A three-phase pwm rectifier energy state adaptive disturbance control method and system

Abstract

The present application relates to the technical field of power electronic converter control, in particular to a three-phase PWM rectifier energy state adaptive disturbance control method and system, taking the instantaneous energy state quantity of the three-phase PWM rectifier as an outer ring control object, establishing an extended energy state equation through energy balance relationship, using an extended state observer to estimate and compensate the lumped disturbance in real time, and through parameter adaptive optimization agent, the bandwidth of the extended state observer and the parameters of the current inner loop PI controller are optimized cooperatively, so as to improve the dynamic response speed, DC bus voltage stability and robustness of the three-phase PWM rectifier under the working conditions of load mutation, power grid disturbance and parameter perturbation.

Description

Technical Field

[0001] This invention relates to the field of power electronic converter control technology, specifically to a three-phase PWM rectifier energy state adaptive disturbance rejection control method and system. Background Technology

[0002] With the development of new energy power generation, electric vehicle charging, DC microgrids and energy storage systems, three-phase PWM rectifiers, as an important interface for AC-DC energy conversion, are widely used in various power electronic systems. Their control performance directly affects the stability of DC bus voltage, input current quality and system dynamic response capability.

[0003] Existing three-phase PWM rectifiers mostly employ a dual-closed-loop PI control structure with an outer voltage loop and an inner current loop. This method is simple in structure and widely used, but it suffers from slow dynamic response, large overshoot, and insufficient robustness under complex operating conditions such as load abrupt changes, grid disturbances, parameter perturbations, and nonlinear loads. Furthermore, the traditional outer voltage loop primarily adjusts based on voltage error, making it difficult to reflect the overall dynamic characteristics of the rectifier from the perspective of system energy storage and power balance. Active disturbance rejection control (ADRC) can treat both internal uncertainties and external disturbances as lumped disturbances and perform real-time estimation and compensation through an extended state observer, exhibiting strong disturbance rejection and robustness. Introducing the energy balance concept into three-phase PWM rectifier control, using DC-side capacitor energy and AC-side inductor energy to construct state variables, helps improve system dynamic performance and disturbance adaptability. However, the performance of energy state disturbance rejection control largely depends on the bandwidth of the extended state observer and the proper tuning of the PI parameters in the inner current loop. Fixed parameters are difficult to adapt to different operating conditions, and there is a coupling relationship between the observer parameters and the current loop parameters. The manual tuning process is complicated and it is difficult to balance speed, stability and disturbance rejection.

[0004] Therefore, there is an urgent need for an energy state adaptive disturbance rejection control method and system for three-phase PWM rectifiers to solve the problems of insufficient dynamic performance, difficult parameter tuning, and limited robustness of traditional PI control and fixed parameter disturbance rejection control under complex operating conditions. Summary of the Invention

[0005] One of the objectives of this invention is to provide an energy state adaptive disturbance rejection control method for a three-phase PWM rectifier. By combining energy state modeling, active disturbance rejection estimation and compensation, and deep reinforcement learning parameter optimization, this method enables high dynamics, high robustness, and adaptive control of the three-phase PWM rectifier under complex operating conditions. This addresses the problems of insufficient dynamic performance, difficult parameter tuning, and limited robustness of traditional PI control and fixed parameter disturbance rejection control under complex operating conditions.

