Multi-agent based virtual power plant hierarchical voltage coordination control system and method

Abstract

The application discloses a multi-agent-based virtual power plant hierarchical voltage coordination control system and method, relates to the technical field of coordination control, and comprises the following steps: acquiring original operation data of distributed energy units in a virtual power plant, and constructing a dynamic mapping model representing the relationship between node voltage and control input; decomposing the dynamic mapping model into a fast dynamic submodel and a slow dynamic submodel; generating a fast-loop control law based on the fast dynamic submodel, and performing residual error calculation and online correction on the slow dynamic submodel according to the fast-loop control law; establishing a global optimization objective function according to the corrected slow dynamic submodel, and solving node reference values through a distributed algorithm; and issuing the node reference values to local inverters, and generating control input signals in combination with the fast-loop control law; and the application solves the problems of voltage oscillation, tracking error and slow convergence caused by fast-slow control conflicts in the virtual power plant through fast-slow dynamic hierarchical collaborative control.

Description

Technical Field

[0001] This invention relates to the field of coordinated control technology, and more specifically, to a hierarchical voltage coordinated control system and method for a virtual power plant based on multiple agents. Background Technology

[0002] Virtual power plants (VPPs) integrate different types of distributed energy resources (solar power, energy storage, electric vehicle charging stations, flexible loads, etc.) to achieve unified scheduling and voltage control. The control objectives are: to maintain the voltage of each node within acceptable ranges; to ensure consistent voltage variation trends among multiple resources (avoiding some nodes having excessively high voltages while others have excessively low voltages); and to guarantee stable operation after renewable energy is integrated.

[0003] However, due to resource diversity, communication delays, and inaccurate modeling, common control methods exhibit shortcomings at the control level. For example, the simultaneous action of fast control from the local inverter and slow commands from the virtual power plant coordinator on a time scale can cause conflicts, resulting in oscillations (periodic voltage deviations), tracking errors, or slow convergence. To address these issues, this invention proposes a solution. Summary of the Invention

[0004] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a hierarchical voltage coordination control system and method for virtual power plants based on multi-agent systems. By using fast and slow dynamic hierarchical collaborative control, the system solves the problems of voltage oscillation, tracking error, and slow convergence caused by fast and slow control conflicts in virtual power plants, thereby achieving rapid voltage response of each node in the virtual power plant, global coordination and stability, and efficient integration of multiple types of distributed energy sources.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] Firstly, this application provides a hierarchical voltage coordination control method for virtual power plants based on multi-agent systems. This method includes: acquiring raw operating data of distributed energy units in a virtual power plant and constructing a dynamic mapping model characterizing the relationship between node voltage and control input; decomposing the dynamic mapping model into a fast dynamic sub-model and a slow dynamic sub-model; generating a fast-loop control law based on the fast-loop sub-model, and performing residual calculation and online correction on the slow-dynamic sub-model according to the fast-loop control law; establishing a global optimization objective function based on the corrected slow-dynamic sub-model, and solving for the reference values ​​of each node using a distributed algorithm; and distributing the node reference values ​​to the local inverter and generating control input signals in conjunction with the fast-loop control law.

[0007] In one embodiment, a dynamic mapping model characterizing the relationship between node voltage and control input is constructed, specifically by: preprocessing the raw operating data and performing time series analysis on the preprocessed data to distinguish system dynamic variables with different time scales and extracting key feature vectors; determining the model structure based on the key feature vectors, wherein the model structure includes an autoregressive part related to historical output and an exogenous input part related to control input; constructing a regression model by combining the node voltage output time series with the input signal according to the determined model structure; solving the model structure parameters of the regression model using a parameter identification algorithm on the preprocessed data; and generating a dynamic mapping model between the node voltage and control input based on the solved parameters.

[0008] In one embodiment, the dynamic mapping model is decomposed into a fast dynamic sub-model and a slow dynamic sub-model. Specifically, the singular perturbation small parameters are adaptively adjusted based on the Lyapunov adaptive law; the mapping model is scaled according to the adjusted singular perturbation small parameters to obtain a fast dynamic subsystem containing fast dynamic variables and a slow dynamic subsystem containing slow dynamic variables; the fast dynamic subsystem is subjected to variable freezing to obtain a fast dynamic sub-model that evolves with a fast time scale; and the slow dynamic subsystem is subjected to asymptotic expansion to obtain a slow dynamic sub-model that evolves with a slow time scale.

