An improved incremental deadbeat current prediction control method with parameter self-tuning for cutting motor

Abstract

The present application belongs to the technical field of coal mine tunneling machine cutting motor control, and specifically discloses a parameter self-tuning improved incremental deadbeat current prediction control method for cutting motor. First, the incremental deadbeat current prediction control IDPCC is adopted, so that the influence of slowly changing parameters such as permanent magnet flux linkage and back electromotive force is greatly offset after subtraction, and on this basis, an improved ESO is designed. By introducing a nonlinear error feedback mechanism, the robustness to inductance and resistance parameter perturbation, digital delay and unmodeled dynamics is improved. In addition, on the basis of wideband disturbance compensation of the improved ESO, a parallel quasi-resonant controller is connected, which makes up for the short board of the wideband ESO in the specific frequency harmonic suppression ability, effectively reduces the low-speed current ripple and speed fluctuation, and proposes to use the DDPG algorithm to self-adaptively set the current loop control parameters, avoiding the defect that the traditional artificial trial-and-error method is easy to fall into local optimum. The present application improves the adaptive ability to the low-speed complex operating environment.

Description

Technical Field

[0001] This invention belongs to the field of control technology for cutting motors in coal mine tunneling machines, and specifically relates to an improved incremental zero-delay current prediction control method for cutting motors with parameter self-tuning. Background Technology

[0002] Tunnel boring machines (TBMs) are the core equipment for tunneling in coal mines. Their cutting sections endure complex and variable impact loads, placing stringent demands on the reliability, dynamic response, and low-speed, high-torque output capabilities of the drive system. As the power source for the TBM's cutting section, the control performance of the cutting motor directly affects tunneling efficiency and equipment safety. Permanent magnet synchronous motors (PMSMs) are considered ideal for TBM cutting drives due to their simple structure, high efficiency, high power density, and excellent low-speed, high-torque output characteristics. However, mining equipment operates under extreme conditions such as heavy loads, humidity, and explosion-proof environments year-round, and the engineering application of permanent magnet drive technology in such environments still faces many control-level challenges.

[0003] During the cutting operation of the tunneling machine, the cutting motor operates under low-speed and heavy-load conditions for a long time. The load torque fluctuates frequently with the change in coal and rock hardness, and the impact load of the cutting head exhibits obvious non-periodic and strong nonlinear characteristics.

[0004] Therefore, the control system of the cutting motor of a tunneling machine needs to possess both rapid current tracking capability and strong parameter robustness to cope with the impact of nonlinear changes in motor parameters on control performance under harsh operating conditions. In traditional permanent magnet synchronous motor vector control systems, the inner current loop often uses a proportional-integral (PI) controller, which is essentially a linear controller based on error feedback. It performs well when motor parameters are constant and the load is stable. However, when the motor operates under low-speed, heavy-load conditions, the stator resistance drifts significantly with increasing temperature, and the effects of inductor magnetic saturation and cross-coupling exhibit nonlinear changes. The rotor flux may also decay due to the temperature rise of the permanent magnet. Under these time-varying parameter conditions, the fixed-parameter PI controller struggles to maintain optimal control performance, and the dynamic response and steady-state accuracy of the current loop decrease significantly, thus affecting the overall control quality of the machine.

[0005] With the development of control technology, deadbeat predictive current control (DPCC) has attracted widespread attention in the field of high-performance permanent magnet synchronous motor drives due to its advantages such as fast dynamic response and no overshoot in current tracking. This method is based on a discrete mathematical model of the motor and calculates the voltage vector to be applied in the current cycle, theoretically ensuring that the current error in the next cycle is zero, thus achieving current tracking within one cycle. However, the nature of DPCC dictates that it is highly dependent on the accuracy of the motor model parameters. When the control system faces parameter mismatch and external disturbances, it is prone to deviations in voltage vector calculation.

[0006] Studies have shown that DPCC (Dynamic Permanent Magnet Control) is extremely sensitive to motor parameters, especially inductance parameters. When there is a deviation between the controller parameters and the actual motor parameters, the current tracking error increases significantly, and in severe cases, it can even lead to system instability. In the actual operation of the cutting motor of a tunneling machine, the motor parameters change in real time with the working conditions, and the mismatch between the fixed model parameters in the controller and the actual object is particularly prominent. The robustness bottleneck of traditional DPCC in engineering applications urgently needs to be overcome. In addition, the inherent delay of the control system (such as current sampling delay and inverter action delay) and external torque disturbances will exacerbate the current tracking error and deteriorate the robustness of the control. Therefore, it is urgent to design an improved DPCC method for the special working conditions of the low-speed permanent magnet cutting motor of the tunneling machine.

[0007] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art. Summary of the Invention

[0008] To address the aforementioned problems of traditional DPCC methods in the control of low-speed permanent magnet cutting motors in tunneling machines, this invention proposes an improved incremental zero-delay current predictive control method for cutting motors using parameter self-tuning. By employing incremental zero-delay current predictive control (IDPCC), the current prediction formula and voltage calculation formula of traditional DPCC are differentially processed, significantly canceling out the influence of slowly varying parameters such as permanent magnet flux linkage and back electromotive force after subtraction. Based on this, an improved ESO is designed, introducing a nonlinear error feedback mechanism to improve... The system demonstrates robustness against inductance and resistance parameter perturbations, digital delays, and unmodeled dynamics. Building upon the improved ESO wideband disturbance compensation, a resonant controller connected in parallel with the improved ESO is designed to compensate for the insufficient suppression capability of the wideband ESO for specific frequency harmonics, thereby reducing low-speed current ripple and speed fluctuations. Furthermore, the proposed DDPG algorithm is used to globally optimize the current loop control parameter set, ultimately facilitating real-time observation and compensation of multi-source disturbances such as parameter mismatch, load disturbances, and speed fluctuations in the low-speed operation of the permanent magnet cutting motor system.

[0009] To achieve the above objectives, the present invention adopts the following technical solution: An improved incremental zero-delay current prediction control method for cutting motors with parameter self-tuning includes the following steps: Step 1. Real-time acquisition of the three-phase current of the tunneling machine's cutting motor, and conversion of it into current values ​​in the dq coordinate system through Clark transformation and Park transformation; simultaneously, real-time acquisition of the motor rotor position and actual speed; Step 2. Input the setpoint speed and the actual speed, which are then calculated by the speed loop PI controller. Shaft reference current, using shaft voltage Control strategy settings The shaft reference current is 0, and the output... Shaft reference current; Step 3. Establish a current loop controller based on improved incremental deadbeat current predictive control. The design process is as follows: First, the current prediction formula and voltage calculation formula of traditional DPCC are differentiated to construct the IDPCC architecture; Under the IDPCC architecture, the current observation formula and disturbance observation formula of the traditional extended state observer (ESO) are differentially processed to construct a nonlinear incremental extended disturbance observer, namely the improved ESO. A nonlinear error feedback mechanism is introduced into the improved ESO, that is, in the error correction stage of the improved ESO, a variable gain nonlinear feedback function is used to replace the fixed gain linear feedback function of the traditional ESO. Design a quasi-resonant controller QRC in parallel with an improved ESO, and design the QRC as... Narrowband high-gain structure; Regarding the current moment and the previous moment The actual current and voltage of the shaft are differentially calculated and input into the improved ESO to obtain the current increment observation value and the total system disturbance increment observation value. The feedforward compensation is then applied to the IDPCC formula to obtain the initial reference voltage. Calculate the current tracking error and input it into QRC to generate the resonance compensation amount, then superimpose them to obtain the final result. Shaft reference voltage; Step 4. Final The shaft reference voltage and rotor electrical angle input are obtained by inverse Park transformation. The coordinate system voltage is used to drive the motor by outputting three-phase voltage from the inverter controlled by SVPWM, and the operating status is collected. Step 5. Design the Deep Deterministic Strategy Gradient (DDPG) algorithm to adaptively tune the current loop control parameters.

