Microgrid hierarchical fractional order robust control method based on neural network order estimation

Abstract

The application provides a micro-grid hierarchical fractional order robust control method based on neural network order estimation, and belongs to the technical field of micro-grid control. The application establishes the mapping relationship between the micro-grid operation state parameters and the fractional order PID controller order parameters by constructing a neural network order estimator, adopts the fractional order PID controller which has better dynamic performance than the integer order PID, increases the adjustable integral order and differential order, and combines the self-optimization ability of the LSTM neural network, so that the fractional order PID controller can obtain more suitable parameter configuration under different working conditions. Meanwhile, the application establishes a small signal model of the micro-grid closed-loop system, constructs the stability criterion and the stability domain boundary based on the frequency domain analysis method, and checks and corrects the order parameters. The application significantly improves the adaptability of the system to complex working condition changes, improves the transient control performance of the micro-grid, reduces the overshoot and shortens the regulation time.

Description

Technical Field

[0001] This invention belongs to the field of microgrid control technology, and in particular relates to a hierarchical fractional-order robust control method for microgrids based on neural network order estimation. Background Technology

[0002] Microgrids, as an effective integration method for distributed power sources, can flexibly operate in both grid-connected and islanded modes, and are an important means to improve the absorption capacity of renewable energy and the reliability of power supply. However, distributed power sources in microgrids, containing a large number of power electronic interfaces, exhibit complex dynamic characteristics such as multi-timescale, strong nonlinearity, and uncertainty, posing severe challenges to their stable control. Especially in real-world scenarios with continuously increasing renewable energy penetration, increasingly complex load types, and frequent switching of operating modes, the dynamic behavior of microgrid systems becomes even more complex, and traditional control methods face higher requirements in terms of response speed, control accuracy, and disturbance rejection capability. Therefore, research on high-performance control methods for the complex dynamic characteristics of microgrids has become one of the important research directions in this field.

[0003] Currently, microgrids widely adopt a hierarchical control architecture. The primary control layer, typically based on droop control, is responsible for rapid response to load changes and maintaining power distribution. The secondary control layer primarily compensates for frequency and voltage deviations introduced by the primary control layer to restore the system to its rated operating state. In this hierarchical control structure, the primary control layer directly acts on the local control unit of the distributed power source, and its control performance directly affects the system's frequency, voltage stability, and power distribution accuracy. Therefore, the design level of the primary control layer controller largely determines the dynamic response quality and stable operation performance of the entire microgrid system. However, traditional integer-order PID control strategies often fail to achieve ideal control results when facing the fractional-order characteristics of microgrids (such as the memory characteristics of energy storage elements and the frequency-varying characteristics of transmission lines). When the microgrid is affected by load disturbances, parameter changes, and external interference during operation, if the controller parameters in the primary control layer are not designed properly, it may lead to problems such as large overshoot, increased oscillations, prolonged settling time, and even decreased stability margin during the system's dynamic adjustment process, thereby affecting the compensation effect of the secondary control layer and the overall operating quality of the microgrid system. Summary of the Invention

[0004] This invention aims to solve the technical problems of fixed order of fractional-order controllers in microgrids, inability to adaptively adjust online, and lack of strict stability criteria in the prior art. It provides a hierarchical fractional-order robust control method for microgrids based on neural network order estimation, which estimates the order parameters of the fractional-order PID controller in real time and establishes system stability criteria by combining frequency domain analysis, thereby improving the stability and robustness of microgrids under complex operating conditions.

[0005] This invention provides a hierarchical fractional-order robust control method for microgrids based on neural network order estimation. The microgrid adopts a hierarchical control architecture, including a primary control layer and a secondary control layer. In the primary control layer, a fractional-order PID controller is used as a voltage / current dual-loop tracking controller. The fractional-order PID controller includes proportional coefficient, integral coefficient, derivative coefficient, integral order, and derivative order. A neural network order estimator is constructed, using a Long Short-Term Memory (LSTM) network as input, with real-time state parameters representing the current operating conditions of the microgrid as input, and fractional-order and differential-order estimates of a fractional-order PID controller as output; the real-time operating status data includes at least the distributed power output, load power, bus frequency deviation, and voltage deviation. During the online operation phase, the neural network order estimator outputs the corresponding order value in real time according to the current microgrid operating status and updates the fractional-order PID controller. A small-signal model of a microgrid closed-loop system including a fractional-order PID controller is established. The system stability constraints are derived by frequency domain analysis, the stability domain boundary of the controller parameters is determined, and the stability of the output order value is verified and corrected.

