Optical storage micro-grid lead correction active disturbance rejection control method and system

Abstract

The invention, which relates to the electric energy quality control field, discloses an optical storage micro-grid lead correction active disturbance rejection control system comprising a reduced order observer that generates an observation value of total disturbance of the system based on differential information of a voltage signal; the phase compensator is used for performing phase lead correction on the observation signal; the feedback controller is used for generating a control signal for the energy storage converter; the parameter optimizing unit is configured to obtain voltage state information of the system; outputting an optimization action for adjusting adjustable parameters of the reduced order observer, the phase compensator and / or the feedback controller; and according to a reward function taking the voltage stability as a target, evaluating an action effect and optimizing a strategy through learning. According to the invention, collaborative optimization and online adaptive adjustment of parameters of the phase compensation network and the active-disturbance-rejection controller are realized, and the voltage stability and the electric energy quality of the optical storage DC micro-grid under complex disturbance are effectively improved.

Description

Technical Field

[0001] This invention belongs to the technical field, and in particular relates to a method and system for advance correction and active disturbance rejection control of photovoltaic-storage microgrids. Background Technology

[0002] Photovoltaic-storage DC microgrids are an important form of absorbing distributed energy. To improve the stability and power quality of photovoltaic-storage DC microgrids, control strategies need to be applied to the energy storage converter to maintain the stability of the DC bus voltage. However, the existence of uncertain time-varying disturbances such as temperature changes, load switching, and random noise may cause large overshoot and undesirable oscillations in the bus voltage during the control process.

[0003] Linear active disturbance rejection control (ADRC), as a mature power electronic converter control technology, is widely used in various fields of power quality improvement and integrated energy system automatic control in new power systems. In existing technologies, the linear extended state observer, the core component of the linear ADRC, suffers from significant phase lag in observing time-varying disturbance signals, affecting the dynamic response characteristics and tracking accuracy of the control system and making it difficult to cope with various complex internal and external disturbances that photovoltaic-storage DC microgrids may face. Therefore, optimizing the response speed of the ADRC control system to fast time-varying disturbances and reducing disturbance observation lag is crucial for improving the power quality of photovoltaic-storage DC microgrids. Based on this, this invention designs a lead-compensation ADRC control method and system for photovoltaic-storage microgrids. Summary of the Invention

[0004] The purpose of this invention is to solve the problems in the prior art, and to propose a method and system for advance correction and active disturbance rejection control of photovoltaic-storage microgrids.

[0005] This invention first discloses a lead-correction active disturbance rejection control system for photovoltaic-storage microgrids, comprising: A reduced-order observer is configured to receive the DC bus voltage signal of the energy storage converter and generate an observation of the total system disturbance based on the differential information of the voltage signal. A phase compensator, connected to the perturbation observation channel of the reduced-order observer, is used to perform phase lead correction on the observation signal; The feedback controller is configured to generate control signals for the energy storage converter based on the deviation between the reference voltage and the DC bus voltage, and the observed signal corrected by the phase compensator. The parameter optimization unit, built upon a reinforcement learning framework, is configured as follows: It interacts with the microgrid operating environment to obtain the system's voltage status information; The output is an optimized action used to adjust the adjustable parameters of the reduced-order observer, the phase compensator, and / or the feedback controller; Based on a reward function aimed at voltage stability, the system evaluates the effectiveness of actions and learns optimization strategies to achieve adaptive tuning of system parameters.

[0006] In the above system, the reduced-order observer reduces the number of state variables to be observed by directly acquiring the derivative of the DC bus voltage signal as a feedback quantity.

[0007] In the above system, the phase compensator is configured to provide a leading phase angle in its frequency response to compensate for the phase lag caused by the reduced-order observer.

[0008] In the above system, the phase compensator is a first-order lead compensation network.

[0009] In the above system, the feedback controller is a linear state feedback controller based on the voltage deviation and its derivative signal.

[0010] In the above system, the reinforcement learning algorithm used by the parameter optimization unit is the actor-critic algorithm.