[0006] The basic solution provided by this invention is: an adaptive disturbance rejection control method for the energy state of a three-phase PWM rectifier, comprising the following: S1. Collect the three-phase AC side voltage, three-phase AC side current and DC bus voltage of the three-phase PWM rectifier, and transform the three-phase AC side voltage and three-phase AC side current to obtain the d-axis voltage component, q-axis voltage component, d-axis current component and q-axis current component in the two-phase rotating coordinate system, and set the q-axis current reference value. S2. Construct the instantaneous energy state of the three-phase PWM rectifier based on the DC bus voltage, d-axis current component, and q-axis current component. S3. Based on the DC bus voltage setpoint, current reference value, and energy storage parameters of the three-phase PWM rectifier, construct the target energy state variables; where the current reference value includes: d-axis current reference value and q-axis current reference value; S4. Input the target energy state quantity and the instantaneous energy state quantity into the energy balance active disturbance rejection controller, estimate the energy state and lumped disturbance of the three-phase PWM rectifier through the extended state observer, and generate the d-axis current reference value according to the disturbance compensation control law. S5. Input the d-axis current reference value and the q-axis current reference value into the inner loop PI controller. After decoupling calculation of the d-axis and q-axis feedforward, the modulation voltage command is obtained. Then, the switching drive signal is generated by PWM modulation to control the three-phase PWM rectifier. S6. Construct a PPO agent and use the operating state of the three-phase PWM rectifier as the state space of the PPO agent. Use the bandwidth of the extended state observer of the energy balance active disturbance rejection controller and the parameters of the current inner loop PI controller as the action space of the PPO agent. Optimize and update the bandwidth of the extended state observer and the parameters of the current inner loop PI controller through the reward function.

[0007] The second objective of this invention is to provide an energy state adaptive disturbance rejection control system for a three-phase PWM rectifier. By combining energy state modeling, active disturbance rejection estimation and compensation, and deep reinforcement learning parameter optimization, the system enables high dynamics, high robustness, and adaptive control of the three-phase PWM rectifier under complex operating conditions. This addresses the problems of insufficient dynamic performance, difficult parameter tuning, and limited robustness of traditional PI control and fixed parameter disturbance rejection control under complex operating conditions.

[0008] This invention provides a second basic solution: a three-phase PWM rectifier energy state adaptive disturbance rejection control system, comprising: The sampling module is used to collect three-phase AC side voltage, three-phase AC side current and DC bus voltage; The coordinate transformation module is used to transform the voltage and current in the three-phase stationary coordinate system into the d-axis and q-axis components in the two-phase rotating coordinate system. The energy calculation module is used to calculate instantaneous energy state quantities based on DC bus voltage, d-axis current, and q-axis current. The energy balance active disturbance rejection control module is used to estimate the lumped disturbance based on the target energy state quantity and the instantaneous energy state quantity and generate the d-axis current reference value. The inner current control module is used to generate modulation voltage commands based on the d-axis current reference value, the q-axis current reference value, and the actual current. The PWM modulation module is used to generate a switching drive signal according to the modulation voltage command; The PPO parameter optimization module is used to output the bandwidth of the extended state observer and the parameters of the current inner loop PI controller based on the operating status of the three-phase PWM rectifier.

[0009] Compared with existing technologies, this solution has the following advantages: 1. This scheme uses the energy state (target energy state quantity and instantaneous energy state quantity) of the three-phase PWM rectifier as the outer loop control object, and incorporates the DC side capacitor energy and AC side inductor energy into the control variables. Compared with the traditional voltage outer loop PI control, it can more comprehensively reflect the rectifier's energy storage state and power flow characteristics. 2. This scheme establishes an extended energy state equation based on the energy balance relationship, and unifies load changes, grid disturbances, parameter perturbations and unmodeled dynamics into a unified equivalent of lumped disturbances. The extended state observer is used for real-time estimation and compensation, which improves the system's disturbance immunity.

[0010] 3. This scheme uses a parameter adaptive optimization agent to collaboratively optimize the bandwidth of the extended state observer and the PI parameters of the current inner loop, thus avoiding the problem of insufficient adaptability of fixed parameters under varying operating conditions.

[0011] In summary, this invention uses the instantaneous energy state of a three-phase PWM rectifier as the outer loop control object, establishes an extended energy state equation through the energy balance relationship, uses an extended state observer to estimate and compensate for lumped disturbances in real time, and uses a parameter adaptive optimization agent to coordinately optimize the bandwidth of the extended state observer and the PI parameters of the current inner loop, thereby improving the dynamic response speed, DC bus voltage stability and robustness of the three-phase PWM rectifier under load changes, grid disturbances and parameter perturbations. Attached Figure Description