[0009] In one embodiment, the singular perturbation small parameter is adaptively adjusted based on the Lyapunov adaptive law. Specifically, the system control error is obtained and a Lyapunov function is constructed; the derivative of the Lyapunov function is calculated, and the derivative result is used as a stability constraint; the adaptive law is set according to the stability constraint, and the value of the singular perturbation small parameter is updated in real time according to the adaptive law for dynamic application.

[0010] In one embodiment, a fast-loop control law is generated based on the fast dynamic sub-model, specifically as follows:

[0011] Based on the fast dynamic sub-model, the fast dynamic variables and discretized scheduling variables of the fast dynamic subsystem are obtained. Based on the fast dynamic variables and scheduling variables, the fast dynamic subsystem is locally linearized to establish an LPV state-space model. For each LPV state-space model, the closed-loop pole positions are set according to the fast dynamic response performance. The corresponding state feedback gain matrix is ​​solved using the pole placement method combined with the closed-loop pole positions, and each gain matrix is ​​integrated to form a gain library. The real-time control gain matrix is ​​obtained by selecting from the gain library or by multi-point interpolation based on the scheduling variables, and the fast-loop control law is calculated in conjunction with the fast dynamic variables.

[0012] In one embodiment, residual calculation and online correction are performed on the slow-dynamic sub-model based on the fast-loop control law. Specifically, the slow-dynamic prediction residual signal is obtained based on the difference between the actual node voltage under the action of the fast-loop control law and the predicted output value of the slow-dynamic sub-model. The prediction residual signal is initially screened through wavelet transformation to obtain a preliminary residual signal. The preliminary residual signal is input into a low-pass filter for processing to obtain a low-pass residual signal. The slow-dynamic sub-model is corrected based on the low-pass residual signal, and the residual is added as a feedback correction term to the state equation of the slow-dynamic sub-model to obtain the corrected slow-dynamic sub-model.

[0013] In one embodiment, the predicted residual signal is initially screened using wavelet transform to obtain a preliminary residual signal. Specifically, the sampling frequency and typical fast dynamic frequency are obtained, and the number of wavelet decomposition levels is calculated. Based on the number of wavelet decomposition levels, the predicted residual signal is decomposed into multi-scale wavelet coefficients to obtain wavelet coefficients for each level. The wavelet coefficients include high-frequency coefficients and low-frequency coefficients. According to the characteristics of the fast and slow system of the virtual power plant, the high-frequency coefficients are regarded as fast dynamic disturbances, and the low-frequency coefficients are regarded as slow dynamic components. The high-frequency coefficients are removed, and the low-frequency coefficients are retained as the preliminary residual signal.

[0014] In one embodiment, a global optimization objective function is established based on the modified slow-dynamic sub-model, and the reference values ​​of each node are solved using a distributed algorithm. Specifically, the global optimization objective function is constructed based on the state vector output by the modified slow-dynamic sub-model and the voltage measurement values, and constraints are set. The global optimization objective function includes node voltage deviation, power loss, and economic indicators, and the constraints include node voltage constraints, energy storage SOC constraints, and inverter power upper and lower limits. The global optimization problem is decomposed into local optimization sub-problems for each node, resulting in the local objective function and neighbor node constraints for each node. A distributed optimization algorithm is used to iteratively optimize each node based on the local objective function and neighbor node constraints to obtain the local reference value for the current iteration. Each node sends its local reference value to its neighboring nodes through a communication network. The local constraints are updated based on the local reference value, and local optimization is performed again to obtain a new round of reference values ​​until a preset global convergence condition is met. The final node reference values ​​for each node are then output, including node voltage reference values, current reference values, or power reference values.

[0015] In one embodiment, the node reference value is sent to the local inverter and combined with the fast-loop control law to generate a control input signal. Specifically, the node reference value is smoothed to generate a smoothed reference value; the smoothed reference value is sent to the local inverter controller through the control communication interface; the local inverter performs data verification after receiving the smoothed reference value; the control input signal is calculated based on the fast-loop control law and the verified smoothed reference value; after receiving the control input signal, the local inverter adjusts the output power, current, or voltage.

[0016] Secondly, this application provides a hierarchical voltage coordination control system and method for virtual power plants based on multi-agent systems, the system comprising:

[0017] The model building module is used to acquire the raw operating data of distributed energy units in the virtual power plant and build a dynamic mapping model that represents the relationship between node voltage and control input.