[0006] Preferably, the primary control layer is based on a droop control mechanism, and the droop control equation is as follows: ; ; in, This is the frequency reference value for the droop control output. This is a reference value for the voltage amplitude of the droop control output. and These are the rated reference values ​​for frequency and voltage, respectively. and These are the active power-frequency droop coefficient and the reactive power-voltage droop coefficient, respectively. and These represent the actual active power and actual reactive power output by the inverter, respectively. and These are the active power reference value and the reactive power reference value, respectively; the fractional-order PID controller is used for tracking control in the voltage outer loop and / or current inner loop, and its transfer function is as follows: ; in, , and These are the proportional, integral, and differential coefficients, respectively. For the order of integration, It is the order of the differential, and satisfies .

[0007] Preferably, the neural network order estimator is trained using offline generated working condition-optimal order samples, and the sample construction process is as follows: Design typical microgrid operating conditions, including load step, photovoltaic fluctuation, and grid-connected / islanded operation switching; Under various typical operating conditions, one or more of the following state parameters are collected: active load fluctuation, reactive load fluctuation, photovoltaic output power change rate, current frequency deviation, frequency change rate, and common coupling point voltage amplitude deviation. The collected state parameters are normalized, and time-series input samples are constructed according to a preset time window. Using an intelligent optimization algorithm, with the goal of minimizing the preset system dynamic response performance index, the integral order of the fractional-order PID controller is optimized. and differential order Optimization is performed to obtain the optimal integral order corresponding to each working condition sample. and optimal differential order ; Using the time-series input samples as input, the corresponding optimal integration order is... and optimal differential order As output labels, a training dataset is constructed for training the LSTM neural network order estimator.

[0008] Preferably, the neural network order estimator employs a Long Short-Term Memory (LSTM) network, and its input layer feature vector includes at least one of the following: active load fluctuation, reactive load fluctuation, distributed power output power change rate, frequency deviation, frequency change rate, and voltage deviation; its output layer nodes correspond to the integration order. and differential order The LSTM neural network is based on the current input. and the hidden state at the previous moment Calculate the forget gate, input gate, output gate, and candidate memory states, and update the memory unit state and output the hidden state based on the forget gate, input gate, and output gate to obtain the order of integration for estimation. and differential order The temporal characteristics are represented.

[0009] Preferably, the specific process of establishing a small-signal model of a microgrid closed-loop system including a fractional-order PID controller and deriving the system stability constraints through frequency domain analysis includes: Near the rated operating point of the microgrid, small-signal linearization is performed on the distributed power inverter, the output-side LC filter, and the grid connection line impedance to obtain a small-signal model of the microgrid controlled object. This small-signal model is used to characterize the dynamic response relationship between the control reference input and the inverter output voltage, and its transfer function is expressed as follows: ; in, For filtering inductors, For filtering capacitors, The equivalent damping resistance or line equivalent resistance; based on the fractional-order PID controller and the small-signal model of the microgrid controlled object, an open-loop transfer function is constructed: ; in, The transfer function of a fractional-order PID controller is represented. Represents the transfer function of the controlled object. , , These are the proportional coefficient, integral coefficient, and differential coefficient, respectively. and These are the integral and derivative orders, respectively. Under unity negative feedback, a small-signal model of the microgrid closed-loop system including a fractional-order PID controller is obtained, and its closed-loop transfer function is expressed as: ; Mapping the open-loop transfer function to the frequency domain, let ,in, The imaginary unit, For angular frequency, the fractional-order operator and Mapped to respectively and The expression for the fractional-order PID controller in the frequency domain is obtained. Based on the open-loop transfer function formed by the fractional-order PID controller and the controlled object of the microgrid, Nyquist curves and / or Bode plots are plotted to analyze the amplitude-frequency and phase-frequency characteristics of the system at different frequencies, and based on the phase margin... and gain margin Construct system stability constraints, wherein the stability constraints are: in the order of neural network estimation , Within the range of variation, the following conditions must be met simultaneously: ; ; in, The minimum allowable phase margin; This represents the minimum allowable gain margin.

[0010] Preferably, the process of performing stability verification and correction on the output order value specifically involves: taking the current integral order estimate of the neural network order estimator... and differential order estimate proportional coefficient with current controller Integral coefficient and differential coefficients The system combines parameters and determines whether the corresponding parameter combinations satisfy stability constraints. If the parameter combinations do not satisfy the stability constraints, the system adjusts the current output integral order estimate according to a preset projection rule, minimum distance criterion, or boundary search method. and differential order estimate The fractional-order PID controller is mapped or corrected to a feasible region within the stability domain, and the corrected order value is used to update the controller.

[0011] Preferably, the secondary control layer acquires the microgrid bus frequency and voltage information, compares them with the corresponding rated values, and generates compensation signals through an integral controller. and The compensation signal is sent to each primary controller to correct its frequency and voltage reference values; the corrected reference values ​​are expressed as follows: ; ; in, This is the frequency reference value for the droop control output. This is a reference value for the voltage amplitude of the droop control output. This is the corrected frequency reference value. This is the corrected voltage reference value; through the compensation effect of the secondary control layer, the steady-state frequency deviation and voltage deviation caused by the primary droop control are gradually eliminated.