[0012] Figure 1 This is a flowchart illustrating an embodiment of the energy state adaptive disturbance rejection control method for a three-phase PWM rectifier according to the present invention. Figure 2 This is a topology diagram of a three-phase PWM rectifier in an embodiment of the energy state adaptive disturbance rejection control method for a three-phase PWM rectifier according to the present invention. Figure 3 This is a schematic diagram of the Actor-Critic network structure of the PPO algorithm in an embodiment of the energy state adaptive disturbance rejection control method for a three-phase PWM rectifier according to the present invention. Figure 4 This is a schematic diagram of the three-phase PWM rectifier control structure for optimizing energy balance self-disturbance rejection parameters using the PPO algorithm in an embodiment of the energy state adaptive disturbance rejection control method for a three-phase PWM rectifier according to the present invention. Figure 5 This is a PPO training reward change curve in an embodiment of the energy state adaptive disturbance rejection control method for a three-phase PWM rectifier according to the present invention; Figure 6 This is a diagram showing the three-phase voltage and current waveforms in an embodiment of the energy state adaptive disturbance rejection control method for a three-phase PWM rectifier according to the present invention. Figure 7 This is a DC bus voltage waveform diagram in an embodiment of the energy state adaptive disturbance rejection control method for a three-phase PWM rectifier according to the present invention. Figure 8 This is a graph showing the FFT analysis results of the A-phase current in an embodiment of the energy state adaptive disturbance rejection control method for a three-phase PWM rectifier according to the present invention. Detailed Implementation

[0013] The following detailed description illustrates the specific implementation method: Example 1 This embodiment provides an adaptive disturbance rejection control method for the energy state of a three-phase PWM rectifier, as shown in the attached figure. Figure 1 As shown, it includes the following steps: S1. Collect the three-phase AC side voltage, three-phase AC side current and DC bus voltage of the three-phase PWM rectifier, and transform the three-phase AC side voltage and three-phase AC side current to obtain the d-axis voltage component, q-axis voltage component, d-axis current component and q-axis current component in the two-phase rotating coordinate system. Specifically, such as Figure 2 The three-phase PWM rectifier topology shown can be represented by the AC side inductor current, DC bus voltage, and switching function in a three-phase stationary coordinate system. Since the three-phase AC quantities (including the three-phase AC side voltage and current) are time-varying, designing the controller directly in the three-phase stationary coordinate system is complex. Therefore, this embodiment uses Clark and Park transformations to convert the three-phase AC quantities to a two-phase synchronous rotating d,q coordinate system. In the d-q coordinate system, the mathematical model of a three-phase PWM rectifier can be expressed as:

[0014] in, For AC side filter inductance, The equivalent resistance on the AC side. , These are the d-axis and q-axis current components, respectively. , These are the d-axis and q-axis components of the grid voltage, respectively. The angular frequency of the power grid. , These are the equivalent control voltage components on the AC side of the rectifier. For DC-side filter capacitors, This is the DC bus voltage. Inject DC-side current into the rectifier. This is the load current.

[0015] When the d-axis of the synchronous rotating coordinate system is oriented to the grid voltage vector, it usually has the following: In this embodiment, to achieve unity power factor operation, the q-axis current reference value can be set as follows: .

[0016] S2. Construct the instantaneous energy state of the three-phase PWM rectifier based on the DC bus voltage, d-axis current component, and q-axis current component. Specifically, the energy storage components of the three-phase PWM rectifier include an AC-side filter inductor and a DC-side filter capacitor. In order to describe the dynamic behavior of the rectifier from the perspective of system energy, this embodiment constructs instantaneous energy state variables:

[0017] in, This is the energy coefficient, and its value depends on the power conservation form used in the coordinate transformation. When using the power-invariant Park transformation, A value of 1 can be taken; when using the amplitude-invariant Park transform, Two-thirds is acceptable. In engineering implementation, Pre-set according to the definition of the coordinate transformation module in the controller.

[0018] Through the aforementioned energy state variables, the DC bus voltage change, inductor current dynamics, and AC side energy exchange are unified into the same control variable, which helps to improve the outer loop control's ability to describe the overall dynamics of the system.