[0018] The model decomposition module is used to decompose the dynamic mapping model into fast dynamic sub-models and slow dynamic sub-models;

[0019] The fast-loop control law generation and model correction module is used to generate fast-loop control laws based on fast-dynamic sub-models, and to perform residual calculation and online correction on slow-dynamic sub-models based on fast-loop control laws.

[0020] The slow-dynamic optimization module is used to establish a global optimization objective function based on the corrected slow-dynamic sub-model and solve for the reference values ​​of each node through a distributed algorithm.

[0021] The control execution module is used to send node reference values ​​to the local inverter and generate control input signals in conjunction with the fast-loop control law.

[0022] As can be seen from the above technical solutions, the embodiments of this application have the following advantages:

[0023] 1. By using a mapping model based on system identification, fast and slow dynamic decomposition of singular perturbation theory, and LPV fast loop control law, the system achieves fast response of virtual power plant node voltage in the millisecond to second range and slow dynamic optimization coordination in the minute to hour range. This not only takes into account the transient and steady-state characteristics of voltage and avoids oscillations and tracking errors caused by fast and slow control conflicts, but also improves the global regulation accuracy and real-time control capability of various types of distributed energy sources while maintaining system stability.

[0024] 2. By combining fast-loop control law with residual correction of slow-dynamic sub-model, the virtual power plant node voltage can be rapidly responded to in milliseconds to seconds. At the same time, wavelet decomposition and low-pass filtering are used to extract slow-dynamic error signals for accurate correction. Then, a distributed optimization algorithm is used to solve for global node reference values ​​and smoothly distribute them to the local inverter. This not only improves the model prediction accuracy and system stability, but also takes into account the minimization of power loss and economic optimization, achieving efficient coordination and global optimization control of fast and slow dynamics. Attached Figure Description

[0025] Figure 1 A schematic diagram of the hierarchical voltage coordination control method for a virtual power plant based on multiple agents provided in this application embodiment.

[0026] Figure 2 A schematic diagram of the hierarchical voltage coordination control system for a virtual power plant based on multiple agents provided in this application embodiment.

[0027] Figure 3 The scatter plot showing the scale separation of the fast and slow dynamic subsystems provided in the embodiments of this application. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0029] Reference Figure 1 As shown in the diagram, the hierarchical voltage coordination control method for virtual power plants based on multi-agent systems provided by this invention includes the following steps:

[0030] S1 acquires the raw operating data of various distributed energy units in the virtual power plant, and constructs a dynamic mapping model representing the relationship between node voltage and control input based on the parameter identification algorithm.

[0031] In this embodiment, the various distributed energy units in the virtual power plant include photovoltaics, energy storage, electric vehicle charging piles, and flexible loads. The original operating data includes control inputs (such as inverter reference power and current commands) and node voltage outputs.

[0032] Specifically, a dynamic mapping model representing the relationship between node voltage and control input is constructed based on a parameter identification algorithm, as follows:

[0033] The raw operating data is cleaned to obtain preprocessed data. The cleaning includes noise reduction, outlier removal, missing data imputation and normalization.

[0034] Time series analysis is performed on the preprocessed data to distinguish between fast-dynamic and slow-dynamic variables, and key feature vectors are extracted. These key feature vectors include signal trends, periodicity, fluctuation amplitude, and lag characteristics.

[0035] Based on key feature vectors, an ARX (Autoregressive with Exogenous Input) model structure is selected, wherein the ARX model structure includes ARX model parameters consisting of autoregressive coefficients related to historical outputs and exogenous input coefficients related to control inputs.

[0036] Based on the ARX model structure, the node voltage output time series and input signal are used to construct a regression equation, and the ARX model parameters are calculated by the parameter identification algorithm based on the preprocessed data and the ARX model structure.

[0037] The specific calculation formula for the regression equation is as follows:

[0038]

[0039] In the formula, Let be the observed value of the node voltage at time t. Node voltage first Historical values ​​at each sampling time, These are the autoregressive coefficients for the historical voltage values ​​of the corresponding nodes. , ..., To control the current and historical values ​​of the input signal after a lag of d, where d is the input delay. For exogenous input coefficients, This is the error term.

[0040] A dynamic mapping model of node voltage and control input is generated based on the ARX model parameters.