[0012] Preferably, the multiple distributed power sources in the primary control layer are connected through a sparse communication network, and use a multi-agent consensus algorithm to exchange output frequency, voltage and power information, and coordinate to adjust their respective output power to achieve power distribution.

[0013] Compared with the prior art, the present invention has the following innovative features and beneficial effects: Adaptive order adjustment: This invention establishes a mapping relationship between microgrid operating state parameters and fractional-order PID controller order parameters by constructing a neural network order estimator, thereby realizing integral order adjustment. and differential order The online estimation and dynamic update enable the controller to dynamically adjust the fractional order according to the real-time operating conditions of the microgrid, overcoming the shortcomings of fixed parameters in traditional fractional order control and significantly improving the system's adaptability to complex operating conditions.

[0014] High control precision: The fractional-order PID controller adopted in this invention has better dynamic performance than integer-order PID controllers, and increases the adjustable integral order. and differential order By combining the self-optimization capability of the LSTM neural network, the fractional-order PID controller can obtain more suitable parameter configurations under different operating conditions, further improving the control accuracy of microgrid frequency and voltage, improving the transient control performance of microgrid, reducing overshoot and shortening the settling time.

[0015] Stability is guaranteed: This invention establishes a small-signal model of a microgrid closed-loop system, rigorously constructs stability criteria and stability domain boundaries based on frequency domain analysis, and verifies and corrects the order parameters output by the neural network to ensure that the system maintains stable operation during the online parameter adjustment process of the neural network, thus avoiding the instability risk that may be caused by relying solely on data-driven methods.

[0016] Combining data-driven and model-based analysis: This invention combines the online estimation capabilities of neural networks with frequency domain stability constraint criteria, achieving a synergistic fusion of data-driven parameter estimation and model constraint analysis, thereby improving the reliability and engineering feasibility of control strategies.

[0017] Layered Coordination and Optimization: This invention adopts a layered control architecture. In the primary control layer, online updates and rapid responses of fractional-order PID control parameters are achieved, while in the secondary control layer, frequency and voltage deviation compensation is achieved. This ensures the speed and steady-state accuracy of the microgrid's dynamic response and meets the control requirements of the microgrid at different time scales. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the following description is only one embodiment of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a diagram illustrating the overall architecture of the hierarchical fractional-order robust control method for microgrids based on neural network order estimation as described in this invention.

[0020] Figure 2 This is a schematic diagram of the fractional-order PID controller and neural network order estimator in this invention.

[0021] Figure 3 This is a flowchart of the frequency domain stability analysis in this invention.

[0022] Figure 4 The figure shows the simulation results in the example. Detailed Implementation

[0023] This invention proposes a hierarchical fractional-order robust control method for microgrids based on neural network order estimation, which is executed within a hierarchical control framework for microgrids and includes the following processes: A hierarchical control architecture is constructed: the hierarchical control architecture includes at least a primary control layer and a secondary control layer; wherein, the primary control layer is based on a droop control mechanism, and a local controller is set for each distributed power unit to realize rapid local adjustment and power distribution of the distributed power; the secondary control layer is used to compensate for the frequency deviation and voltage deviation introduced by the primary control layer.

[0024] Design of a fractional-order PID primary controller: In the primary control layer, a fractional-order PID controller is used as a voltage / current dual-loop tracking controller, forming the control core of the distributed power supply local control unit. The transfer function form of the fractional-order PID controller is: ; in, , , These are the proportional, integral, and differential coefficients, respectively. For the order of integration, It is the order of the differential, and By introducing two adjustable parameters, the integral order and the derivative order, the controller's ability to describe the complex dynamic characteristics of the microgrid and its control flexibility are improved.

[0025] Constructing a neural network order estimator: Collect real-time operating status data of the microgrid, including but not limited to: distributed generation output power, load power, bus frequency deviation, and voltage deviation; construct a multi-layer feedforward neural network, using the real-time operating status data as input, and using the optimal integral order of a fractional-order PID controller. and optimal differential order For output, the neural network is trained offline; in actual operation, the trained neural network is used to output the optimal order estimate of the fractional-order PID controller online according to the current operating conditions, and the primary controller parameters are updated in real time accordingly to improve the controller's adaptability to changes in operating conditions.

[0026] Frequency Domain Stability Analysis and Criterion Construction: A small-signal model of the closed-loop system, including the fractional-order PID controller and the microgrid controlled object, is established. The stability of the closed-loop system is analyzed in the frequency domain, and the controller parameters are derived based on the gain margin and phase margin requirements. Stability constraints must be met. These stability constraints constrain the order parameters of the neural network output to ensure that the controller parameter adjustment process meets the requirements for stable operation of the closed-loop system.