[0010] Furthermore, based on the aforementioned improved incremental zero-delay current prediction control method for parameter self-tuning of cutting motors, this invention also proposes a corresponding improved incremental zero-delay current prediction control system for parameter self-tuning of cutting motors.

[0011] This improved incremental zero-delay current prediction control system for the cutting motor with parameter self-tuning includes the following modules: The acquisition module is used to acquire the three-phase current of the cutting motor of the tunneling machine in real time, and convert it into the current value in the dq coordinate system through Clark transformation and Park transformation; at the same time, it acquires the position and actual speed of the motor rotor in real time. The reference current calculation module is used to input the speed setpoint and the actual speed, which are then calculated by the speed loop PI controller. Shaft reference current, using shaft voltage Control strategy settings The shaft reference current is 0, and the output... Shaft reference current; The controller design module is used to build a current loop controller based on the improved incremental deadbeat current predictive control IDPCC; where IDPCC performs differential processing on the current prediction formula and voltage calculation formula of the traditional DPCC. Under the IDPCC architecture, a nonlinear incremental extended disturbance observer, namely the improved ESO, is constructed by differential processing of the current observation formula and disturbance observation formula of the traditional extended state observer (ESO). A nonlinear error feedback mechanism is introduced into the improved ESO, that is, in the error correction stage of the improved ESO, a variable gain nonlinear feedback function is used to replace the fixed gain linear feedback function of the traditional ESO. Design a quasi-resonant controller QRC in parallel with an improved ESO, and design the QRC as... Narrowband high-gain structure; Regarding the current moment and the previous moment The actual current and voltage of the shaft are differentially calculated and input into the improved ESO to obtain the current increment observation value and the total system disturbance increment observation value. The feedforward compensation is then applied to the IDPCC formula to obtain the initial reference voltage. Calculate the current tracking error and input it into QRC to generate the resonance compensation amount, then superimpose them to obtain the final result. Shaft reference voltage; The motor operation control and operation status acquisition module is used to collect the final data. The shaft reference voltage and rotor electrical angle input are obtained by inverse Park transformation. The coordinate system voltage is used to drive the cutting motor of the tunneling machine through the three-phase voltage output of the inverter controlled by SVPWM, and its operating status is collected. It also includes a parameter tuning module for designing the Deep Deterministic Strategy Gradient (DDPG) algorithm to adaptively tune the current loop control parameters.

[0012] Furthermore, based on the above-mentioned improved incremental zero-delay current prediction control method for parameter self-tuning of cutting motor, this invention also proposes a computer device, which includes a memory and one or more processors.

[0013] Executable code is stored in the memory. When the processor executes the executable code, it implements the steps of the above-described improved incremental deadbeat current prediction control method for parameter self-tuning of the cutting motor.

[0014] Furthermore, based on the aforementioned improved incremental zero-delay current predictive control method for cutting motors using parameter self-tuning, this invention also proposes a computer-readable storage medium storing a program. When executed by a processor, this program implements the steps of the aforementioned improved incremental zero-delay current predictive control method for cutting motors using parameter self-tuning.

[0015] The present invention has the following advantages: As described above, this invention discloses an improved incremental deadbeat current predictive control method for cutting motors with parameter self-tuning. In this invention, the current loop controller adopts an incremental deadbeat current predictive control (IDPCC) structure. By differentially processing the current and voltage prediction equations of the traditional DPCC, the influence of slowly varying parameters such as permanent magnet flux linkage and back electromotive force is significantly canceled out after subtraction. This method reduces parameter sensitivity at the mathematical structure level of the prediction model, significantly reduces prediction errors caused by flux linkage mismatch, alleviates the amplification of parameter mismatch effects under low-speed conditions, and improves the system's parameter robustness. Based on the IDPCC architecture, this invention proposes an improved extended state observer (i.e., an improved ESO). Since the improved ESO is a system extended state model established under an incremental framework, it maintains a high degree of consistency with the IDPCC architecture in its dynamic description method, thus enabling accurate observation and compensation of internal and external disturbances. The improved ESO, by introducing a nonlinear error feedback mechanism, achieves increased gain to accelerate disturbance convergence during large error phases and reduced gain to suppress high-frequency noise and observation chatter during small error phases. This overcomes the drawback of traditional linear ESOs, where fixed gain makes it difficult to balance convergence speed and noise suppression, significantly improving the system's robustness to inductor and resistance parameter perturbations, digital delays, and unmodeled dynamics. Building upon the improved ESO's wideband disturbance compensation, this invention further proposes a parallel quasi-resonant controller (QRC). The QRC utilizes its high gain and narrow bandwidth at 6 times the fundamental frequency to accurately suppress periodic errors caused by inverter nonlinearity and flux harmonics under low-speed conditions. This compensates for the shortcomings of wideband ESOs in suppressing specific frequency harmonics without significantly reducing the dynamic response speed of the IDPCC, thus reducing low-speed current ripple and speed fluctuations. Finally, this invention proposes to adaptively tune the parameters of a composite current loop control architecture built by combining an improved ESO, QRC, and IDPCC using the Deep Deterministic Strategy Gradient (DDPG) algorithm. Through continuous interaction between the agent and the control environment, the complex nonlinear mapping relationship between the system state and control parameters is learned, and the optimal control parameters are directly output in the continuous action space. This avoids the shortcomings of traditional manual trial-and-error methods that are prone to getting trapped in local optima. It can simultaneously take into account the system's dynamic response speed, steady-state control accuracy, parameter robustness, and harmonic suppression capability, thereby improving the system's adaptability to low-speed and complex operating environments. Attached Figure Description

[0016] Figure 1 This is a flowchart of the improved incremental zero-delay current prediction control method for parameter self-tuning of cutting motor in an embodiment of the present invention. Figure 2 This is a structural diagram of the nonlinear incremental extended state observer in an embodiment of the present invention; Figure 3 This is a structural diagram of the composite current loop control architecture in an embodiment of the present invention; Figure 4 This is a block diagram of parameter collaborative optimization of the composite current loop control architecture based on the DDPG algorithm in an embodiment of the present invention; Figure 5 The graph shows a comparison of the speed reduction effects of the method of the present invention and the unmodified conventional method on complex disturbance suppression; wherein (a) uses conventional PI control, (b) uses the IDPCC of the improved ESO of the present invention, and (c) uses the method of the present invention. Figure 6 The diagram shows a comparison of the phase current of the method of the present invention and the unmodified traditional method for suppressing complex disturbances; wherein (a) uses traditional PI control, (b) uses the IDPCC of the improved ESO of the present invention, and (c) uses the method of the present invention. Figure 7 This is a comparison chart showing the effect of the DDPG parameter tuning method of the present invention on the suppression of complex disturbances with other tuning methods; Figure 8 This is a comparison chart of the dynamic performance of the method of the present invention with other control methods; wherein... Figure 8 (a) in the diagram is a schematic comparison of the whole; Figure 8 (b) in the image is an enlarged version of part A in (a); Figure 8 (c) in the diagram is an enlarged version of part B in (a); Figure 9 This is a comparison chart of the d-axis current response speed of the method of this invention with other control methods; in Figure 9 (a) in the figure shows a comparison between the method of the present invention and PI control and the IDPCC of the improved ESO. Figure 9 (b) in the figure is a comparison between the method of the present invention and the heuristic algorithm tuning method; Figure 10 This is a comparison chart of the q-axis current response speed of the method of this invention with other control methods; in Figure 10 (a) in the figure shows a comparison between the method of the present invention and PI control and the IDPCC of the improved ESO. Figure 10 (b) in the figure is a comparison between the method of the present invention and the heuristic algorithm tuning method. Detailed Implementation

