physical response; Rolling update: Remove the first time step of the current forecast time window and roll forward one time step.

[0079] Through the above process, MPC can maintain high control accuracy and system stability in dynamic environments.

[0080] By using the S4 optimization solution method, the control system can effectively reduce the value of the cost function while satisfying physical constraints, thereby achieving energy consumption optimization, temperature control, and stability improvement during motor operation.

[0081] S5. Approximate solution of control strategy: Use deep neural networks or support vector regression methods to approximate the solution of partial differential equations and obtain the optimal control input; To achieve dynamic response in the closed-loop control system, step S5 focuses on closely integrating the optimized control strategy with system operation. Through real-time execution, feedback correction, and dynamic updates of the predictive model, a complete rolling implementation mechanism for model predictive control is constructed. This step is crucial for the implementation of the control strategy and the formation of closed-loop operating logic, and it is also the foundation for achieving system robustness, adaptability, and stability.

[0082] Generally, the implementation of model predictive control relies on a dynamic cycle of four stages: prediction, optimization, execution, and feedback. Step S5 mainly covers the "execution" and "feedback" parts. During control execution, the system directly applies the first control input from the previous cycle's optimization results to the motor drive system, and at the end of the current cycle, it collects system response information and updates the system state variables as the prediction starting point for the next cycle.

[0083] In one possible implementation, this embodiment introduces a feedback error compensation mechanism to improve the controller's resilience to external disturbances, model uncertainties, and sensor biases. This mechanism calculates the residual between the predicted state and the actual measured state to correct the system's initial state conditions in real time. The specific process can be formalized into the following two steps: Residual estimation formula: e(t) = x meas (t)-x pred (t); State correction formula: x corr (t)=x pred(t)+K e ·e(t); Where: e(t) is the prediction error residual vector at the current time t; x meas (t) represents the system state variable vector obtained through sensor measurements; x pred (t) represents the system state estimate at the current time, predicted by the controller's internal model; x corr (t) represents the state variable vector corrected based on the residuals, serving as the prediction starting point for the next control cycle; K e This is the residual correction gain matrix.

[0084] In some embodiments, the system may introduce a first-order low-pass filter to denoise the measured feedback signal, thereby suppressing the impact of high-frequency interference signals on control accuracy. The filter form is as follows: x filt (t)=α1·x meas (t)+(1-α1)·x filt (t-Δt2); Where: x filt (t) represents the filtered state variable; x meas (t) is the original state vector obtained from the sensor measurement at the current time t; x filt (t-Δt2) is the state vector output by the filter in the previous control cycle, i.e. the filtering result at the previous moment; α1 is the filter coefficient, ranging from (0, 1), usually set by experience, such as 0.2 to 0.5; Δt2 is the control sampling period, in seconds.

[0085] In the actual application of the control signal, to meet the input characteristics of the actual execution unit, this embodiment introduces a control signal conversion module to convert the voltage command V(t) in the optimization result into a specific control signal form, such as a PWM signal. This conversion relationship has been given in the previous description and will not be repeated here to avoid repetition.

[0086] As an alternative, when the system has influencing factors such as nonlinear dead zones, actuator saturation, or signal lag, this embodiment can use a controller feedforward compensation mechanism to predict and correct in advance, thereby reducing the deviation caused by actual response lag. For example, for a system with a known transmission delay τ... d The optimal control signal can be shifted forward by τ. d The duration is output, i.e.: u exec (t)=u * (t+τ d ); Where: u exec (t) represents the actual control signal executed; u * (t+τ d ) represents the optimal control quantity after delay compensation.

[0087] During control implementation, it is also necessary to ensure the precise and consistent operation cycle of the entire MPC structure to avoid instability in system control caused by state estimation delays, optimization calculation lags, or signal output jitter. To this end, this embodiment sets up a control synchronization manager to ensure that the following modules all complete within a unified control cycle Δt2: Status acquisition; Model prediction; Solving for optimal control; Control signal output; Status feedback updated.

[0088] The control period Δt2 is generally set in the range of 1ms to 10ms, depending on the system response speed and computing resources.

[0089] In summary, step S5 not only realizes the complete transformation process of the control signal from numerical optimization results to physical output, but also ensures the stability and responsiveness of the control system during dynamic operation through multiple sub-mechanisms such as residual correction, filtering and noise reduction, delay compensation, and synchronization scheduling. Through the above complete control closed-loop implementation scheme, the engineering application of this invention in highly dynamic, strongly coupled motor control systems can be effectively realized.

[0090] S6. Control Parameter Adjustment and Execution: Adjust the control parameters of the brushless motor in real time according to the optimal control input to complete the control execution; S6 further introduces a dynamic update and error adaptation mechanism for the system prediction model, thereby improving the self-adjustment capability of the entire closed-loop system to external disturbances, internal nonlinear changes, and switching of operating conditions.