[0027] Secondary control layer compensation: The secondary control layer detects the steady-state deviations in frequency and voltage caused by the primary control layer, generates a compensation signal through a slow integral circuit, and superimposes the compensation signal onto the reference value of the primary control layer to correct the reference value of the primary control layer, thereby achieving error-free regulation of the microgrid frequency and voltage.

[0028] The implementation process of the present invention will be further described below with reference to specific embodiments.

[0029] I. System Architecture Construction like Figure 1 As shown, the microgrid system in this embodiment includes several distributed power sources (such as photovoltaics, energy storage, and micro gas turbines). Each distributed power source is connected in parallel to a common AC bus via an inverter, collectively forming the microgrid power supply system. To achieve effective control of the microgrid's frequency, voltage, and power distribution, this embodiment adopts a hierarchical control architecture, which is divided into two layers: 1. Primary Control Layer (Local Control Layer): Each distributed power source is equipped with a local controller, which is integrated into the inverter controller and includes a power calculation module, a droop control module, and a voltage and current dual closed-loop module. The power calculation module calculates the actual active and reactive power of the inverter based on the sampled output voltage and current. The droop control module generates frequency and voltage reference values ​​based on the power deviation. The voltage and current dual closed-loop control module drives the inverter to track the reference values ​​and outputs corresponding control signals.

[0030] The improvement of this invention lies in the fact that both the voltage outer loop and the current inner loop adopt fractional-order PID controllers to replace the traditional integer-order PID controllers, thereby enhancing the controller's adaptability to the complex dynamic characteristics of the microgrid and improving the dynamic response performance and robustness of the system.

[0031] In this embodiment, the droop control equation is as follows: ; ; in, This is the frequency reference value for the droop control output. This is a reference value for the voltage amplitude of the droop control output. and These are the rated reference values ​​for frequency and voltage, respectively. and These are the active power-frequency droop coefficient and the reactive power-voltage droop coefficient, respectively. and These represent the actual active power and actual reactive power output by the inverter, respectively. and These are the active power reference value and the reactive power reference value, respectively. Through the aforementioned droop control relationship, the frequency and voltage reference values ​​can be automatically adjusted according to changes in the output power of the distributed power source, thereby achieving initial power allocation among multiple distributed power sources. The voltage amplitude and frequency reference values ​​output by the droop control are further fed into a subsequent fractional-order voltage / current controller for tracking control.

[0032] 2. Secondary control layer (centralized / distributed control layer): The secondary control layer is used to further compensate for frequency and voltage deviations introduced by the primary control layer, in order to restore the microgrid system to its rated operating state. The secondary controller can adopt a centralized or distributed structure, collecting global or local voltage and frequency information through a low-speed communication network, calculating the deviation from the corresponding rated values, and generating compensation amounts via a PI controller. and .

[0033] The compensation amount and This information is sent to each primary controller to correct its frequency and voltage reference values. The corrected reference values ​​can be expressed as: ; ; in, This is the corrected frequency reference value. This is the corrected voltage reference value. Through the compensation effect of the secondary control layer, the steady-state frequency deviation and voltage deviation caused by the primary droop control can be gradually eliminated, achieving zero steady-state error regulation of the microgrid frequency and voltage.

[0034] II. Fractional-order PID controller design like Figure 2 As shown, the fractional-order PID controller is specifically implemented using the Oustaloup filter approximation method. The fractional-order operator... In the selected frequency band The internal approximation is a high-order integer transfer function to facilitate the implementation of fractional integral and fractional differential operations in existing digital controllers or simulation platforms. In this embodiment, the selected frequency band is... The approximation order is 5 rad / s. Through the above approximation process, the fractional-order controller can be converted into an engineering-implementable rational transfer function form while ensuring approximation accuracy within the target frequency band, thus balancing control performance and implementation complexity.

[0035] In this embodiment, the transfer function of the fractional-order PID controller is expressed as: ; in, , and These are the proportional, integral, and differential coefficients, respectively. For the order of integration, This is the derivative order. Compared with traditional integer-order PID controllers, this controller, by introducing non-integer-order integral and derivative terms, can more flexibly adjust the system's frequency domain characteristics and time domain dynamic response, thereby improving the speed, stability, and robustness of the microgrid inverter control process.

[0036] Controller initial parameters , , Preliminary values ​​are obtained through engineering tuning or offline particle swarm optimization. The key is the integration order. and differential order The values ​​are not fixed offline, but determined online in real time by a neural network estimator. Specifically, during the operation of the microgrid, based on the real-time collected system status information, the neural network estimator outputs the estimated integral and derivative orders corresponding to the current operating conditions, and updates these estimated values ​​to the fractional-order PID controller in real time, so that the controller parameters can be dynamically adjusted according to changes in operating conditions.