[0011] In the above system, the actor-critic algorithm employs a deep deterministic strategy gradient algorithm.

[0012] In the above system, the state space of the parameter optimization unit includes the DC bus voltage, its differential value, and the voltage deviation and its differential value; its action space includes the time constant and scale factor of the phase compensator, and / or the bandwidth of the reduced-order observer.

[0013] Secondly, the present invention provides a method for advance correction and active disturbance rejection control of photovoltaic-storage microgrids, which employs the above-mentioned system.

[0014] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is used to implement the above-described active disturbance rejection control system.

[0015] Through the above steps, this invention achieves coordinated optimization and online adaptive adjustment of the parameters of the phase compensation network and the active disturbance rejection controller, effectively improving the voltage stability and dynamic quality of the photovoltaic-storage DC microgrid under complex disturbances. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of a photovoltaic-storage DC microgrid structure.

[0017] Figure 2 This is a block diagram of a phase-compensated reduced-order active disturbance rejection control system.

[0018] Figure 3This is a schematic diagram of the parameter optimization model for a phase-compensated reduced-order active disturbance rejection control system based on a deep deterministic strategy gradient algorithm provided by the present invention.

[0019] Figure 4 The diagram shows the structure of the deep neural network for the deep deterministic policy gradient algorithm provided by this invention.

[0020] Figure 5 The simulation curves of bus voltage under the influence of noise are shown for various models.

[0021] Figure 6 The simulation curves of capacitor power under noise influence are shown for various models.

[0022] Figure 7 The output voltage curves of various models after being disturbed by an applied ramp disturbance are shown.

[0023] Figure 8 The output voltage curves of various models after being disturbed by an applied sinusoidal disturbance are shown.

[0024] Figure 9 The figures show the system output voltage and dynamic performance indicators for various models under load power change conditions.

[0025] Figure 10 This is a chart analyzing the evaluation metrics for various models. Detailed Implementation

[0026] To facilitate understanding of this application and to make the aforementioned objectives, features, and advantages of this application more apparent, a detailed description of specific embodiments of this application is provided below in conjunction with the accompanying drawings. Numerous specific details are set forth in the following description to provide a thorough understanding of this application, and preferred embodiments are shown in the accompanying drawings. However, this application can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the disclosure of this application. This application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of this application; therefore, this application is not limited to the specific embodiments disclosed below. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified. In the description of this application, "several" means at least one, such as one, two, etc., unless otherwise explicitly specified. It should be noted that when an element is referred to as being "fixed to" another element, it can be directly attached to the other element or there may be an intervening element. When an element is referred to as being "connected to" another element, it can be directly connected to the other element or there may be an intervening element. The terms "vertical," "horizontal," "left," "right," and similar expressions used herein are for illustrative purposes only and do not represent the only possible implementations. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is only for describing particular implementations and is not intended to limit the scope of this application. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0027] Reference Figures 1-4 This invention provides a proactive disturbance rejection control system for photovoltaic-storage microgrids, comprising: A reduced-order observer is configured to receive the DC bus voltage signal of the energy storage converter and generate an observation of the total system disturbance based on the differential information of the voltage signal. A phase compensator, connected to the perturbation observation channel of the reduced-order observer, is used to perform phase lead correction on the observation signal; The feedback controller is configured to generate control signals for the energy storage converter based on the deviation between the reference voltage and the DC bus voltage, and the observed signal corrected by the phase compensator. The parameter optimization unit, built upon a reinforcement learning framework, is configured as follows: It interacts with the microgrid operating environment to obtain the system's voltage status information; The output is an optimized action used to adjust the adjustable parameters of the reduced-order observer, the phase compensator, and / or the feedback controller; Based on a reward function aimed at voltage stability, the system evaluates the effectiveness of actions and learns optimization strategies to achieve adaptive tuning of system parameters.