[0041] Fast dynamic variables refer to signals that change significantly within a short timescale (usually milliseconds to seconds), and their rate of change is much faster than the overall system regulation or slow loop regulation capability, such as inverter current and voltage transients; slow dynamic variables refer to signals that change significantly only within a longer timescale (usually minutes to hours), and their rate of change is slower than fast dynamic responses, such as energy storage SOC and photovoltaic power prediction.

[0042] It should be noted that the mapping model between node voltage and control input constructed using the ARX system identification method can directly utilize the input and output data collected in the virtual power plant without establishing a complex physical mechanism model, thus capturing the dynamic relationship between node voltage and control input. Furthermore, the ARX model has a clear structure, low computational cost, and its parameters can be quickly obtained through parameter identification algorithms, facilitating updates and applications in real-time scheduling. In addition, by distinguishing between fast and slow dynamic variables and extracting key feature vectors, the ARX model can take into account both transient and steady-state voltage characteristics, thereby ensuring both the sensitivity of the control response and improving the accuracy of system operation prediction and regulation.

[0043] S2 decomposes the dynamic mapping model into fast dynamic sub-models and slow dynamic sub-models using singular perturbation theory.

[0044] In this example, the dynamic mapping model is decomposed into a fast dynamic sub-model and a slow dynamic sub-model using singular perturbation theory, specifically:

[0045] Small parameters of singular perturbations based on Lyapunov adaptive laws Adaptive adjustments are made, and the adjusted singular perturbation parameters are introduced into the dynamic mapping model to characterize the time scale difference between fast and slow dynamics.

[0046] like Figure 3 As shown, the mapping model is scaled according to the singular perturbation small parameter to obtain a fast dynamic subsystem containing fast dynamic variables and a slow dynamic subsystem containing slow dynamic variables.

[0047] The fast dynamic subsystem is subjected to variable freezing, that is, the slow dynamic variables are treated as constants, to obtain a fast dynamic sub-model that evolves only with a fast time scale;

[0048] The specific calculation formula for the fast dynamic sub-model is as follows:

[0049]

[0050] In the formula, For fast dynamic sub-model, These are fast dynamic variables, including inverter current, voltage transients, and other fast dynamic variables. Indicates a fast time scale. For slow-dynamic variables, such as energy storage SOC and photovoltaic power prediction, To control the input vector, This is a fast dynamic output function that maps fast dynamic variables, slow dynamic variables, and control input vectors to fast dynamic node voltage outputs.

[0051] The slow-dynamic subsystem is asymptotically expanded by treating the fast-dynamic variables as algebraic constraints that reach a steady state instantaneously, resulting in a slow-dynamic sub-model that evolves only with a slow time scale.

[0052] The specific calculation formula for the slow-dynamic sub-model is as follows:

[0053]

[0054] In the formula, The state vector output by the slow-dynamic sub-model includes the energy storage SOC (battery state of charge), the predicted photovoltaic power, and the power trend of flexible loads. For the algebraic solution of the fast dynamic subsystem, that is, after the dynamic subsystem reaches the quasi-steady-state condition, the fast variables... From slow variables and control input vector The function represented This is a slow dynamic system function.

[0055] The fast dynamic subsystem reflects the rapid transient regulation of voltage on a time scale, while the slow dynamic subsystem reflects the long-term evolution trend of voltage on a time scale. The singular perturbation small parameter is a dimensionless small quantity in singular perturbation theory used to represent the difference in time scale between the fast and slow dynamic states in the system.

[0056] Furthermore, based on the Lyapunov adaptive law, the singular perturbation small parameters are adaptively adjusted, specifically as follows:

[0057] Obtain the system control error and construct the Lyapunov function;

[0058] The Lyapunov function is calculated using the following formula:

[0059]

[0060] In the formula, For Lyapunov functions, This refers to the system control error, specifically the vector deviation between the node voltage output and the desired voltage. For singular perturbation small parameters, For the preset range of desired singular perturbation parameters, As a regulating factor, This is the quadratic form of the system control error, i.e., the sum of squares of the errors. This is the transpose of the system error vector.

[0061] The derivative of the Lyapunov function is calculated, and the result is used as a stability constraint. ;

[0062] The derivative is then performed, and the specific calculation formula is as follows:

[0063]

[0064] In the formula, The derivative of the Lyapunov function. This represents the rate of change of the system error over time. is the rate of change of the singular perturbation parameter over time.