[0037] III. Construction and Training of Neural Network Order Estimators 1. Input Feature Selection: Feature quantities that characterize the operating conditions of the microgrid are selected as the input to the neural network. : Active load fluctuation ; reactive load fluctuation ; Photovoltaic output power change rate ; Current frequency deviation ; Rate of change of frequency ; Common Coupling Point Voltage Amplitude Deviation ; The aforementioned input features can reflect the current operating status of the microgrid from different perspectives, including load disturbances, distributed power generation output changes, and dynamic responses of system voltage and frequency. Specifically, active and reactive load fluctuations characterize load-side disturbances, the photovoltaic output power change rate reflects the fluctuation characteristics of renewable energy output, the current frequency deviation and frequency change rate describe the dynamic behavior of system frequency, and the point of common coupling voltage amplitude deviation reflects the system voltage regulation status. By selecting these features as neural network inputs, the dynamic characteristics of the microgrid under different operating conditions can be comprehensively depicted, providing a data foundation for online estimation of the order parameters of the fractional-order PID controller.

[0038] 2. Network Structure and Output: Construct a neural network model based on Long Short-Term Memory (LSTM): The input feature dimension is 6, corresponding to the 6 operating condition features mentioned above, and the input is fed into the LSTM network in time series form, where time... The input feature vector is defined as: ; superscript Represents the transpose of a vector; Input feature vectors at consecutive time points The time-series input samples constitute the LSTM network to characterize the changes in the microgrid's operating conditions over time. Let the input time window length be... Then at time The corresponding input feature sequence can be represented as: ; The number of hidden layer units in the LSTM was experimentally determined to be 15 to balance temporal feature extraction capability and online computational complexity. The LSTM network extracts temporally relevant features from the operating condition sequence through a gating mechanism; its internal state update process can be represented as follows: ; ; ; ; ; ; in, , and These are the outputs of the forget gate, input gate, and output gate, respectively. For the unit state, It is in a hidden state. The candidate cell state is... , , , For the corresponding weight matrix, , , , For the corresponding bias term, This represents the Sigmoid activation function. This represents the Hadamard product. The output layer has 2 nodes, corresponding to the optimal integration order. and optimal differential order The output layer uses the linear activation function purelin, and its output can be expressed as: ; in, This is the output layer weight matrix. This serves as the output layer bias vector; and the network output is subjected to amplitude limiting or mapping constraints to ensure that the output range is within... Within the range.

[0039] In this embodiment, the neural network is used to establish a nonlinear mapping relationship between the operating conditions of the microgrid and the order parameters of the fractional-order PID controller, so that the integral and derivative orders of the controller can be dynamically adjusted as the system operating state changes.

[0040] 3. Training data generation: During the offline training phase, a microgrid simulation model is built. For different operating conditions (such as load step changes, photovoltaic fluctuations, and islanding switching), a particle swarm optimization algorithm is used to traverse and search for the optimal order that minimizes the comprehensive performance index (such as the ITAE index) under each operating condition. and The neural network is trained under supervised conditions by using the operating condition feature data as input samples and the corresponding optimal order as the label.

[0041] Specifically, under different typical operating conditions, the input feature quantities at corresponding times are first collected through simulation to form neural network input samples; then, within a given parameter search range, the integral order of the fractional-order PID controller is optimized using the particle swarm optimization algorithm. and differential order The optimization process aims to achieve the optimal preset comprehensive performance index. Finally, the optimal order parameter obtained through optimization is used as the output label, which, together with the corresponding operating condition characteristics, constitutes the training sample set. Through offline training with a large number of typical operating condition samples, the neural network learns to establish the mapping relationship between the microgrid operating state characteristics and the optimal order parameter of the fractional-order PID controller.

[0042] 4. Online applications: The trained and converged neural network is deployed in the microgrid monitoring system. In each control cycle (e.g., 0.1s), current operating condition characteristics are collected. Forward calculation yields and The order parameters of the fractional-order PID controller are updated in real time. Specifically, during the online operation of the microgrid, the monitoring system periodically collects the current operating condition characteristics and inputs them into the trained neural network model. Through forward propagation, the estimated integral order value under the current operating condition is obtained. and differential order estimate Subsequently, the estimated order value is sent to the fractional-order PID controller in the primary control layer, and the fractional-order integral and derivative terms in the controller are updated in real time. Through this online application method, the order parameters of the fractional-order PID controller can be dynamically adjusted according to changes in the microgrid's operating conditions, thereby improving the system's adaptability and control performance to complex operating conditions such as load disturbances, photovoltaic fluctuations, and operating mode switching.