[0091] As a crucial extension mechanism, step S6 primarily involves periodically updating, correcting, and optimizing the dynamic model used for prediction to better reflect the actual operating state of the system. Its goal is not only to improve the model's prediction accuracy but also to enhance its versatility and adaptability, providing a more accurate basis for subsequent rolling optimization.

[0092] In this embodiment, the correction mechanism of the prediction model is based on the error residual between the actual state measurement value and the model's predicted state value at the current moment, and the state is updated by constructing an augmented model. To avoid repetition, the basic state transition equation and residual calculation formula will not be described again here. The focus is on disclosing the correction and update model structure introduced in step S6, as follows: Where: x(k) is the system state vector at the current time k; x(k+1) is the predicted state variable at the next time; A is the state transition matrix, used to linearly map the active dynamic characteristics from the current state to the next state, and its unit is dimensionless; B is the control input matrix, reflecting the direct influence of control input on state changes; u(k) is the current control input vector, which usually includes bus voltage, current reference value, etc., and its unit is defined as V or A depending on the actual input. The state correction gain matrix is ​​used to adjust the influence of the prediction error feedback. It is usually a diagonal matrix with dimensionless units; r(k) is the residual vector; δ(k) is the system modeling error term or disturbance estimation term, used to reflect unmodeled nonlinear terms, load disturbances, or dynamic lags in the system. It can be estimated by statistical averaging or sliding window filtering of historical deviation residuals.

[0093] In general, the gain matrix It can be set to a constant diagonal matrix, or it can adaptively adjust based on historical residual trends. For example, if the residuals are consistently large, the system can automatically increase... The value of is used to enhance the feedback correction effect, and vice versa to prevent overcompensation. Specifically, this can be expressed as: Where: η1 is the gain adjustment rate coefficient, which controls the response speed, and the unit is dimensionless; γ2 is the set residual tolerance threshold.

[0094] In some embodiments, to further enhance model robustness, this embodiment combines the above-described update model with a Kalman filter structure, employing an extended Kalman filter to dynamically update the state estimate and covariance matrix. Kalman gain K f (k) is calculated as follows: K f (k)=P(k|k-1)H T [HP(k|k-1)H T +R3] -1 ; Where: P(k|k-1) is the state prediction covariance matrix, in units of the square of the state unit; H is the observation matrix, describing how the state variables are mapped to the observable output; R3 is the observation noise covariance matrix, in units of the square of the observation unit; K f (k) is the Kalman filter gain matrix at the current time.

[0095] This filter structure can effectively filter out the uncertainty caused by measurement noise and enhance the model's ability to estimate the actual system state.

[0096] As a possible extension, this embodiment can also introduce a disturbance observer to estimate the unmeasurable disturbance term δ(k) in the system, which can be used to further correct the state prediction process and enhance the self-compensation capability of the control system for model biases. Under this framework, the disturbance term can be observed as an additional state dimension and embedded in the state transition function.

[0097] To ensure the update mechanism operates correctly under real-time requirements, this embodiment establishes a dedicated model correction thread or module in the controller operating platform, which executes in parallel with the main control task. Data synchronization between modules is achieved through timestamps and state buffers to avoid the problem of asynchrony between predicted state data and actual control data.

[0098] S6 constructs a dynamically adjustable model self-updating mechanism by introducing multiple correction strategies, including state residual feedback, disturbance term modeling, Kalman filter estimation, and adaptive gain adjustment. This mechanism not only dynamically corrects the prediction model during operation and improves control accuracy, but also provides a key guarantee for the long-term stable operation of the controller, serving as an important supporting structure for the closed-loop robustness of the model predictive control system of this invention.

[0099] S7. Feedback and Parameter Correction: Obtain the execution results of feedback control and correct the parameters of the control model based on the feedback information; S7 constructs a rolling execution strategy based on model predictive control. This step is the core control structure for the continuous dynamic operation of the control system. Its function is to embed the aforementioned optimization and correction results into a rolling time window, realizing a closed-loop control process of multi-cycle iterative optimization and continuous control input and output.

[0100] Step S7 not only addresses the phased execution of optimization results but also encapsulates a time-logic approach to the dynamic closed loop between state estimation, model correction, and control input / output. Through systematic design of parameters such as control cycle, rolling prediction time domain, and optimization frequency, the control system achieves an adaptive rolling response to state changes.

[0101] In this embodiment, the core of step S7 lies in defining the rolling execution mechanism of predictive control. Its basic process is as follows: within each control cycle, only the first predicted control input is executed, and the remaining predicted values ​​are recalculated in the next cycle. Under this structure, the optimization problem is reconstructed and solved in each control cycle, thereby achieving a dynamic fusion of predictive control and real-time feedback.