[0017] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments: Example 1 like Figure 1As shown, this embodiment describes an improved incremental zero-delay current predictive control method for cutting motors with parameter self-tuning. In this method, the current loop controller adopts an incremental zero-delay current predictive control (IDPCC) structure. The current prediction formula and voltage calculation formula of the traditional DPCC are differentially processed, so that the influence of slowly varying parameters such as permanent magnet flux linkage and back electromotive force is significantly canceled out after subtraction. Based on this, a current loop controller based on the improved IDPCC is further designed. First, the current observation formula and disturbance observation formula of the traditional ESO are differentially processed under the incremental framework to match the dynamic structure of the IDPCC. Furthermore, in the error correction stage of the improved ESO, a variable gain nonlinear feedback function is used instead of the traditional fixed gain linear feedback function, thus designing the improved ESO. Next, the quasi-resonant controller QRC is designed as… The system employs a narrowband, high-gain structure, forming a parallel complementary architecture with the improved ESO total disturbance compensation channel. It performs differential calculations on the actual dq-axis current and voltage at the current and previous moments, inputting these values ​​into the improved ESO to obtain the observed current increment and total system disturbance increment. These are then fed forward to the IDPCC formula to obtain the initial reference voltage. The current tracking error is calculated and input into the quasi-resonant controller QRC to generate the resonant compensation amount, which is then superimposed to obtain the final dq-axis reference voltage. The final reference voltage and rotor electrical angle are then subjected to inverse Park transformation to obtain the α-β coordinate system voltage. This voltage is then used by the SVPWM module to generate six drive pulses, controlling the inverter to output three-phase voltage to drive the motor and collect its operating status. The operating status is input into the DDPG algorithm, which outputs the optimal current loop control parameters from the DDPG agent.

[0018] The improved incremental zero-delay current prediction control method for the cutting motor using parameter self-tuning in this embodiment includes the following steps: Step 1. Collect the three-phase current of the tunneling machine cutting motor in real time, and convert it into current value in dq coordinate system through Clark transformation and Park transformation; at the same time, obtain the position and actual speed of the motor rotor in real time.

[0019] This invention uses a current sensor to collect the three-phase current of a motor in real time. And the sampled current is converted through Clark transformation. Current value converted to α-β coordinate system .

[0020] Further through Park transformation Convert the current value to the dq coordinate system.

[0021] Simultaneously, the rotor electrical angle of the motor is acquired in real time via a photoelectric encoder. and rotor angular velocity , That is, the actual rotational speed.

[0022] Step 2. Input the setpoint speed. Compared with actual speed ,use shaft voltage Control strategy settings The shaft reference current is 0, calculated by the speed loop PI controller. Shaft reference current Output Shaft reference current , .

[0023] use The control method involves setting the d-axis current value to a constant 0.

[0024] A traditional PI controller is used on the speed loop to measure the actual speed. The error between the speed and the target speed is fed into the PI controller, which calculates the reference value of the q-axis current using proportional-integral calculations. .

[0025] Step 3. Establish a current loop controller based on the improved incremental deadbeat current predictive control (IDPCC).

[0026] The construction process of the improved IDPCC is as follows: First, the current prediction formula and voltage calculation formula of traditional DPCC are differentiated to construct the IDPCC architecture.

[0027] Then, under the IDPCC architecture, an improved ESO is constructed by differentially processing the current observation formula and disturbance observation formula of the traditional extended state observer (ESO) to match the dynamic structure of IDPCC.

[0028] Furthermore, a nonlinear error feedback mechanism is introduced into the improved ESO. That is, in the error correction stage of the improved ESO, a variable gain nonlinear feedback function is used instead of the fixed gain linear feedback function of the traditional ESO.

[0029] Following this, a QRC was designed to be connected in parallel with the improved ESO, and the QRC was designed as follows: Narrowband high-gain structure.

[0030] Regarding the current moment and the previous moment The actual current and voltage of the shaft are differentially calculated and input into the improved ESO to obtain the current increment observation value and the total system disturbance increment observation value. The feedforward compensation is then applied to the IDPCC formula to obtain the initial reference voltage.

[0031] Calculate the current tracking error and input it into QRC to generate the resonance compensation amount, then superimpose them to obtain the final result. Shaft reference voltage.

[0032] Specifically, the processing flow of the current loop controller based on the improved IDPCC in this embodiment is as follows: Compare the current time with the previous time. After performing differential calculations on the actual shaft current and voltage, we obtain Current increment and Voltage increment and Input improved ESO.

[0033] The improved ESO internally estimates the total system disturbance increment observations, composed of inductor-resistance perturbations, digital delays, and unmodeled dynamics, online through a nonlinear error feedback mechanism. and and current increment observations and .

[0034] The total disturbance increment observation value of the system and Current increment observations and The initial reference voltage is obtained from the current prediction and voltage calculation formulas for feedforward compensation to the IDPCC. and .

[0035] in and These represent the initial reference voltage along the d-axis and the initial reference voltage along the q-axis, respectively.

[0036] Simultaneous calculation Axis current tracking error , The input quasi-resonant controller QRC generates a low-speed 6th harmonic periodic resonant compensation quantity. , .

[0037] in Indicates the d-axis reference current. Represents the actual current along the d-axis. This indicates the d-axis current tracking error. Indicates the q-axis reference current. This represents the actual current along the q-axis. This represents the q-axis current tracking error.

[0038] The initial reference voltage after feedforward compensation and The final result is obtained by superimposing the low-speed 6th harmonic periodic resonance compensation amount. Shaft reference voltage and The formula is as follows: , ;in , They represent the final , Shaft reference voltage, , Representing the initial , Shaft reference voltage.

[0039] Step 3.1. Assuming the motor operates under linear conditions, its permanent magnet motor in The mathematical model in the coordinate system is represented as follows: (1) in and They are respectively shaft and Shaft stator voltage, and They are respectively shaft and Shaft stator current, and For magnetic flux in Components in coordinate system The electric angular velocity of the motor rotor. This is the stator winding resistance.

[0040] and for shaft and Shaft stator inductance, It is a permanent magnet flux linkage.

[0041] (2) It should be noted that the above ideal linear model serves only as the theoretical basis for the derivation of subsequent control algorithms.

[0042] The tunneling machine cutting motor targeted by this invention suffers from serious problems such as time-varying parameters, magnetic circuit saturation, load impact, and inverter nonlinearity during actual operation, which completely deviates from the linear assumption. This is the fundamental reason why the control performance of the traditional DPCC control method based on the ideal linear model deteriorates under low-speed conditions of the tunneling machine cutting motor.

[0043] Step 3.2. Establish a current loop controller based on IDPCC.

[0044] Taking surface-mounted PMSM as the target, i.e. By discretizing formula (1) using the forward Euler method, the traditional deadbeat current predictive control DPCC is obtained. The formula for predicting the current at time t is as follows: (3) In the formula , They represent The predicted current along the d-axis and the predicted current along the q-axis at time t; , , , For time k d-axis current, q-axis current, d-axis voltage, and q-axis voltage in the coordinate system.

[0045] ; in This represents the electric angular velocity of the motor rotor at time k. This is the sampling time of the controller.

[0046] As can be seen from equation (3), a traditional DPCC includes a stator inductor. Stator resistance and permanent magnet chain The accuracy of motor parameters directly affects the current prediction results, which in turn determines the steady-state control accuracy and dynamic response performance of the system.