[0043] IV. Frequency Domain Stability Analysis and Criterion Generation To ensure online adjustment of neural networks and The system remains stable. This invention first establishes a small-signal model of a microgrid closed-loop control system containing a fractional-order PID controller near the rated operating point. Based on this, an open-loop transfer function for frequency domain analysis is derived. Furthermore, stability constraints are generated based on phase margin and gain margin, and these constraints are used as the basis for verifying and correcting the order parameters of the neural network's online output. This avoids the risk of closed-loop instability that may arise from updating parameters solely based on data-driven results. The overall process is as follows: Figure 3 As shown.

[0044] 1. Establish a small-signal model of a microgrid closed-loop control system including a fractional-order PID controller: Taking a single distributed power source as an example, a closed-loop transfer function model from the inverter output voltage to the reference voltage is established, including an LC filter and line impedance characteristics. In this embodiment, the controlled object model can be composed of the inverter, a small-signal linearized filter, and an equivalent model of the grid-connected line, used to reflect the dynamic response relationship of the inverter output voltage to the control reference input. To facilitate frequency domain stability analysis, small-signal modeling of the microgrid system can be performed near the rated operating point to obtain a linear approximation model suitable for controller parameter design and stability determination.

[0045] The controlled object model can be represented in general rational fraction form: ; in and These are coefficients determined by the inverter parameters, LC filter parameters, and line impedance parameters. If the influence of the LC filter on the inverter output side is mainly considered, the controlled object model can be approximated as: ; in, For filtering inductors, For filtering capacitors, It is the equivalent damping resistance or the equivalent resistance of the line.

[0046] Based on the fractional-order PID controller and the small-signal model of the microgrid controlled object, the open-loop transfer function is constructed as follows: ; in, The transfer function of a fractional-order PID controller is represented. Represents the transfer function of the controlled object. , , These are the proportional coefficient, integral coefficient, and differential coefficient, respectively. and These represent the integral and derivative orders, respectively. Under unity negative feedback, a small-signal model of a microgrid closed-loop system including a fractional-order PID controller is obtained, and its closed-loop transfer function is expressed as follows: ; because and Since the parameters are changed online, the open-loop frequency response of the system will change in real time with the output of the neural network. Therefore, it is necessary to dynamically analyze and constrain its stability margin.

[0047] 2. Frequency domain stability margin analysis: To determine the integral order obtained by the online estimation of the neural network and differential order To determine whether the closed-loop system stability requirements are still met after combining the current controller parameters, the microgrid closed-loop system containing the fractional-order PID controller is mapped to the frequency domain. Stability criteria are constructed based on phase margin and gain margin, providing a basis for online verification and correction of subsequent order parameters.

[0048] make ,in, The imaginary unit, If ω is the angular frequency, then ; Similarly, ; Therefore, the expression for the fractional-order PID controller in the frequency domain can be written as: ; Further expansion yields: ; Combined with the frequency domain transfer function of the controlled object The open-loop transfer function of the system in the frequency domain is expressed as: ; Based on this, Nyquist plots or Bode plots of the open-loop transfer function are drawn to analyze the amplitude-frequency and phase-frequency characteristics of the system at different frequencies. The phase margin of the system is defined. and gain margin This is used as an important indicator for evaluating the stability and robustness of a closed-loop system.

[0049] Let the gain crossover frequency be set. satisfy: ; The phase margin of the system is defined as: ; Let the phase crossover frequency be... satisfy: ; The gain margin of the system is then defined as: ; If expressed in decibels, it can be represented as: ; Based on the above definition, the stability constraints for the controller parameters are constructed as follows: In the order of the neural network estimation... , Within the range of variation, the following conditions must be met simultaneously: ; ; in, The minimum allowable phase margin is typically taken as 30° to 60°; The minimum allowable gain margin is typically set to no less than 6 dB. This constraint ensures that the system maintains sufficient stability margin and disturbance rejection capability even as the controller order parameters change. For different types of microgrid systems, the values ​​of the minimum phase angle margin and minimum gain margin can be adjusted appropriately according to specific dynamic performance requirements.

[0050] 3. Determination of the stability region and its constraint boundaries: Based on the above stability constraints, the controller parameter combination can be determined. Stability region The stability region is defined as follows: ; The stability region represents the set of parameters that satisfy the preset minimum phase margin and minimum gain margin requirements. The critical parameter set of the stability region constitutes the stability region boundary, which characterizes the critical conditions under which the controller parameter combination satisfies the closed-loop stability requirements. The stability region boundary can be further represented by a stability region boundary function as follows: ; The output of the neural network Should be consistent with current controller parameters All orders satisfy the stability constraints and reside within the stability region. If the order combination output by the neural network exceeds the stability boundary, a parameter clamping mechanism is activated to correct the order to the nearest point on the stability boundary.