[0102] To describe the control logic of this step in detail, the system in each cycle t k Perform the following optimization: Among them, the cost function J is used to comprehensively balance state error, control energy consumption, and input change rate, ensuring control smoothness and suppressing frequent modulation; x(t k +i|t k ) is at time t k The predicted system state variable for the i-th future step is given in state units; x ref (t k +i) represents the prediction time step t k +i is the reference state trajectory, used to guide the system to converge toward the target state; u(t) k +i|t k ) is at time t k The predicted control input for the i-th future step is expressed in control input units; Q4 is the state error weight matrix, typically a positive definite diagonal matrix, expressed as the reciprocal of the square of the state units; R4 is the control input weight matrix, expressed as the reciprocal of the square of the control input units; W is the rate of change penalty weight matrix; N2 is the prediction time-domain step size, a positive integer used to define the prediction time range, expressed as dimensionless. The weighted sum of squares controlling the rate of change of input represents the smoothness constraint term, specifically: t k This is the current control cycle time, in seconds, and the time index for each controller operation. To control the time derivative of the input, the unit is the input unit divided by seconds.

[0103] In one possible implementation, to further satisfy the dynamic stability of the controller and the constraints of the actuator, this embodiment also introduces input constraints and state boundary constraints, forming a complete optimization problem.

[0104] In each control cycle t k Within the cycle, the controller executes only the first optimal control input, and the remaining control inputs are discarded in the next cycle, and the optimization problem is resolved based on the latest state information.

[0105] To improve real-time performance, this embodiment adopts a fixed-time-step scrolling window structure, executing within each cycle: Condition measurement and correction; Model update; Optimize problem reconstruction and solution; Control input extraction and application; Scroll the window to the right and enter t k+1 cycle.

[0106] As a supplement, to prevent actuator jitter caused by excessively rapid control switching, this embodiment introduces a control smoothing buffer mechanism. If the rate of change of the current control input exceeds a preset threshold, the output control command is processed with a slow transition: S7 completes the closed-loop iterative design of model predictive control in real-time systems by introducing a constrained rolling optimization structure, state trajectory prediction, control smoothing processing, and dynamic execution mechanism.

[0107] The brushless motor energy-saving control and adjustment system described below can be referred to in correspondence with the brushless motor energy-saving control and adjustment method described above.

[0108] Please see the appendix Figure 2 The present invention also provides a brushless motor energy-saving control and adjustment system, comprising: The status acquisition module is used to collect the operating status information of the brushless motor in real time, including current, voltage, temperature and load data, with a acquisition frequency greater than 100Hz. The modeling and optimization module is used to construct a dynamic model of the motor based on the collected state information using the Runge-Kutta fourth-order method, define the control cost function using the linear quadratic adjustment method, and derive the control strategy using the Euler-Lagrange variational method; the control output and execution module uses the DNN or SVR method to generate the optimal control input and output the control signal to the motor driver. The feedback learning and update module updates and optimizes key parameters in the model based on feedback data using an online gradient descent algorithm to achieve closed-loop adaptive control.

[0109] The system in this embodiment can be used to execute the above method embodiments, and its principle and technical effect are similar, so they will not be described again here.

[0110] The brushless motor energy-saving controller described below and the brushless motor energy-saving control adjustment method described above can be referred to each other.

[0111] Please see the appendix Figure 3 The present invention also includes a brushless motor energy-saving controller, the controller comprising: The input interface unit is used to receive current, voltage, temperature and load data uploaded by the status acquisition module; The processing unit is configured to perform the following operations: The state-space model of the brushless motor is established using the Runge-Kutta fourth-order method; Construct the control cost function based on LQR control theory; Solving the control strategy based on the Euler–Lagrange variational method; Use a deep neural network or support vector regression model to fit the control input; The output interface unit is used to send the control input generated by the processing unit to the motor actuator. The storage unit is used to store state data, cost function model, network training weights, feedback information, and parameter update history for online learning and updating.

[0112] The controller in this embodiment can be used to execute the above method embodiments, and its principle and technical effect are similar, so it will not be described again here.

[0113] Please see the appendix Figure 4 The present invention also includes a storage medium storing a computer program thereon, which, when executed, runs the above-described brushless motor energy-saving control and adjustment method.