[0019] S3. Based on the DC bus voltage setpoint, current reference value, and energy storage parameters of the three-phase PWM rectifier, construct the target energy state variables; where the current reference value includes: d-axis current reference value and q-axis current reference value; Specifically, based on the DC bus voltage setpoint d-axis current reference value and q-axis current reference value Construct the target energy state variables:

[0020] in, For the target energy state quantity, The DC bus voltage setpoint. This is the reference value for the d-axis current. This is the reference value for the q-axis current; under unity power factor rectification operation mode, The d-axis current reference value can be calculated using the power balance relationship to obtain the d-axis current feedforward value as the initial value. This value can be output by the energy balance active disturbance rejection controller, or it can be combined with the power feedforward estimate for limiting correction. For example, when the grid d-axis voltage is... At that time, the d-axis current feedforward corresponding to the DC-side power demand can be expressed as:

[0021] in, The DC-side load power or power setpoint is used; the d-axis current feedforward can be combined with the output of the energy balance active disturbance rejection controller to form the final d-axis current reference value.

[0022] S4. Input the target energy state quantity and the instantaneous energy state quantity into the energy balance active disturbance rejection controller, estimate the energy state and lumped disturbance of the three-phase PWM rectifier through the extended state observer, and generate the d-axis current reference value according to the disturbance compensation control law. Specifically, based on the power balance relationship of a three-phase PWM rectifier, the rectifier's energy change rate can be expressed as the difference between input power, output power, and loss power. Considering external disturbances, internal losses, parameter uncertainties, and unmodeled dynamics, the energy balance relationship of a three-phase PWM rectifier can be uniformly expressed as:

[0023] in, The instantaneous rate of change of energy. For equivalent control gain, For energy control input, The lumped disturbances include, but are not limited to: grid voltage fluctuations, load abrupt changes, negative impedance characteristics of constant power loads on the DC side, deviations in filter inductor parameters, deviations in filter capacitor parameters, switching device losses, and sampling noise. Since each of these disturbances is difficult to measure or model precisely in practice, this solution uses an extended state observer to uniformly estimate them as lumped disturbances. It eliminates the need to acquire each disturbance component separately, thus achieving unified compensation for multi-source uncertainties. In actual operation, ESO uniformly estimates all the above uncertainties (external disturbances + internal parameter perturbations + unmodeled dynamics) into a real-time value of a lumped disturbance, without distinguishing between the components. ESO only needs the system's input (control quantity) and output (energy state) to estimate the lumped disturbance and feedforward compensation in real time.

[0024] To estimate the energy state and lumped disturbances in real time, this embodiment uses a second-order linear extended state observer:

[0025] in, Instantaneous energy state quantity The estimated value, For aggregated disturbance The estimated value, , For the observer gain, a bandwidth parameterization tuning method is used:

[0026] in, To expand the bandwidth of the state observer.

[0027] Based on the extended state observer, the disturbance compensation control law of the energy balance active disturbance rejection controller is:

[0028] in, For energy error feedback gain, This represents the rate of change of the target energy state variable. When the target energy is given as a constant value or changes slowly... It can be obtained by taking zero or through differential filtering. This control law is obtained through... The term compensates for lumped disturbances, making the controlled object approximately transform into an integral element, thereby achieving rapid tracking of the target energy by the instantaneous energy.

[0029] Energy control input It can be further converted into The d-axis current reference value, for example under grid voltage-oriented control, can be obtained as follows:

[0030] in, For the amplitude limiting function, , These represent the lower and upper limits of the allowable d-axis current, respectively. Limiting measures prevent overcurrent and meet the safe operation requirements of power devices.

[0031] S5. Input the d-axis current reference value and the q-axis current reference value into the inner loop PI controller. After decoupling calculation of the d-axis and q-axis feedforward, the modulation voltage command is obtained. Then, the switching drive signal is generated by PWM modulation to control the three-phase PWM rectifier. Specifically, the current inner-loop PI controller adopts d-axis and q-axis feedforward decoupling control. Its modulation voltage command includes d-axis modulation voltage command and q-axis modulation voltage command. The grid voltage feedforward term and cross-coupling compensation term are added to the d-axis modulation voltage command and q-axis modulation voltage command respectively to eliminate the coupling effect between the d-axis and q-axis currents.