[0051] 4. Online verification and correction of order parameters After each time the neural network outputs a new estimate of the integral and derivative orders, it first checks whether the corresponding parameter combination satisfies the stability constraints by combining the current proportional, integral, and derivative coefficients. If the result is that the constraints are not met, the currently output order parameters are mapped or corrected to a feasible region within the stability domain according to a preset projection rule, minimum distance criterion, or boundary search method. Then, the corrected order parameters are used to update the fractional-order PID controller to ensure that the updated fractional-order PID controller still meets the requirements for stable operation of the closed-loop system. By setting the above stability verification and correction mechanism, the online estimation capability of the neural network and the frequency domain robust stability constraints can be organically combined, thereby improving the safety and engineering feasibility of the control strategy.

[0052] V. Simulation Verification A microgrid model was built in Matlab / Simulink for verification. The microgrid model includes distributed power generation units, inverters, LC filters, a common AC bus, load modules, and a hierarchical controller module. The primary control layer employs a fractional-order PID control strategy, while the secondary control layer is used for frequency and voltage deviation compensation. To verify the dynamic control performance and stability of the method under disturbance conditions, time-domain simulation analysis was performed on the built model.

[0053] Setting operating conditions: In At a certain time, the system load suddenly increases by 20% to simulate a typical load disturbance scenario during microgrid operation, and to examine the impact of different control strategies on the system frequency, voltage, and dynamic response performance under this disturbance. Three control methods are compared: Traditional integer-order PI control; Fractional order with fixed parameters control; The present invention relates to an adaptive fractional-order control based on neural network order estimation.

[0054] Among them, traditional integer-order PI control is used as a benchmark to reflect the dynamic adjustment capability of conventional control strategies under disturbance conditions; fractional-order PI control with fixed parameters... The control is used to demonstrate the improvement in dynamic performance of the fractional-order controller compared to the integer-order controller; the adaptive fractional-order control based on neural network order estimation described in this invention is used to verify the further improvement in system control performance after introducing online order estimation and stability constraint mechanisms on the basis of fractional-order control.

[0055] Simulation results (such as) Figure 4 As shown in the figure, the method of the present invention reduces the frequency drop depth by about 30% and the adjustment time by 40% compared with the traditional method. It can also maintain good dynamic response under different working conditions, which verifies the effectiveness of the neural network adaptive parameter tuning and the correctness of the stability criterion.

[0056] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0057] While the above description illustrates specific embodiments of the present invention, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A hierarchical fractional-order robust control method for microgrids based on neural network order estimation, wherein the microgrid adopts a hierarchical control architecture, including a primary control layer and a secondary control layer, characterized in that: In the primary control layer, a fractional-order PID controller is used as a voltage / current dual-loop tracking controller. The fractional-order PID controller includes proportional coefficient, integral coefficient, derivative coefficient, integral order, and derivative order. A neural network order estimator is constructed, using a Long Short-Term Memory (LSTM) network as input, with real-time state parameters representing the current operating conditions of the microgrid as input, and fractional-order and differential-order estimates of a fractional-order PID controller as output; the real-time operating status data includes at least the distributed power output, load power, bus frequency deviation, and voltage deviation. During the online operation phase, the neural network order estimator outputs the corresponding order value in real time according to the current microgrid operating status and updates the fractional-order PID controller. A small-signal model of a microgrid closed-loop system including a fractional-order PID controller is established. The system stability constraints are derived by frequency domain analysis, the stability domain boundary of the controller parameters is determined, and the stability of the output order value is verified and corrected.

2. The hierarchical fractional-order robust control method for microgrids based on neural network order estimation as described in claim 1, characterized in that: The primary control layer is based on a droop control mechanism, and the droop control equation is as follows: ； ； in, This is the frequency reference value for the droop control output. This is a reference value for the voltage amplitude of the droop control output. and These are the rated reference values ​​for frequency and voltage, respectively. and These are the active power-frequency droop coefficient and the reactive power-voltage droop coefficient, respectively. and These represent the actual active power and actual reactive power output by the inverter, respectively. and These are the active power reference value and the reactive power reference value, respectively; the fractional-order PID controller is used for tracking control in the voltage outer loop and / or current inner loop, and its transfer function is as follows: ； in, , and These are the proportional, integral, and differential coefficients, respectively. For the order of integration, It is the order of the differential and satisfies .