[0047] As the motor operates for extended periods, parameters such as resistance, inductance, and flux linkage are susceptible to mismatch due to factors like temperature rise, magnetic saturation, and changes in mechanical operating conditions. This mismatch leads to deviations between the prediction model and the actual system. Especially under low-speed operating conditions, the impact of parameter mismatch on current prediction errors is amplified due to weakened back electromotive force, decreased system immunity, and more pronounced parameter perturbations. In severe cases, this can even cause the failure of deadbeat constraints, resulting in increased current fluctuations and decreased system stability. To reduce the controller's dependence on an accurate model, this invention designs an incremental DPCC, or IDPCC.

[0048] IDPCC differentiates the traditional DPCC current and voltage prediction formulas, transforming slowly changing terms such as permanent magnet flux linkage and back electromotive force into incremental terms. This effectively reduces the role of permanent magnet flux linkage parameters in the control model, thereby weakening the impact of flux linkage parameter mismatch on current prediction accuracy.

[0049] Traditional DPCC Time and By subtracting the discretized formulas for the current at each time step, the predicted current of the IDPCC is designed as follows: (4) Since electrical angular velocity is a mechanical variable of the motor, its rate of change is relatively slow compared to the sampling period. Within several sampling periods, the electrical angular velocity is usually approximately constant. .

[0050] Therefore, , Equivalent to a constant matrix , Equation (4) can be rearranged as follows: (5) in ; .

[0051] The d-axis current increment at time k. The q-axis current increment at time k. The voltage increment along the d-axis at time k. This represents the q-axis voltage increment at time k.

[0052] , , , For the first time Current and voltage in a coordinate system , , , For the first time Current and voltage in a coordinate system.

[0053] According to the forward Euler method, we get time, The discrete formula for voltage at any given time is given by equation (6): (6) in Indicates the first d-axis voltage at time t, Indicates the first q-axis voltage at time t; Indicates the first d-axis current at time t, Indicates the first The q-axis current at time t.

[0054] Considering that the analog-to-digital conversion and voltage vector calculation processes take time, the calculated voltage cannot be immediately applied to the motor. Without proper compensation, this can negatively impact control performance.

[0055] Taking into account the one-step delay compensation, the actual current can track the reference value within two control cycles, that is: (7) in , Indicates the first The predicted d-axis current and q-axis current at time t; This represents the d-axis reference current at time k. This represents the q-axis reference current at time k.

[0056] To calculate the controller reference voltage that enables the actual current to track the reference current within two control cycles, the following steps are taken: After subtracting the voltage formulas at time k and time k, we can use equation (7) to obtain... The predicted current value at time k is replaced with the reference current value at time k to obtain the IDPCC. Formula for calculating the voltage reference value at a given time: (8) in , Indicates the first The increments of the d-axis and q-axis current tracking errors at time t, i.e., the... The reference current given at time t and the model prediction of the first time The difference in current between moments.

[0057] , Indicates the first The predicted current increments along the d-axis and q-axis at time t, i.e., at the t-th time... Under the applied voltage at time t, the current flows from the first t. Actual value at time 1 to 2 The natural variation of the predicted value at any given time.

[0058] ; .

[0059] As can be seen from equations (5) and (8), the flux linkage value has been eliminated from the current prediction and voltage reference value calculation at the control level, thereby reducing the impact of slow changes in flux linkage value on control performance.

[0060] Step 3.2. Establishing the improved ESO.

[0061] IDPCC weakens the influence of slowly changing model terms of permanent magnet flux linkage by differentiating and improves the robustness of parameter mismatch. However, it is still sensitive to fast-changing disturbances such as inductance and resistance perturbations, digital delays and unmodeled dynamics. The low-speed heavy-load and frequent impact operation characteristics of the cutting motor lead to frequent deep saturation of its magnetic circuit and large fluctuations in winding temperature. The model parameters are severely mismatched and strong coupling of multiple sources of disturbances. Simple IDPCC control is difficult to meet engineering requirements.

[0062] Existing technologies mostly employ Extended State Observer (ESO) to classify various disturbances into a total disturbance for online estimation and compensation. However, traditional ESOs are based on absolute mathematical models of the motor, using absolute state variables as the observation objects. This leads to a dynamic description mismatch with Incremental Dynamic Controlled Circular Curve (IDPCC), limiting the compensation effect. To address these shortcomings, this invention designs an improved ESO. It establishes an extended state model within an incremental framework, setting the observer's state variables as state increments between adjacent sampling periods. This achieves dynamic structural unification with IDPCC. Furthermore, by leveraging the weak changes in slow variables such as permanent magnet flux linkage within adjacent periods, differential analysis further weakens their influence on the incremental model, reducing the observer's dependence on precise model parameters.

[0063] Meanwhile, a nonlinear error feedback mechanism is introduced into the improved ESO to replace the fixed gain of the traditional ESO. The equivalent gain of the nonlinear feedback is adjusted according to the magnitude of the observation error. In the large error stage, the excessive amplification of the nonlinear feedback function is suppressed, and in the small error stage, a high local feedback gain is maintained, thus taking into account both observation error convergence and steady-state accuracy.

[0064] This invention, through a combination of incremental structure and nonlinear feedback, weakens the influence of flux linkage parameter perturbations, achieving architectural unification and dynamic matching between the improved ESO and IDPCC, enhancing the ability to observe and compensate for rapidly changing disturbances, and providing support for low-speed, high-precision current control and torque ripple suppression. The constructed structure is as follows: Figure 2 As shown, the specific implementation method is as follows: The second-order nonlinear ESO under the absolute model of the motor is expressed as follows: (9) and This represents the difference between the actual and observed current values ​​on the d-axis and q-axis.

[0065] and These are the current observations along the d-axis and q-axis. and This represents the actual current values ​​along the d-axis and q-axis. and These are the perturbation observations for the d-axis and q-axis; The feedback gain for current observation; Feedback gain of perturbation observation.

[0066] , , , Here is a nonlinear feedback function, where , , , All are expressed using equation (10): (10) nonlinear functions middle, For symbolic functions, The input observation error (e.g.) , ). It is a nonlinear factor (such as) , ), It is a filter factor (such as) , ).

[0067] when At this time, the function can achieve the effect of large error and small gain, and small error and large gain.

[0068] Discretize the state equations of the nonlinear ESO using the forward Euler method to obtain the observer expressions at time k and time k+1: (11) (12) In the formula, and Representing the d-axis and q-axis respectively The difference between the actual and observed current values ​​at any given moment; and Representing the d-axis and q-axis respectively The difference between the actual and observed current values ​​at any given time.

[0069] and Representing the d-axis and q-axis respectively Actual current value at any given time.

[0070] and Representing the d-axis and q-axis respectively Actual current value at any given time.

[0071] and Representing the d-axis and q-axis respectively Actual voltage value at any given time.

[0072] and Representing the d-axis and q-axis respectively Actual voltage value at any given time.

[0073] and The d-axis and q-axis are respectively Current observation at any given time; and The d-axis and q-axis are respectively Current observation at any given time; and The d-axis and q-axis are respectively Current observation at any given time.

[0074] and The d-axis and q-axis are respectively Perturbation of observation values ​​at any time; and For the d-axis and q-axis Perturbation of observation values ​​at any time; and For the d-axis and q-axis The observed values ​​are constantly disturbed.

[0075] Subtracting equation (12) from equation (11) achieves the adaptation of the observer to the IDPCC structure and eliminates the influence of magnetic flux in the observation process, resulting in the improved ESO equation expression: (13) ; ; ; In the formula and Representing the d-axis and q-axis respectively The increment of the current observation at any given time; and Representing the d-axis and q-axis respectively The increment of the current observation value at time.