[0032] The inner current loop uses a PI controller, and the d-axis and q-axis current errors are respectively:

[0033] The output of the inner current loop PI controller is:

[0034] in, , These are the proportional gain and integral gain of the inner current loop, respectively. This refers to the d-axis modulation voltage control quantity (i.e., the d-axis voltage control quantity on the AC side of the rectifier / the d-axis component of the modulation wave). It is the q-axis modulation voltage control quantity (i.e., the q-axis voltage control quantity on the AC side of the rectifier / the q-axis component of the modulation wave).

[0035] The inner current loop PI controller includes a d-axis PI controller and a q-axis PI controller, which together constitute the inner current loop PI controller. The d-axis PI controller adjusts the d-axis current component, and the q-axis PI controller adjusts the q-axis current component.

[0036] To eliminate d-axis and q-axis coupling and improve current response speed, grid voltage feedforward and cross-coupling compensation are added to obtain the modulation voltage command:

[0037] Then , The three-phase modulation reference quantity is obtained through inverse Park transformation, and six-channel switch drive signals are generated using space vector PWM or sinusoidal PWM to control the switching devices of the three-phase PWM rectifier to turn on and off.

[0038] S6. Construct a PPO agent, and use the operating state of the three-phase PWM rectifier as the state space of the PPO agent, and use the observer bandwidth of the energy balance active disturbance rejection controller and the current inner loop PI controller parameters as the action space of the PPO agent. Optimize and update the observer bandwidth and PI controller parameters through the reward function. Specifically, to address the insufficient adaptability of the energy balance active disturbance rejection controller and the current inner loop PI controller due to fixed parameters, the PPO algorithm is introduced to adaptively optimize key control parameters. These key control parameters include the extended state observer bandwidth, the proportional gain and integral gain of the current inner loop PI controller, which are usually fixed values ​​in existing technologies. This scheme uses PPO reinforcement learning to adaptively optimize them online. The state space is as follows: The PPO agent state space is used to characterize the current operating state of the three-phase PWM rectifier. The state vector in this embodiment is:

[0039] That is, the state space includes DC bus voltage, DC bus voltage error, d-axis current component, d-axis current error (the difference between the d-axis current reference value and the actual value), q-axis current component and q-axis current error (the difference between the q-axis current reference value and the actual value). The state space can reflect the main dynamic characteristics of the voltage outer loop and the current inner loop. in, Let be the state vector observed by the PPO agent at the current moment.

[0040] The motion space is as follows: The PPO agent's action space is used to output the control parameters to be optimized. In this embodiment, the action vector is:

[0041] Or it may include the incremental form of the above parameters; wherein, To expand the bandwidth of the state observer, , These are the proportional gain and integral gain of the current inner-loop PI controller, respectively.

[0042] The reward function is as follows: To guide the PPO agent in learning a fast, stable, and low-overshoot control strategy, the reward function includes at least a DC bus voltage error penalty term, a current error penalty term, an overshoot penalty term, and a control parameter change penalty term. This embodiment designs the following reward function:

[0043] in, For instant rewards, to The positive weighting coefficients are... This represents the change in action between adjacent time points. This is a metric for evaluating DC bus voltage overshoot. A positive reward is given when both DC bus voltage error and current error are within acceptable limits. This encourages agents to learn control strategies with zero steady-state error, low overshoot, and fast convergence.

[0044] The PPO network structure and training are as follows: like Figure 3 As shown, the PPO agent employs an Actor-Critic network structure. The Actor network outputs actions based on the current state vector, i.e., control parameters or control parameter increments; the Critic network estimates the current state value, used to evaluate the quality of the Actor network's output actions. The PPO agent updates the Actor network by pruning the surrogate objective function and updates the Critic network using the mean squared error loss function.

[0045] The PPO algorithm (PPO agents are decision entities built based on the PPO algorithm) updates the policy network by pruning the agent's objective function, and its form is as follows:

[0046] in, For Actor network parameters, This is the cutting factor. This is the estimated value of the dominance function. The probability ratio between the old and new strategies:

[0047] The Critic network updates by minimizing the mean square error of the value function:

[0048] in, For Critic network parameters, This is the estimated value of the current state. For target value.