[0065] An adaptive law is set based on stability constraints, and the values ​​of singular perturbation parameters are updated in real time according to the adaptive law. This value is then applied dynamically to ensure that the fast dynamic sub-model and the slow dynamic sub-model always maintain a reasonable separation of time scales.

[0066] The adaptive law is calculated using the following formula:

[0067]

[0068] In the formula, The regression vector, composed of preprocessed data, reflects the sensitivity of the system control error to singular perturbation parameters.

[0069] It should be noted that decomposing the mapping model into fast dynamic sub-models and slow dynamic sub-models can effectively distinguish between the fast transient response and the slow long-term evolution trend in the system, thereby achieving time-scale decoupling in analysis and control and simplifying system complexity. At the same time, based on the Lyapunov adaptive law, the small parameters of singular perturbation can be adjusted in real time to adaptively optimize the time-scale separation of the fast and slow dynamic subsystems according to the system control error. This ensures that the fast dynamic sub-model and the slow dynamic sub-model always maintain a reasonable scale difference, which not only guarantees the stability of the system but also improves control accuracy and dynamic performance, achieving a balance between fast response and robust control.

[0070] S3 establishes an LPV state-space model dependent on the adjustment parameters for the fast dynamic sub-model and outputs a fast-loop control law.

[0071] In this embodiment, an LPV state-space model dependent on the adjustment parameters is established for the fast dynamic sub-model, and a fast-loop control law is output, specifically as follows:

[0072] Based on the fast dynamic sub-model, the fast dynamic variables of the fast dynamic subsystem are obtained, including inverter current and voltage transients.

[0073] Obtain scheduling variables defined by uncertainties and time-varying characteristic parameters (such as node load changes, photovoltaic power output fluctuations, and battery SOC) in the fast dynamic subsystem;

[0074] Among them, scheduling variables refer to external or internal parameters that affect the dynamic characteristics of the system but are not control inputs themselves;

[0075] Based on fast dynamic variables and discretized scheduling variables, a local linearization of the fast dynamic subsystem is performed to establish an LPV (linear parameter variation) state-space model;

[0076] The specific calculation formula for the LPV state-space model is as follows:

[0077]

[0078]

[0079] In the formula, For fast dynamic state change rate, For scheduling variables, It is a linear mapping matrix of the fast dynamic state to its own changes. To control the linear influence matrix of the input on fast dynamic state changes, For fast dynamic output, For the linear mapping matrix of fast dynamic variables to the output, To control input For output The direct influence matrix.

[0080] For each LPV state-space model, the closed-loop pole positions are set according to the fast dynamic response performance, which includes the damping ratio and natural frequency.

[0081] The specific formula for calculating the closed-loop poles is as follows:

[0082]

[0083] In the formula, These are the two closed-loop poles of the system. For the damping ratio, is the natural frequency, and j is the imaginary unit.

[0084] The state feedback gain matrix is ​​obtained by combining the pole placement method with the solution of the closed-loop poles. Each gain matrix is ​​integrated to form a gain library;

[0085] The specific formula for calculating the gain matrix is ​​as follows:

[0086]

[0087] In the formula, Let be the linear mapping matrix of the fast dynamic state under the i-th scheduling variable to its own changes. Let be the linear influence matrix of the control input under the i-th scheduling variable on the fast dynamic state change. This is the nth closed-loop pole. For the closed-loop system matrix, Let be the set of eigenvalues ​​of the closed-loop system matrix.

[0088] The real-time control gain matrix is ​​obtained by selecting or multi-point interpolating the scheduling variables from the gain library. The fast-loop control law is obtained by combining the fast dynamic variables.

[0089] The specific calculation formula for the fast-loop control law is as follows:

[0090]

[0091] In the formula, This is a fast-loop control law.

[0092] Among them, the fast-loop control law refers to the rule or formula that maps fast dynamic variables to control inputs in real time. It is the core of the fast-loop controller to achieve rapid stabilization of node voltage. It is used to quickly adjust the node voltage, so that the system can quickly adjust the node voltage within the millisecond to second time scale, so that the voltage is close to the desired value.

[0093] S4. The residual signal is obtained by calculating the residual of the slow dynamic sub-model according to the fast loop control law, and the slow dynamic sub-model is then corrected.