[0076] and Representing the d-axis and q-axis respectively The increment of the perturbation observation at any given time; and Representing the d-axis and q-axis respectively The increment of the observed value at any given time.

[0077] , Here, is the nonlinear gain parameter, and E is the 2D identity matrix.

[0078] and Representing the d-axis and q-axis respectively The increment of the actual voltage value at any given moment; and Representing the d-axis and q-axis respectively The increment of the difference between the actual value and the observed value of the current at any given time.

[0079] Step 3.3. Quasi-resonant controller design.

[0080] This invention employs an improved ESO (i.e. Figure 3 The IDPCC (Improved ESO-IDPCC) that performs feedforward compensation for the ESO can perform online observation and compensation for the total system disturbance. However, its compensation capability is still limited for specific frequency errors caused by inverter nonlinearity, dq axis coupling and periodic load disturbances under low-speed operating conditions.

[0081] To further improve the system's ability to suppress periodic disturbances, this invention adds a parallel quasi-resonant controller (QRC) to the improved ESO-IDPCC current loop compensation. The QRC provides high gain near the target frequency, enabling selective compensation for periodic error components. The control structure is as follows: Figure 3 As shown.

[0082] Compared to traditional first-order filter structures, the second-order structure of QRC can achieve higher resonant gain and narrower frequency selectivity bandwidth near the target frequency. This allows for effective and precise suppression of current harmonics and torque ripple under low-speed conditions without significantly reducing the dynamic response speed of incremental DPCC. The specific implementation method is as follows: The error value between the reference current and the actual current is input into the QRC module; by introducing a damping term in the resonant circuit, the frequency selectivity and stability of the system are improved, and the QRC transfer function is... Represented as: (14) In the formula To control the angular frequency of the target harmonic, For resonant gain, Here, s represents the bandwidth adjustment parameter, and s denotes the continuous domain operator. This is the sampling time of the controller.

[0083] Using a bilinear transformation to discretize equation (14), the continuous domain operator s is mapped to the discrete domain operator z as follows: ; The discretized QRC expression is obtained as follows: (15) In the formula, The discretized QRC expression; .

[0084] Simultaneous calculation Axis current tracking error , The input quasi-resonant controller QRC generates a low-speed 6th harmonic periodic resonant compensation quantity. , .

[0085] The periodic resonance is tracked and compensated on the current loop. The compensation amount for the low-speed 6th harmonic periodic resonance is superimposed with the voltage after feedforward compensation by the improved ESO to obtain the final dq-axis reference voltage. and The formula is as follows: , ;in , They represent the final , Shaft reference voltage, , Representing the initial , Shaft reference voltage.

[0086] Step 4. Final The shaft reference voltage and rotor electrical angle input are obtained by inverse Park transformation. The coordinate system voltage is used to generate six drive pulses via SVPWM, which control the inverter to output three-phase voltage to drive the cutting motor of the tunneling machine.

[0087] Specifically, it will eventually Shaft reference voltage and With rotor electrical angle Input the inverse Park transform module to convert to a static state. Coordinate system voltage U α U β Then, the space vector pulse width modulation module (SVPWM) generates six inverter drive pulse signals, thereby controlling the inverter to output three-phase voltage and drive the tunneling machine's cutting motor to run.

[0088] Simultaneously, this invention also collects the operating status of the tunneling machine's cutting motor. Specifically, it simultaneously collects the system's operating status signals in real time, including... Shaft current value and its tracking error, rotational speed and its tracking error.

[0089] Step 5. Design the Deep Deterministic Strategy Gradient (DDPG) algorithm to adaptively tune the current loop control parameters.

[0090] Input the real-time operating status of the system into the Deep Deterministic Policy Gradient (DDPG) algorithm.

[0091] The DDPG agent learns the nonlinear mapping relationship between system state and control parameters through continuous interaction with the motor control environment, and outputs an optimized set of current loop control parameters to achieve global collaborative optimization of parameters in the composite current loop control architecture.

[0092] The current loop control parameter set includes two types: control parameters for the improved ESO and control parameters for the QRC.

[0093] The control parameters of the improved ESO include the feedback gain of current observation. Feedback gain of perturbation observation Nonlinear factors , Filtering factor , The control parameters of QRC include resonant gain. and bandwidth .

[0094] To further enhance the overall performance of the composite current loop control architecture built from the improved ESO, QRC, and IDPCC, the DDPG algorithm is introduced to collaboratively optimize the parameters of the composite current loop control architecture. The composite current loop control architecture achieves fast current tracking through IDPCC, while the improved ESO simultaneously performs online compensation for wideband dynamic disturbances and inductor and resistance parameter perturbations. QRC specifically suppresses low-speed periodic harmonic errors, and the three are deeply integrated to form a unified control closed loop.

[0095] In the entire composite current loop control architecture, the control parameters of the improved ESO and the control parameters of the QRC are strongly coupled. Adjusting a single parameter will simultaneously affect the dynamic response, steady-state accuracy, parameter robustness, and harmonic suppression performance.

[0096] For example, while increasing the feedback gain of the improved ESO can enhance disturbance compensation capability, it amplifies high-frequency noise and weakens the harmonic suppression effect of QRC; adjusting the QRC parameters can reduce torque ripple, but it may change the system damping characteristics and affect the dynamic tracking performance of IDPCC. Therefore, the composite control of this composite current loop control architecture is essentially a high-dimensional nonlinear multi-objective optimization problem, and traditional empirical trial-and-error or single-index tuning methods are difficult to simultaneously meet multi-dimensional performance requirements.

[0097] To address the aforementioned issues, this invention proposes a DDPG parameter self-tuning method based on a multi-objective reward function. This method integrates current tracking error, response time, and overshoot as dynamic indicators, along with current ripple and speed fluctuation as steady-state indicators, into the reward function to calculate the reward value for the current cycle, achieving global collaborative optimization of the current loop control parameter set. During training, the DDPG agent continuously interacts with the motor control environment, learning the nonlinear mapping relationship between system state and control parameters. It adaptively adjusts the current loop control parameters of the composite current loop control architecture, achieving an optimal balance between dynamic response, steady-state accuracy, parameter robustness, and harmonic suppression. This method fully leverages the synergistic advantages of the composite architecture while optimizing the disturbance compensation capability of the ESO and the harmonic suppression effect of the QRC, thus improving the overall control quality and robustness of the permanent magnet synchronous motor at low speeds.

[0098] The DDPG parameter self-tuning process is as follows: Figure 4 As shown. The process of DDPG parameter self-tuning is as follows: After defining the input and output, two core modules, the policy network and the value network, are built. On this basis, the online policy network, the online Q network, and the corresponding target policy network and target Q network are built. The weight parameters of each network are initialized, and an experience playback storage unit with a capacity of 10,000 is built. A circular overwrite mechanism is used to store interaction samples. When the number of samples exceeds the capacity, the new sample automatically overwrites the oldest sample stored.

[0099] The control environment includes state variables ( The DDPG algorithm inputs the shaft current and its tracking error, rotational speed and its tracking error, and reward value into the control environment. The DDPG algorithm uses the current loop control parameter set as the action quantity.