[0114] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A brushless motor energy-saving control and adjustment method, characterized in that, Includes the following steps: S1. Collect motor operating status information: Collect brushless motor operating status information in real time, including current, voltage, temperature and load data; S2. Dynamic Modeling and Numerical Solution: Based on the collected operating status information, a dynamic model of the brushless motor system is constructed. A high-order differential equation system is used for modeling, and the high-order differential equation system is numerically solved using the Runge-Kutta fourth-order method. S3. Construction of Control Cost Function: Define the control cost function, which includes state error term, control input term and control smoothness term, and construct it using a linear quadratic adjustment model; S4. Derivation of the optimal control strategy: Based on the control cost function, the partial differential equations corresponding to the optimal control strategy are derived using the Euler-Lagrange variational method. S5. Approximate solution of control strategy: Use deep neural networks or support vector regression methods to approximate the solution of partial differential equations and obtain the optimal control input; S6. Control Parameter Adjustment and Execution: Adjust the control parameters of the brushless motor in real time according to the optimal control input to complete the control execution; S7. Feedback and Parameter Correction: Obtain the execution results of feedback control and correct the parameters of the control model based on the feedback information.

2. The brushless motor energy-saving control and adjustment method according to claim 1, characterized in that, The real-time acquisition of the brushless motor's operating status information includes: A current sensor is used to collect current data from a brushless motor. Voltage sensor, used to collect voltage data of brushless motor; Temperature sensor used to collect temperature data of brushless motor; A torque sensor is used to collect load data from the brushless motor.

3. The brushless motor energy-saving control and adjustment method according to claim 1, characterized in that, The steps for constructing the dynamic model of the motor include: Establish a first-order differential equation describing the change in motor current; Establish a second-order thermal balance equation describing the thermal characteristics of the motor; Establish a second-order dynamic equation that reflects the effect of load disturbance on motor speed; The above equations are numerically solved using the Runge-Kutta fourth-order method to obtain the discrete-time state sequence.

4. The brushless motor energy-saving control and adjustment method according to claim 1, characterized in that, The steps for defining the control cost function include: The control cost function is constructed using a linear quadratic adjustment model; The cost function includes: a weighted term based on the squared state error; a weighted term based on the squared control input; and a weighted term based on the squared rate of change of control. The weight matrix is ​​obtained through empirical parameter tuning or optimization based on a genetic algorithm.

5. The brushless motor energy-saving control and adjustment method according to claim 1, characterized in that, The steps for deriving the control strategy using the Euler-Lagrange variational method include: The control cost function is in the form of a Lagrangian function; Apply variation to the control input and state function within the time interval; Solve the corresponding Euler–Lagrange equations to obtain the partial differential control strategy expression; The derivation process is implemented in MATLAB using the symbolic differentiation engine.

6. The brushless motor energy-saving control and adjustment method according to claim 1, characterized in that, The steps for approximate solution using neural networks or kernel space methods include: Construct a deep neural network model, with the current system state as the input and the optimal control input as the output; Alternatively, a support vector regression model can be constructed, using historical state-control pairs as training samples; The neural network is trained using the Adam optimizer, and the loss function is the residual of the cost function. Cross-validation is used during model training to avoid overfitting.

7. The brushless motor energy-saving control and adjustment method according to claim 1, characterized in that, The steps of feedback control execution results and dynamic adjustment of the control model include: Compare the current control input with the obtained state feedback; Based on the magnitude and direction of the error, the modeling parameters are corrected in real time using an online gradient descent algorithm. Each update retains a snapshot of the historical model and records it in the storage unit for rollback or training of the auxiliary network.

8. A brushless motor energy-saving control and adjustment system, characterized in that, The brushless motor energy-saving control adjustment method according to any one of claims 1-7 includes: The status acquisition module is used to collect the operating status information of the brushless motor in real time, including current, voltage, temperature and load data, with a acquisition frequency greater than 100Hz. The modeling and optimization module is used to construct a dynamic model of the motor based on the collected state information using the Runge-Kutta fourth-order method, define the control cost function using the linear quadratic adjustment method, and derive the control strategy using the Euler-Lagrange variational method. The control output and execution module uses DNN or SVR methods to generate the optimal control input and outputs the control signal to the motor driver. The feedback learning and update module updates and optimizes key parameters in the model based on feedback data using an online gradient descent algorithm to achieve closed-loop adaptive control.

9. A brushless motor energy-saving controller, characterized in that, Using the brushless motor energy-saving control adjustment method according to any one of claims 1-7, the controller includes: The input interface unit is used to receive current, voltage, temperature and load data uploaded by the status acquisition module; The processing unit is configured to perform the following operations: The state-space model of the brushless motor is established using the Runge-Kutta fourth-order method; Construct the control cost function based on LQR control theory; Solving the control strategy based on the Euler–Lagrange variational method; Use a deep neural network or support vector regression model to fit the control input; The output interface unit is used to send the control input generated by the processing unit to the motor actuator. The storage unit is used to store state data, cost function model, network training weights, feedback information, and parameter update history for online learning and updating.

10. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed, it performs the brushless motor energy-saving control adjustment method according to any one of claims 1-7.

Images