[0049] The PPO algorithm training process is as follows: Step 1: Randomly initialize the Actor network parameters and Critic network parameters, and initialize the old policy network parameters; where the old policy network parameters... The parameters of the Actor network (neural network weights and bias parameters) before the last parameter update are used to calculate the probability ratio between the old and new policies. In this embodiment, the Old Policy Network refers to the Actor network before the last parameter update. This refers to the set of neural network weights and bias parameters of the Actor neural network in the previous training iteration; PPO calculates the probability ratio of the new and old policies. To perform cropping and updates, among which For the current strategy (new parameters), The old strategy (parameters before the update); It is copied from the current Actor network before each PPO update and remains unchanged during an update. Step 2: The PPO agent interacts with the three-phase PWM rectifier simulation environment or hardware in the environment to collect empirical data consisting of state, action, reward and next state. Step 3: Calculate the advantage function and target value based on the collected experience data; The advantage function is typically calculated using generalized advantage estimation (GAE):

[0050] in, , It is a discount factor. It is the GAE coefficient. It is the clipping hyperparameter.

[0051] The target value, i.e., the advantage function plus the current Critic network's state. Value estimation:

[0052] Step 4: Update the Actor network using the pruning agent objective function and update the Critic network using the value loss function; Step 5: Synchronize the current strategy with the old strategy, clear the experience pool, and proceed to the next round of training; Step six: When the average reward reaches a preset threshold or the number of training iterations reaches the upper limit, output the trained policy network; the overall control structure is as follows: Figure 4 As shown.

[0053] During online operation, the state variables of the three-phase PWM rectifier are acquired in real time and input into the trained PPO agent. The PPO agent outputs the observer bandwidth and PI parameters based on the current state. Using the energy state of the three-phase PWM rectifier (target energy state and instantaneous energy state) as the outer loop control object, the DC-side capacitor energy and AC-side inductor energy are uniformly incorporated into the control variables. Compared with traditional voltage outer loop PI control, this approach can more comprehensively reflect the rectifier's energy storage state and power flow characteristics. The DC-side capacitor energy is the instantaneous energy state variable. The first item This refers to the electric field energy stored in the DC bus filter capacitor; it is calculated in S2 by collecting the DC bus voltage and the known DC side filter capacitor; the AC side inductance energy is an instantaneous energy state quantity. The second item This refers to the magnetic field energy stored in the AC-side filter inductor; it is calculated in S2 using the d-axis current component, the q-axis current component, and the known AC-side filter inductor; both are uniformly constructed in S2 as an instantaneous energy state quantity, which serves as the outer loop control object.

[0054] After the controller performs amplitude limiting, filtering, and smoothing on the parameters, it inputs them into the energy balance active disturbance rejection controller and the current inner-loop PI controller. The training process is as follows: Figure 5 As shown, this demonstrates that the above method can converge.

[0055] The DC bus voltage, three-phase voltage, and current waveforms are shown below. Figure 6 and Figure 7 This indicates that the method can quickly stabilize the output voltage and has good immunity to disturbances. Figure 8 It can be seen that the THD of the input current is 2.86%, indicating that this method is relatively good.

[0056] This embodiment also provides a three-phase PWM rectifier energy state adaptive disturbance rejection control system, including: The sampling module is used to collect three-phase AC side voltage, three-phase AC side current and DC bus voltage; The coordinate transformation module is used to transform the voltage and current in the three-phase stationary coordinate system into the d-axis and q-axis components in the two-phase rotating coordinate system. The energy calculation module is used to calculate instantaneous energy state quantities based on DC bus voltage, d-axis current, and q-axis current. The energy balance active disturbance rejection control module is used to estimate the lumped disturbance based on the target energy state quantity and the instantaneous energy state quantity and generate the d-axis current reference value. The current inner loop control module is used to generate modulation voltage commands based on the d-axis current reference value, the q-axis current reference value, and the actual current. The PWM modulation module is used to generate a switching drive signal according to the modulation voltage command; The PPO parameter optimization module is used to output the bandwidth and current inner loop PI parameters of the extended state observer based on the operating status of the three-phase PWM rectifier.