[0100] Online policy networks based on current state variables Generate control actions And superimposed exploration noise The control environment after (Gaussian noise) acts on the composite current loop control architecture: (16) In the formula, For online policy networks In parameters Below, the state quantity corresponding to the current time t. The action output; The DDPG algorithm outputs the current time-to-time reward from the environment. and the state quantity at the next moment , and the current state quantity and amount of movement Forming samples The samples are stored in the experience unit. When the number of training epochs exceeds 10 (the cumulative number of samples in the experience unit reaches 100), the DDPG algorithm randomly selects 64 samples from the experience replay unit to form a batch training set and performs a network update operation. The online Q network outputs parameters from the value network module. Initialization, while the online Q-network transmits the value gradient back to the value network module. During training, the DDPG algorithm randomly draws a batch of samples from empirical units. The target policy network is based on the state at the next time step. Generate the target action for the next moment. And input the target Q network Combined with the current time reward of the sample and conversion factor The time-series difference target value of the sample was calculated. The time-series difference target value Compared with the current estimate of the online Q network The difference is calculated to obtain the temporal difference error. The algorithm uses gradient descent to minimize the mean squared loss of the temporal difference error of the batch samples and updates the parameters of the online Q-network. The value gradient of the online Q-network for each action is calculated, and the policy gradient is calculated by combining the parameter gradients of the online policy network. The parameters of the online policy network are then updated using the gradient ascent method. The optimization objective is to maximize the mean value estimate of the policy output action by the online Q-network. After each parameter update of the online Q-network and the online policy network, a soft update strategy is immediately used to synchronize the target policy network parameters. and target Q network parameters This allows the target network to slowly track the parameter changes of the online network with extremely small steps, thereby effectively improving training stability and promoting algorithm convergence.

[0101] Based on the motion adjustment amount output by the DDPG algorithm, the parameters are iteratively updated within a unified normalized space and then restored to the actual physical parameters to ensure that all parameters are always within the preset feasible range during the optimization process.

[0102] The DDPG of this invention outputs an optimized current loop control parameter set after completing parameter optimization in the simulation environment. During actual operation, fixed parameters are directly loaded to avoid the risk of system instability that may be caused by online training.

[0103] This invention designs the reward function R as a multi-objective comprehensive function, which comprehensively considers performance indicators such as current tracking error, response speed, overshoot and current ripple, and achieves synergistic optimization of system dynamic performance and steady-state performance.

[0104] Because of the severe coupling relationship between parameters in the entire composite current loop control architecture of this invention, this invention optimizes each control parameter in a unified manner through a multi-objective reward function, so as to achieve comprehensive coordination between dynamic response speed, disturbance compensation capability and steady-state harmonic suppression, thereby improving the control accuracy and system robustness under low-speed operating conditions.

[0105] (17) In the formula, As a reward indicator value, This is a comprehensive index value for dynamic performance. The steady-state performance index value is t1; t1 is the time to reach the set speed. For the speed start-up overshoot recovery time, For speed-initiated overshoot, This refers to the speed recovery time during a sudden load change. This represents the speed disturbance caused by sudden load changes. This represents the dq-axis current ripple value. This represents the difference between the observed and actual values ​​of the dq-axis current. This represents the difference between the reference and actual values ​​of the dq-axis current. c1~c8 are the weighting coefficients for each indicator, and s1~s8 are the normalized values. This invention uses negative reward values ​​for training and optimization; the optimal optimization result is obtained when the R value is maximized.

[0106] The method of this invention obtains an optimized current loop control parameter set during the online interactive training phase of simulation. First, the DDPG algorithm is trained in the simulation environment to obtain the optimized current loop control parameter set. Every 20 training rounds, the weight parameters of the current main network and target network, as well as the experience replay buffer data, are automatically saved. After all training rounds are completed, the optimized current loop control parameter set is output.

[0107] contrast Figure 5 and Figure 6 As can be seen from (a) and (b) above, compared with PI control, the improved ESO-IDPCC designed in this invention can suppress speed fluctuations and phase current distortions caused by parameter disturbances, effectively suppressing aperiodic disturbances. Further comparison... Figure 5 and Figure 6 As can be seen from (b) and (c), the control structure designed in this invention, on the basis of effectively suppressing non-periodic disturbances, further reduces the speed fluctuations and current distortions caused by periodic disturbances.

[0108] contrast Figure 7The curves of different tuning methods show that the method of this invention (i.e. the parameter tuning method based on the DDPG algorithm adopted in this invention) has the best effect in suppressing speed fluctuations compared with the manual tuning (QRC-IESO-IDPCC) and the heuristic algorithm tuning method (using the PSO algorithm). It also has a wider exploration depth than the PSO algorithm.

[0109] contrast Figure 8 Among the different control methods used in this invention, the overshoot and torque surge drop during startup are minimized. The method of this invention has a faster dynamic response speed during startup and torque surge, and a stronger ability to cope with torque disturbances.

[0110] Figure 9 and Figure 10 As can be seen from the d-axis and q-axis currents of different control methods in (a), the method of this invention, compared with PI control and the improved ESO-IDPCC, has a faster tracking speed of the dq-axis current, the smallest overshoot of current mutations, and reduces the tracking error of the dq-axis current. (Comparison) Figure 9 and Figure 10 It is not difficult to see from (b) that, compared with the heuristic algorithm tuning method (using PSO algorithm tuning), the method of the present invention has a faster dq axis current recovery speed, no oscillation during the recovery process, and better dynamic process.

[0111] Example 2 This embodiment 2 describes an improved incremental zero-delay current prediction control system for parameter self-tuning of a cutting motor. This system is based on the same inventive concept as the improved incremental zero-delay current prediction control method for parameter self-tuning of a cutting motor described above.

[0112] An improved incremental zero-delay current prediction control system for cutting motors with parameter self-tuning includes the following modules: The acquisition module is used to acquire the three-phase current of the cutting motor of the tunneling machine in real time, and convert it into the current value in the dq coordinate system through Clark transformation and Park transformation; at the same time, it acquires the position and actual speed of the motor rotor in real time. The reference current calculation module is used to input the speed setpoint and the actual speed, which are then calculated by the speed loop PI controller. Shaft reference current, using shaft voltage Control strategy settings The shaft reference current is 0, and the output... Shaft reference current; The controller design module is used to build a current loop controller based on the improved incremental deadbeat current predictive control IDPCC; where IDPCC performs differential processing on the current prediction formula and voltage calculation formula of the traditional DPCC. Under the IDPCC architecture, a nonlinear incremental extended disturbance observer, namely the improved ESO, is constructed by differential processing of the current observation formula and disturbance observation formula of the traditional extended state observer (ESO). A nonlinear error feedback mechanism is introduced into the improved ESO, that is, in the error correction stage of the improved ESO, a variable gain nonlinear feedback function is used to replace the fixed gain linear feedback function of the traditional ESO. Design a quasi-resonant controller QRC in parallel with an improved ESO, and design the QRC as... Narrowband high-gain structure; Regarding the current moment and the previous moment The actual current and voltage of the shaft are differentially calculated and input into the improved ESO to obtain the current increment observation value and the total system disturbance increment observation value. The feedforward compensation is then applied to the IDPCC formula to obtain the initial reference voltage. Calculate the current tracking error and input it into QRC to generate the resonance compensation amount, then superimpose them to obtain the final result. Shaft reference voltage; The motor operation control and operation status acquisition module is used to collect the final data. The shaft reference voltage and rotor electrical angle input are obtained by inverse Park transformation. The coordinate system voltage is used to drive the cutting motor of the tunneling machine through the three-phase voltage output of the inverter controlled by SVPWM, and its operating status is collected. It also includes a parameter tuning module for designing the Deep Deterministic Strategy Gradient (DDPG) algorithm to adaptively tune the current loop control parameters.

[0113] It should be noted that the implementation process of the functions and roles of each functional module in the system of this embodiment 2 is detailed in the implementation process of the corresponding steps of the method in the above embodiment 1, and will not be repeated here.