[0028] A photovoltaic-storage DC microgrid typically includes photovoltaic power generation units, energy storage units, loads, and corresponding power electronic converters. This invention primarily focuses on improving the control system of the energy storage converter. Specifically, since the phase lag becomes more severe with higher order in current linear extended state observers, this application reconstructs the traditional linear extended state observer into a reduced-order linear extended state observer. The resulting reduced-order extended state observer is as follows:

[0029] In the formula, , This represents the derivative of the output voltage after order reduction and the observed total disturbance. , Indicates the reduced-order observer gain. This indicates the actual output voltage of the energy storage converter.

[0030] The phase compensator is connected in series in the perturbation observation channel of the reduced-order observer to provide the required phase lead angle. In a preferred embodiment, the phase compensator employs a first-order lead correction network with the following transfer function:

[0031] In the formula, Represents the complex variables in the transfer function. Represents the time constant. This represents the scale factor, which has a value between 0 and 1. By properly configuring the parameters, the inherent phase lag of the reduced-order observer can be effectively compensated, and the adverse effects of high-frequency measurement noise on the observation accuracy can be improved.

[0032] Then, a phase-compensated reduced-order observer is constructed, with the following equation:

[0033] in, , This represents the output differential term and the estimated total disturbance after correction. , This represents the corrected observer gain.

[0034] The feedback controller employs linear state feedback control based on voltage deviation and its derivative signal. Specifically, the control law can be expressed as:

[0035] In the formula, Indicates the proportional gain of the controller; Indicates the differential gain of the controller; This indicates the controller gain.

[0036] The parameter optimization unit is built on a reinforcement learning framework. Its core is to interact with the microgrid operating environment and learn the optimal parameter tuning strategy through continuous trial and error.

[0037] In one feasible embodiment, the reinforcement learning algorithm used by the parameter optimization unit is the actor-critic algorithm, which enables the agent to continuously interact with the environment and explore the optimal policy parameter space of the transformer. The actor-critic algorithm can be the Deep Deterministic Policy Gradient Algorithm (DDPG), which is a deterministic policy gradient method based on deep neural networks and is used to handle continuous action space problems. Its specific principle is as follows: DDPG comprises an actor network, an evaluation network, a target actor network, and a target evaluation network, and its outputs are respectively... , , , ,in, , , , These are the weight coefficients of the neural network. The actor network is used to determine the agent's current state. Actions taken at that time To enhance the ability to explore non-boundary policies, noise is typically added to the actor network, as shown in the following equation:

[0038] In the formula: It is a noise function.

[0039] The critic network is used to predict state-action values. When the agent is in the next state At that time, the target actor network will issue an action. The target evaluation network is then used to calculate state-action space values. This value can be obtained using the following Bellman equation.

[0040]

[0041] In the formula: It is the target value; It is a discount factor; It is a reward value; It is the estimated value function of the target evaluation network; It is the estimation policy function of the target actor network.

[0042] The DDPG algorithm uses an experience replay mechanism to break the association between interaction sequences, and for state transition values ​​already stored in the experience pool... Several sets of data can be randomly selected to train the agent, and the parameters in the network will be updated synchronously. The evaluation of the updated network parameters should ensure that the loss function... Minimum, its expression is

[0043] In the formula: N is the number of training rounds.

[0044] The update of the actor network parameters should enable the output The expected value is maximized, and its expression is:

[0045] In the formula: It is the objective function; Find the gradient.

[0046] Two target networks are used to improve the stability of the DDPG algorithm, and their parameter iterations follow the following formula:

[0047] In the formula: It is a soft update parameter and is less than 1.

[0048] In this scheme, the active disturbance rejection control process exhibits Markov property. The parameter optimization model of the reduced-order active disturbance rejection controller based on the DDPG algorithm is shown in the figure. In the photovoltaic-storage DC microgrid environment, the agent can be trained, and the magnitude and direction of parameter adjustment can be improved based on the reward value of its actions until the training process converges.

[0049] The state space of the parameter optimization unit is defined as follows:

[0050] in, These are the output voltage of the energy storage converter and its derivative, respectively. These are the output voltage deviation and its derivative, respectively.