[0057] The above descriptions are merely embodiments of the present invention. Commonly known structures and characteristics are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A method for adaptive energy state disturbance rejection control of a three-phase PWM rectifier, characterized in that, Includes the following steps: S1. Collect the three-phase AC side voltage, three-phase AC side current and DC bus voltage of the three-phase PWM rectifier, and transform the three-phase AC side voltage and three-phase AC side current to obtain the d-axis voltage component, q-axis voltage component, d-axis current component and q-axis current component in the two-phase rotating coordinate system, and set the q-axis current reference value. S2. Construct the instantaneous energy state of the three-phase PWM rectifier based on the DC bus voltage, d-axis current component, and q-axis current component. S3. Construct the target energy state quantity based on the DC bus voltage setpoint, current reference value, and energy storage parameters of the three-phase PWM rectifier; The current reference values ​​include: d-axis current reference value and q-axis current reference value; S4. Input the target energy state quantity and the instantaneous energy state quantity into the energy balance active disturbance rejection controller, estimate the energy state and lumped disturbance of the three-phase PWM rectifier through the extended state observer, and generate the d-axis current reference value according to the disturbance compensation control law. S5. Input the d-axis current reference value and the q-axis current reference value into the inner loop PI controller. After decoupling calculation of the d-axis and q-axis feedforward, the modulation voltage command is obtained. Then, the switching drive signal is generated by PWM modulation to control the three-phase PWM rectifier. S6. Construct a PPO agent, and use the operating state of the three-phase PWM rectifier as the state space of the PPO agent, and use the observer bandwidth of the energy balance active disturbance rejection controller and the current inner loop PI controller parameters as the action space of the PPO agent. Optimize and update the observer bandwidth and PI controller parameters through the reward function.

2. The three-phase PWM rectifier energy state adaptive disturbance rejection control method according to claim 1, characterized in that, The coordinate transformation employs Clark transformation and Park transformation.

3. The three-phase PWM rectifier energy state adaptive disturbance rejection control method according to claim 1, characterized in that, The instantaneous energy state quantity is: in, This refers to the instantaneous energy state parameters of a three-phase PWM rectifier. For DC-side filter capacitors, This is the DC bus voltage. For AC side filter inductance, and These are the d-axis current component and the q-axis current component, respectively. The energy coefficient is related to the form of coordinate transformation.

4. The three-phase PWM rectifier energy state adaptive disturbance rejection control method according to claim 3, characterized in that, The target energy state quantity is: in, For the target energy state quantity, The DC bus voltage setpoint. The d-axis current reference value is preset with an initial value. This is the reference value for the q-axis current, with a preset initial value.

5. The three-phase PWM rectifier energy state adaptive disturbance rejection control method according to claim 4, characterized in that, The energy balance control system of the three-phase PWM rectifier in the energy balance active disturbance rejection controller is as follows: in, The instantaneous rate of change of energy. For equivalent control gain, For energy control input, This is a lumped disturbance.

6. The three-phase PWM rectifier energy state adaptive disturbance rejection control method according to claim 5, characterized in that, The extended state observer is a second-order linear extended state observer: in, Instantaneous energy state quantity The estimated value, For aggregated disturbance The estimated value, , Let the observer gain be and satisfy: in, To expand the bandwidth of the state observer.

7. The three-phase PWM rectifier energy state adaptive disturbance rejection control method according to claim 6, characterized in that, The disturbance compensation control law of the energy balance active disturbance rejection controller is: in, For energy error feedback gain, The target energy state variable is the rate of change; the control law compensates for the estimated value of the lumped disturbance. , making the instantaneous energy state quantity Tracking target energy state quantity .

8. The three-phase PWM rectifier energy state adaptive disturbance rejection control method according to claim 7, characterized in that, The d-axis current reference value is: in, For the amplitude limiting function, , These represent the lower and upper limits of the allowable d-axis current, respectively.

9. The three-phase PWM rectifier energy state adaptive disturbance rejection control method according to claim 1, characterized in that, The current inner loop PI controller adopts d-axis and q-axis feedforward decoupling control. Its control quantities include d-axis modulation voltage command and q-axis modulation voltage command. The grid voltage feedforward term and cross-coupling compensation term are added to the d-axis modulation voltage command and q-axis modulation voltage command respectively to eliminate the coupling effect between the d-axis and q-axis currents.