[0114] Example 3 This embodiment 3 describes a computer device. The computer device includes a memory and one or more processors. Executable code is stored in the memory. When the processor executes the executable code, it implements the steps of the improved incremental deadbeat current prediction control method for parameter self-tuning of the cutting motor described in embodiment 1 above.

[0115] In this embodiment, the computer device can be any device or apparatus with data processing capabilities, and will not be described in detail here.

[0116] Example 4 This embodiment 4 describes a computer-readable storage medium storing a program that, when executed by a processor, is used to implement the steps of the above-described improved incremental zero-delay current prediction control method for parameter self-tuning of cutting motors.

[0117] The computer-readable storage medium can be an internal storage unit of any device or apparatus with data processing capabilities, such as a hard disk or memory, or an external storage device of any device with data processing capabilities, such as a plug-in hard disk, smart media card (SMC), SD card, flash card, etc.

[0118] Of course, the above description is only a preferred embodiment of the present invention. The present invention is not limited to the above-described embodiments. It should be noted that any equivalent substitutions or obvious modifications made by those skilled in the art under the guidance of this specification fall within the scope of this specification and should be protected by the present invention.

Claims

1. An improved incremental zero-delay current prediction control method for a cutting motor with parameter self-tuning, characterized in that, Includes the following steps: Step 1. Real-time acquisition of the three-phase current of the tunneling machine's cutting motor, and conversion of it into current values ​​in the dq coordinate system through Clark transformation and Park transformation; simultaneously, real-time acquisition of the motor rotor position and actual speed; Step 2. Input the setpoint speed and the actual speed, which are then calculated by the speed loop PI controller. Shaft reference current, using shaft voltage Control strategy settings The shaft reference current is 0, and the output... Shaft reference current; Step 3. Establish a current loop controller based on the improved incremental deadbeat current predictive control (IDPCC); IDPCC differentiates the current prediction formula and voltage calculation formula of traditional DPCC. Under the IDPCC architecture, a nonlinear incremental extended disturbance observer, namely the improved ESO, is constructed by differential processing of the current observation formula and disturbance observation formula of the traditional extended state observer (ESO). A nonlinear error feedback mechanism is introduced into the improved ESO, that is, in the error correction stage of the improved ESO, a variable gain nonlinear feedback function is used to replace the fixed gain linear feedback function of the traditional ESO. Design a quasi-resonant controller QRC in parallel with an improved ESO, and design the QRC as... Narrowband high-gain structure; Regarding the current moment and the previous moment The actual current and voltage of the shaft are differentially calculated and input into the improved ESO to obtain the current increment observation value and the total system disturbance increment observation value. The feedforward compensation is then applied to the IDPCC formula to obtain the initial reference voltage. Calculate the current tracking error and input it into QRC to generate the resonance compensation amount, then superimpose them to obtain the final result. Shaft reference voltage; Step 4. Final The shaft reference voltage and rotor electrical angle input are obtained by inverse Park transformation. The coordinate system voltage is used to drive the cutting motor of the tunneling machine through the three-phase voltage output of the inverter controlled by SVPWM, and its operating status is collected. Step 5. Design the Deep Deterministic Strategy Gradient (DDPG) algorithm to adaptively tune the current loop control parameters.

2. The improved incremental zero-delay current prediction control method for parameter self-tuning of a cutting motor according to claim 1, characterized in that, In step 3, the processing flow of the current loop controller based on the improved IDPCC is as follows: Compare the current time with the previous time. After performing differential calculations on the actual shaft current and voltage, we obtain shaft current increment and Voltage increment and And input the improved ESO; The improved ESO internally estimates the total system disturbance increment observations, composed of inductor-resistance perturbations, digital delays, and unmodeled dynamics, online through a nonlinear error feedback mechanism. and and current increment observations and ; Will and , and The initial reference voltage is obtained from the current prediction and voltage calculation formulas for feedforward compensation to the IDPCC. and ; in and These represent the initial reference voltage along the d-axis and the initial reference voltage along the q-axis, respectively. Simultaneous calculation Axis current tracking error , The input quasi-resonant controller QRC generates a low-speed 6th harmonic periodic resonant compensation quantity. , ; in Indicates the d-axis reference current. Represents the actual current along the d-axis. This indicates the d-axis current tracking error. Indicates the q-axis reference current. This represents the actual current along the q-axis. This indicates the q-axis current tracking error; The initial reference voltage after feedforward compensation and With resonance compensation amount , Superimpose to obtain the final result Shaft reference voltage and The formulas are as follows: ， ； in , They represent the final , Shaft reference voltage.

3. The improved incremental zero-delay current prediction control method for parameter self-tuning of a cutting motor according to claim 2, characterized in that, In step 3, the construction process of the IDPCC architecture is as follows: First, assuming the motor operates under linear conditions, its mathematical model in the dp coordinate system is expressed as follows: （1） in and They are respectively shaft and Shaft-stator voltage; and They are respectively shaft and Shaft stator current, and For magnetic flux in Components in coordinate system The electric angular velocity of the motor rotor. This refers to the stator winding resistance. and for shaft and Shaft stator inductance, For permanent magnet flux linkage; （2） Taking surface-mounted permanent magnet synchronous motors as an example, that is... By discretizing formula (1) using the forward Euler method, the traditional deadbeat current predictive control DPCC is obtained. The formulas for predicting the d-axis and q-axis currents at time t are as follows: (3) In the formula , They represent The predicted current along the d-axis and the predicted current along the q-axis at time t; , , , For time k d-axis current, q-axis current, d-axis voltage, and q-axis voltage in a coordinate system; ； in This represents the electric angular velocity of the motor rotor at time k. The sampling time of the controller; Traditional DPCC Time and By subtracting the discretized formulas for the current at each time step, the predicted current of the IDPCC is designed as follows: (4) Electrical angular velocity is a mechanical variable of the motor, and its rate of change is relatively slow compared to the sampling period. Within several sampling periods, the electrical angular velocity is kept constant. , Equivalent to constant matrices A and C, equation (4) can be rearranged as follows: (5) ， ； in This represents the d-axis current increment at time k. This represents the q-axis current increment at time k. The voltage increment along the d-axis at time k. This represents the q-axis voltage increment at time k. According to the forward Euler method, we get time, The discrete formula for the voltage at any given time is shown in equation (6): (6) in Indicates the first d-axis voltage at time t, Indicates the first q-axis voltage at time t; Indicates the first d-axis current at time t, Indicates the first q-axis current at time t; Taking into account the one-step delay compensation, the actual current can track the reference value within two control cycles, that is: (7) in , Indicates the first The predicted current along the d-axis and the predicted current along the q-axis at time t; This represents the d-axis reference current at time k. This represents the q-axis reference current at time k; Will After subtracting the voltage formulas at time k and time k, we can use equation (7) to obtain... The predicted current value at time k is replaced with the reference current value at time k, resulting in the incremental deadbeat-free current predictive control (IDPCC). Formula for calculating the voltage reference value at a given time: (8) in , Indicates the first The increments of the d-axis and q-axis current tracking errors at time t, i.e., the... The reference current given at time t and the model prediction of the first time The difference in current between moments; , Indicates the first The predicted current increments along the d-axis and q-axis at time t, i.e., at the t-th time... Under the applied voltage at time t, the current flows from the first t. Actual value at time 1 to 2 The natural variation of the predicted value at any given time; ； 。 4. The improved incremental zero-delay current prediction control method for parameter self-tuning of a cutting motor according to claim 3, characterized in that, In step 3, the process of establishing the improved ESO is as follows: First, a second-order nonlinear ESO is established under the absolute model of the motor, expressed as: （9） in and This represents the difference between the actual and observed current values ​​along the d-axis and q-axis. and These are the current observations along the d-axis and q-axis. and This represents the actual current values ​​along the d-axis and q-axis. and These are the perturbation observations for the d-axis and q-axis; The feedback gain for current observation; Feedback gain of perturbation observation; , , , All are nonlinear feedback functions; among them , , , All are expressed using equation (10); (10) nonlinear functions middle, For symbolic functions, The input observation error is... and , It refers to the nonlinear factor. and ; It refers to the filter factor. and ; Discretizing the state equations of the nonlinear ESO using the forward Euler method yields the discrete second-order nonlinear ESO expressions at time k and k+1, as shown in the following formulas: (11) (12) In the formula, and Representing the d-axis and q-axis respectively The difference between the actual and observed current values ​​at any given moment; and Representing the d-axis and q-axis respectively The difference between the actual and observed current values ​​at any given moment; and Representing the d-axis and q-axis respectively Actual current value at any given moment; and Representing the d-axis and q-axis respectively Actual current value at any given moment; and Representing the d-axis and q-axis respectively Actual voltage value at any given moment; and Representing the d-axis and q-axis respectively Actual voltage value at any given moment; and The d-axis and q-axis are respectively Current observation at any given time; and The d-axis and q-axis are respectively Current observation at any given time; and The d-axis and q-axis are respectively Current observation at any given time; and The d-axis and q-axis are respectively Perturbation of observation values ​​at any time; and For the d-axis and q-axis Perturbation of observation values ​​at any time; and For the d-axis and q-axis Perturbation of observation values ​​at any time; Subtracting formula (12) from formula (11), we obtain the equation expression for the improved ESO, as follows: (13) ； ； ； In the formula and Representing the d-axis and q-axis respectively The increment of the current observation at any given time; and Representing the d-axis and q-axis respectively The increment of the current observation at any given time; and Representing the d-axis and q-axis respectively The increment of the perturbation observation at any given time; and Representing the d-axis and q-axis respectively The increment of the perturbation observation at any given time; , Here, E is the nonlinear gain parameter, and E is the 2D identity matrix. and Representing the d-axis and q-axis respectively The increment of the actual voltage value at any given moment; and Representing the d-axis and q-axis respectively The increment of the difference between the actual value and the observed value of the current at any given time.