[0051] To enable the agent to train an effective combination of control parameters in a shorter time, the decomposition space dimension is compressed to improve exploration efficiency while maintaining the functional integrity of the action space. After measuring the converter output voltage and obtaining the voltage deviation, the action space is... Exploration and adaptive adjustment are carried out, among which, It is the scale factor. It is the observer bandwidth, used to improve training convergence speed.

[0052] Specifically, to maintain DC bus voltage stability, the converter output voltage deviation *e* needs to be minimized. This is achieved by evaluating the training effect using the new state values ​​generated after each training round of agent actions, and then adjusting the reward parameter accordingly. , and penalty items The sum as a reward function Its formula can be expressed as:

[0053] When constructing the DDPG network described above, the actor network input layer can be set to contain 4 units; To extract the deep-seated features of complex disturbances in the microgrid layer by layer, the network model constructed is shown in the figure. The actor network consists of an input layer with 4 units, a pre-hidden layer with 128 neurons, and a post-hidden layer with 4 neurons, using functions to enhance expressive power; the output layer contains 2 units, namely... and An activation function is selected to generate actions ranging from -1 to 1. The network's state path and action path are evaluated, each containing two hidden layers and one hidden layer, with 64 neurons in each hidden layer. The features extracted from the two paths are superimposed, and the state-action value is calculated through a fully connected layer. The output layer directly estimates the Q-function, thus eliminating the need for an activation function.

[0054] Stability analysis is performed on the phase-compensated reduced-order active disturbance rejection controller (ADRC) model optimized using a deep deterministic gradient algorithm proposed in this invention. Since the model is based on a parameter optimization adaptive strategy within the ADRC framework, its core architecture remains unchanged; therefore, the stability analysis of linear ADRC can be referenced. The disturbance observation transfer function of the phase-compensated reduced-order ADRC is also discussed. :

[0055] In the formula This refers to the aforementioned disturbance. After performing the Laplace transform, the closed-loop characteristic equation is obtained from the above equation:

[0056] In the formula: .

[0057] The Hurwitz determinant is constructed as follows:

[0058] Accordingly:

[0059] As can be seen from the formula, , , All values ​​are positive. According to the Hurwitz stability criterion, it can be seen that the observation process of the linear extended state observer after the advance order reduction converges for the total disturbance.

[0060] Let tracking error , In the formula, and These represent the reference input and its differential term of the active disturbance rejection controller, respectively. and These represent the actual output of the active disturbance rejection controller and its derivative term, respectively, corresponding to the description at the end of page 6. and Therefore, we can conclude that:

[0061] In the formula: ; Both are estimation errors. Specifically, the former is the estimation error of the output voltage, and the latter is the estimation error of the disturbance.

[0062] make:

[0063]

[0064] The above formula can then be written as:

[0065] Theorem: Assuming the estimation error of the linear extended state observer after lead reduction is bounded, then there exists a constant positive number... and This makes the tracking error of the control system bounded.

[0066] Proof: Solving the above equation, we obtain the dynamic response as follows:

[0067] Due to the characteristic polynomial:

[0068] The matrix in the formula Represents the identity matrix.

[0069] Therefore, it exists. and , making Established, and .

[0070] Since the eigenvalues ​​are distinct, we can Transform into diagonal canonical form, i.e., there exists a nonsingular matrix. , so that:

[0071] Therefore, we can conclude that:

[0072] in, , The following equation holds true:

[0073] In the formula, Since it is a constant, we know that:

[0074] Similarly, we can obtain:

[0075] because:

[0076] Combining these, we can obtain:

[0077] In the above formula, The value indicates that the error converges. Thus, the linear extended state observer after the lead reduction is stable with bounded input and bounded output. Since the change of the control parameters is equivalent to scaling the system time scale, it will not affect the system stability. Therefore, the stability of this model is proven.