3. The hierarchical fractional-order robust control method for microgrids based on neural network order estimation as described in claim 1, characterized in that: The neural network order estimator is trained using offline generated working condition-optimal order samples. The sample construction process is as follows: Design typical microgrid operating conditions, including load step, photovoltaic fluctuation, and grid-connected / islanded operation switching; Under various typical operating conditions, one or more of the following state parameters are collected: active load fluctuation, reactive load fluctuation, photovoltaic output power change rate, current frequency deviation, frequency change rate, and common coupling point voltage amplitude deviation. The collected state parameters are normalized, and time-series input samples are constructed according to a preset time window. An intelligent optimization algorithm is employed to minimize the preset dynamic response performance index of the system, thereby optimizing the integral order of the fractional-order PID controller. and differential order Optimization is performed to obtain the optimal integral order corresponding to each working condition sample. and optimal differential order ; Using the time-series input samples as input, the corresponding optimal integration order is... and optimal differential order As output labels, a training dataset is constructed for training the LSTM neural network order estimator.

4. The hierarchical fractional-order robust control method for microgrids based on neural network order estimation as described in claim 1, characterized in that: The neural network order estimator employs a Long Short-Term Memory (LSTM) network, and its input layer feature vector includes at least one of the following: active load fluctuation, reactive load fluctuation, distributed power output power change rate, frequency deviation, frequency change rate, and voltage deviation; its output layer nodes correspond to the integration order. and differential order The LSTM neural network is based on the current input. and the hidden state at the previous moment Calculate the forget gate, input gate, output gate, and candidate memory states, and update the memory unit state and output the hidden state based on the forget gate, input gate, and output gate to obtain the order of integration for estimation. and differential order The temporal characteristics are represented.

5. The hierarchical fractional-order robust control method for microgrids based on neural network order estimation as described in claim 1, characterized in that: The specific process of establishing a small-signal model of a microgrid closed-loop system including a fractional-order PID controller and deriving the system stability constraints through frequency domain analysis includes: Near the rated operating point of the microgrid, small-signal linearization is performed on the distributed power inverter, the output-side LC filter, and the grid connection line impedance to obtain a small-signal model of the microgrid controlled object. This small-signal model is used to characterize the dynamic response relationship between the control reference input and the inverter output voltage, and its transfer function is expressed as follows: ； in, For filtering inductors, For filtering capacitors, The equivalent damping resistance or line equivalent resistance; based on the fractional-order PID controller and the small-signal model of the microgrid controlled object, an open-loop transfer function is constructed: ； in, The transfer function of a fractional-order PID controller is represented. Represents the transfer function of the controlled object. , , These are the proportional coefficient, integral coefficient, and differential coefficient, respectively. and These are the integral and derivative orders, respectively. Under unity negative feedback, a small-signal model of the microgrid closed-loop system including a fractional-order PID controller is obtained, and its closed-loop transfer function is expressed as: ； Mapping the open-loop transfer function to the frequency domain, let ,in, The imaginary unit, For angular frequency, the fractional-order operator and Mapped to respectively and The expression for the fractional-order PID controller in the frequency domain is obtained. Based on the open-loop transfer function formed by the fractional-order PID controller and the controlled object of the microgrid, Nyquist curves and / or Bode plots are plotted to analyze the amplitude-frequency and phase-frequency characteristics of the system at different frequencies, and based on the phase margin... and gain margin Construct system stability constraints, wherein the stability constraints are: in the order of neural network estimation , Within the range of variation, the following conditions must be met simultaneously: ； ； in, The minimum allowable phase margin; This represents the minimum allowable gain margin.

6. The hierarchical fractional-order robust control method for microgrids based on neural network order estimation as described in claim 5, characterized in that: The process of performing stability verification and correction on the output order value is as follows: The current integral order estimate of the neural network order estimator is then... and differential order estimate proportional coefficient with current controller Integral coefficient and differential coefficients Combine parameters and determine whether the corresponding parameter combinations satisfy the stability constraints. When the parameter combination does not satisfy the stability constraint, the current output integral order estimate is adjusted according to the preset projection rule, minimum distance criterion, or boundary search method. and differential order estimate The fractional-order PID controller is mapped or corrected to a feasible region within the stability domain, and the corrected order value is used to update the controller.

7. The hierarchical fractional-order robust control method for microgrids based on neural network order estimation as described in claim 1, characterized in that: The secondary control layer acquires the microgrid bus frequency and voltage information, compares them with the corresponding rated values, and generates compensation signals through the integral controller. and The compensation signal is sent to each primary controller to correct its frequency and voltage reference values; the corrected reference values ​​are expressed as follows: ； ； in, This is the frequency reference value for the droop control output. This is a reference value for the voltage amplitude of the droop control output. This is the corrected frequency reference value. This is the corrected voltage reference value; through the compensation effect of the secondary control layer, the steady-state frequency deviation and voltage deviation caused by the primary droop control are gradually eliminated.

8. The hierarchical fractional-order robust control method for microgrids based on neural network order estimation as described in claim 1, characterized in that: The multiple distributed power sources in the primary control layer are connected through a sparse communication network. They exchange output frequency, voltage, and power information using a multi-agent consensus algorithm and coordinate to adjust their respective output power to achieve power distribution.

Images