5. The improved incremental zero-delay current prediction control method for parameter self-tuning of a cutting motor according to claim 4, characterized in that, In step 3, the design process of the quasi-resonant controller is as follows: The error value between the reference current and the actual current is input into the QRC module; by introducing a damping term in the resonant circuit, the frequency selectivity and stability of the system are improved. The QRC transfer function... Represented as: (14) In the formula, ω0 is the angular frequency of the target harmonic, and K r For resonant gain, Let s be the bandwidth adjustment parameter, and let s represent the continuous domain operator. Using a bilinear transformation to discretize equation (14), the continuous domain operator s is mapped to the discrete domain operator z as follows: ； The discretized QRC expression is obtained as follows: (15) In the formula, The discretized QRC expression; ； Simultaneous calculation Axis current tracking error , The input quasi-resonant controller QRC generates a low-speed 6th harmonic periodic resonant compensation quantity. , ; Resonance compensation amount , The voltage after being superimposed with the improved ESO feedforward compensation is used to obtain the final dq-axis reference voltage. and The formula is: , .

6. The improved incremental zero-delay current prediction control method for parameter self-tuning of a cutting motor according to claim 5, characterized in that, In step 5, the DDPG parameter self-tuning method based on the multi-objective reward function takes current tracking error, response time, and overshoot as dynamic indicators, and current ripple and speed fluctuation as steady-state indicators, and incorporates them into the reward function to obtain the reward value for this round, so as to achieve global collaborative optimization of the current loop control parameter set. The current loop control parameter set includes two types of parameters: the control parameters of the improved ESO and the control parameters of the QRC. The control parameters of the improved ESO include the feedback gain of the current observation. Feedback gain of perturbation observation Nonlinear factors , Filtering factor , The control parameters of QRC include resonant gain. and bandwidth .

7. The improved incremental zero-delay current prediction control method for parameter self-tuning of a cutting motor according to claim 6, characterized in that, In step 5, the DDPG parameter self-tuning process is as follows: Include in the control environment The state variables, including shaft current and its tracking error, and rotational speed and its tracking error, along with the reward value, are input into the DDPG algorithm. The DDPG algorithm uses the current loop control parameter set as the action quantity and inputs it into the control environment; After defining the inputs and outputs, construct the action generation mechanism for the online policy network; The expression for the online policy network is: ; in For online policy networks The internal parameters of , the current action and state are respectively represented as a. t and s t ; To enhance the agent's ability to explore its environment, Gaussian noise is introduced during the action generation process. The agent will then generate actions based on its current state s. t Through online policy network Select control action a t As shown in the following formula: (16) For the parameters to be tuned in the current loop control parameter set, a normalized iterative correction mechanism is adopted: Based on the motion adjustment amount output by the DDPG algorithm, the parameters are iteratively updated within a unified normalized space and then restored to the actual physical parameters to ensure that all parameters are always within the preset feasible range during the optimization process.

8. An improved incremental zero-delay current prediction control system for a cutting motor with parameter self-tuning, characterized in that, Includes the following modules: The acquisition module is used to acquire the three-phase current of the cutting motor of the tunneling machine in real time, and convert it into the current value in the dq coordinate system through Clark transformation and Park transformation; at the same time, it acquires the position and actual speed of the motor rotor in real time. The reference current calculation module is used to input the speed setpoint and the actual speed, which are then calculated by the speed loop PI controller. Shaft reference current, using shaft voltage Control strategy settings The shaft reference current is 0, and the output... Shaft reference current; The controller design module is used to build a current loop controller based on the improved incremental deadbeat current predictive control IDPCC; where IDPCC performs differential processing on the current prediction formula and voltage calculation formula of the traditional DPCC. Under the IDPCC architecture, a nonlinear incremental extended disturbance observer, namely the improved ESO, is constructed by differential processing of the current observation formula and disturbance observation formula of the traditional extended state observer (ESO). A nonlinear error feedback mechanism is introduced into the improved ESO, that is, in the error correction stage of the improved ESO, a variable gain nonlinear feedback function is used to replace the fixed gain linear feedback function of the traditional ESO. Design a quasi-resonant controller QRC in parallel with an improved ESO, and design the QRC as... Narrowband high-gain structure; Regarding the current moment and the previous moment The actual current and voltage of the shaft are differentially calculated and input into the improved ESO to obtain the current increment observation value and the total system disturbance increment observation value. The feedforward compensation is then applied to the IDPCC formula to obtain the initial reference voltage. Calculate the current tracking error and input it into QRC to generate the resonance compensation amount, then superimpose them to obtain the final result. Shaft reference voltage; The motor operation control and operation status acquisition module is used to collect the final data. The shaft reference voltage and rotor electrical angle input are obtained by inverse Park transformation. The coordinate system voltage is used to drive the cutting motor of the tunneling machine through the three-phase voltage output of the inverter controlled by SVPWM, and its operating status is collected. It also includes a parameter tuning module for designing the Deep Deterministic Strategy Gradient (DDPG) algorithm to adaptively tune the current loop control parameters.

9. A computer device comprising a memory and one or more processors; executable code is stored in the memory; characterized in that, When the processor executes the executable code, it implements the steps of the improved incremental zero-delay current prediction control method for parameter self-tuning of cutting motors as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a program stored thereon; characterized in that, When executed by the processor, the program is used to implement the steps of the parameter self-tuning improved incremental zero-delay current prediction control method for cutting motors as described in any one of claims 1 to 7.

Images