[0078] This paper uses the Matlab 2023a platform for simulation. A photovoltaic-storage DC microgrid system is built in Simulink as the training environment, and its parameter settings are shown in the table below:

[0079] By applying different forms of disturbance in a microgrid, simulations were performed on the linear active disturbance rejection controller (RLADRC) with a reduced-order observer, the phase-compensated reduced-order active disturbance rejection controller (PLADRC) with a phase-compensated reduced-order active disturbance rejection controller (DDPG-PLADRC) optimized under the action of a deep deterministic policy gradient algorithm.

[0080] Based on simulation experiments, the simulated curve of the bus voltage under the influence of noise is as follows: Figure 5 As shown, the simulated capacitor power curve under the influence of noise is as follows: Figure 6 As shown, the output voltage curve of the system after the applied ramp disturbance is as follows: Figure 7 As shown, the output voltage curve of the system after the applied sinusoidal disturbance is as follows: Figure 8 As shown, the system output voltage and dynamic performance indicators under load power sudden change conditions are as follows: Figure 9 As shown, the evaluation index analysis of the three models is as follows: Figure 10 As shown, the index analysis selects the integration of absolute error, Mean absolute error ) and mean squared error Analysis of the indicators demonstrates that the proposed scheme can effectively improve the convergence speed and accuracy of disturbance estimation. Simulation comparisons of the tracking performance and disturbance rejection capabilities of the control strategy before and after the improvement under different operating conditions show that the proposed strategy can accurately observe and rapidly suppress various disturbances in the photovoltaic-storage DC microgrid.

[0081] As is known from common technical knowledge, this invention can be implemented through other embodiments that do not depart from its spirit or essential characteristics. Therefore, the disclosed embodiments described above are merely illustrative in all respects and are not the only ones. All modifications within the scope of this invention or its equivalents are included in this invention.

Claims

1. A light storage micro-grid advanced correction active disturbance rejection control system, characterized in that, The system comprises: a reduced-order observer configured to receive a DC bus voltage signal of an energy storage converter and generate an observation value of total disturbance of the system based on differential information of the voltage signal; a phase compensator connected to a disturbance observation channel of the reduced-order observer, configured to perform phase lead correction on the reduced-order observation signal; a feedback controller configured to generate a control signal of the energy storage converter based on a deviation of a reference voltage from the DC bus voltage and the observation signal corrected by the phase compensator; a parameter optimization unit constructed based on a deep reinforcement learning framework and configured to: interact with the microgrid operating environment to obtain voltage state information of the system; output adaptive optimization of adjustable parameters of the reduced-order observer, the phase compensator and / or the feedback controller; evaluate the effect of actions and optimize the strategy through learning according to a reward function targeting voltage stability, to achieve adaptive setting of system parameters.

2. The system of claim 1, wherein, The reduced-order observer reduces the number of state variables to be observed by directly obtaining the differential of the DC bus voltage signal as a feedback quantity.

3. The system of claim 1, wherein, The reduced-order observer adopts a reduced-order linear extended state observer, and the phase compensator is configured in the observation channel of the total disturbance to provide the required lead phase angle to compensate for the phase lag characteristic inherent in the linear extended state observer when processing time-varying disturbances.

4. The system of claim 3, wherein, The phase compensator is a first-order lead correction network.

5. The system of claim 1, wherein, The feedback controller is a linear state feedback controller based on voltage deviation and its differential signal.

6. The system of claim 1, wherein, The reinforcement learning algorithm adopted by the parameter optimization unit is the actor-critic algorithm.

7. The system of claim 6, wherein, The actor-critic algorithm adopts the deep deterministic policy gradient algorithm.

8. The system of claim 1, wherein, The state space of the parameter optimization unit includes the DC bus voltage, its differential value, and the voltage deviation and its differential value; and the action space thereof includes the time constant and scaling factor of the phase compensator, and / or the bandwidth of the reduced-order observer.

9. A method for advanced correction active disturbance rejection control of a microgrid with photovoltaic energy storage, characterized in that, The system as claimed in claims 1-8 is adopted.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program, when executed by a processor, is used to implement the active disturbance rejection control system as claimed in any one of claims 1-8.

Images