10. The three-phase PWM rectifier energy state adaptive disturbance rejection control method according to claim 9, characterized in that, The d-axis and q-axis current errors in S5 are respectively: The output of the inner current loop PI controller is: in, , These are the proportional gain and integral gain of the inner current loop, respectively. For d-axis modulated voltage control quantity, This is the q-axis modulated voltage control quantity; By incorporating grid voltage feedforward and cross-coupling compensation, the modulation voltage command is obtained: Will , The three-phase modulation reference quantity is obtained through inverse Park transformation, and the switching drive signal is generated by space vector PWM or sinusoidal PWM to control the switching devices of the three-phase PWM rectifier to turn on and off.

11. The three-phase PWM rectifier energy state adaptive disturbance rejection control method according to claim 9, characterized in that, The state space of the PPO agent includes at least: DC bus voltage, DC bus voltage error, d-axis current component, d-axis current component error, q-axis current component, and q-axis current component error. in, Let be the state vector observed by the PPO agent at the current moment.

12. The three-phase PWM rectifier energy state adaptive disturbance rejection control method according to claim 9, characterized in that, The action space of the PPO agent includes at least: Or it may include the incremental form of the above parameters; wherein, To expand the bandwidth of the state observer, , These are the proportional gain and integral gain of the current inner-loop PI controller, respectively.

13. The three-phase PWM rectifier energy state adaptive disturbance rejection control method according to claim 9, characterized in that, The reward function of the PPO agent includes at least a DC bus voltage error penalty term, a current error penalty term, an overshoot penalty term, and a control parameter change penalty term: in, For instant rewards, to The positive weighting coefficients are... This represents the change in action between adjacent time points. This is a metric for evaluating DC bus voltage overshoot.

14. The three-phase PWM rectifier energy state adaptive disturbance rejection control method according to claim 1, characterized in that, The PPO agent adopts an Actor-Critic network structure, where the Actor network outputs actions based on the current state vector; the Critic network is used to estimate the value of the current state; the PPO agent updates the Actor network by pruning the surrogate objective function and updates the Critic network by the mean squared error loss function.

15. The three-phase PWM rectifier energy state adaptive disturbance rejection control method according to claim 9, characterized in that, The optimization of control parameters by the PPO agent includes offline training and online invocation. In the offline training phase, parameter policy learning is completed in a simulation environment or hardware-in-the-environment environment. In the online operation phase, the optimized observer bandwidth and PI controller parameters are output according to the current state, and the parameters are input to the energy balance active disturbance rejection controller and the current inner loop PI controller after being subjected to amplitude limiting processing.

16. The three-phase PWM rectifier energy state adaptive disturbance rejection control method according to claim 9, characterized in that, The three-phase PWM rectifier adaptively adjusts the observer bandwidth and PI controller parameters through the PPO algorithm under conditions such as load abrupt changes, grid voltage fluctuations, inductor parameter perturbation, capacitor parameter perturbation, or constant power load connection on the DC side.

17. A three-phase PWM rectifier energy state adaptive disturbance rejection control system, characterized in that, include: The sampling module is used to collect three-phase AC side voltage, three-phase AC side current and DC bus voltage; The coordinate transformation module is used to transform the voltage and current in the three-phase stationary coordinate system into the d-axis and q-axis components in the two-phase rotating coordinate system. The energy calculation module is used to calculate instantaneous energy state quantities based on DC bus voltage, d-axis current, and q-axis current. The energy balance active disturbance rejection control module is used to estimate the lumped disturbance based on the target energy state quantity and the instantaneous energy state quantity and generate the d-axis current reference value. The current inner loop control module is used to generate modulation voltage commands based on the d-axis current reference value, the q-axis current reference value, and the actual current. The PWM modulation module is used to generate a switching drive signal according to the modulation voltage command; The PPO parameter optimization module is used to output the bandwidth and current inner loop PI parameters of the extended state observer based on the operating status of the three-phase PWM rectifier.

Images