[0094] In this embodiment, residual signals are obtained by calculating the residuals of the slow dynamic sub-model based on the fast-loop control law, and the slow dynamic sub-model is then corrected, specifically as follows:

[0095] Based on the difference between the actual node voltage under the action of the fast loop control law and the predicted output value of the slow dynamic sub-model, the slow dynamic prediction residual signal is obtained. The prediction residual signal represents the magnitude and direction of the slow dynamic sub-model output deviating from the actual operating state.

[0096] The predicted residual signal is initially filtered by wavelet transform to obtain the preliminary residual signal;

[0097] The initial residual signal is input into a low-pass filter for processing to suppress the interference of fast dynamic fluctuations on slow dynamic correction, so that the residual reflects the system error under the slow time scale and obtains the low-pass residual signal.

[0098] The specific calculation formula for the low-pass residual signal is as follows:

[0099] ,

[0100] In the formula, It is a low-pass residual signal. These are the filter coefficients. This is the residual vector for slow dynamic prediction.

[0101] The slow-dynamic sub-model is corrected based on the low-pass residual signal. The residual is then added as a feedback correction term to the state equation of the slow-dynamic sub-model to obtain the corrected slow-dynamic sub-model.

[0102] The specific calculation formula for the modified slow-dynamic sub-model is as follows:

[0103]

[0104] In the formula, This is the state vector output by the corrected slow-dynamic sub-model. The residual feedback gain matrix can be determined through empirical adjustment or robust optimization.

[0105] It should be noted that using the actual response of the fast-loop control law as the basis for slow-dynamic model correction can balance fast and slow dynamic decoupling and improve model prediction accuracy. Low-pass filtering avoids fast-dynamic fluctuation interference and improves the stability of slow-dynamic correction. The corrected slow-dynamic sub-model can be directly used for distributed optimization calculations, ensuring the stability and economy of virtual power plant node voltages on slow time scales.

[0106] Furthermore, the predicted residual signal is preliminarily filtered using wavelet transform to obtain a preliminary residual signal, specifically as follows:

[0107] Based on the characteristics of the virtual power plant system, a wavelet basis with orthogonality and good time-frequency localization is selected, such as Daubechies db4;

[0108] Obtain the sampling frequency and typical fast dynamic frequency, and calculate the wavelet decomposition level;

[0109] The specific formula for calculating the number of wavelet decomposition levels is as follows:

[0110]

[0111] In the formula, The wavelet decomposition level is denoted as . Sampling frequency, For fast dynamic typical frequency, This is a rounding function to ensure that the wavelet decomposition level is a practically feasible integer.

[0112] Based on the wavelet decomposition level, the prediction residual signal is decomposed into multi-scale wavelet coefficients to obtain wavelet coefficients for each level, including high-frequency coefficients and low-frequency coefficients.

[0113] Among them, wavelet coefficients represent the components of the prediction residual signal in different frequency bands;

[0114] Based on the characteristics of the fast and slow systems of the virtual power plant, the high-frequency coefficient is regarded as a fast dynamic disturbance, and the low-frequency coefficient is regarded as a slow dynamic component.

[0115] High-frequency coefficients are removed, and low-frequency coefficients are retained as the initial residual signal.

[0116] It should be noted that, based on the characteristics of the fast and slow systems of the virtual power plant, the high-frequency coefficients are regarded as fast dynamic disturbances and the low-frequency coefficients are regarded as slow dynamic components. This can be understood as using wavelet multi-scale decomposition to decompose the signal into different frequency bands. The fast dynamics are mainly manifested by high-frequency components (obvious transient fluctuations), while the slow dynamics are mainly manifested by low-frequency components (trend changes).

[0117] S5 establishes a global optimization objective function based on the modified slow dynamic sub-model, and solves for the reference values ​​of each node using a distributed algorithm.

[0118] In this example, a global optimization objective function is established based on the modified slow-dynamic sub-model, and the reference values ​​of each node are solved using a distributed algorithm, specifically:

[0119] Based on the state vector output by the corrected slow-dynamic sub-model and the voltage measurement value, a global optimization objective function is constructed and constraints are set. The global optimization objective function includes node voltage deviation, power loss and economic indicators (weighted sum of energy storage charging and discharging costs and renewable energy utilization rate). The constraints include node voltage constraints, energy storage SOC constraints and inverter power upper and lower limits.

[0120] The global optimization objective function is calculated using the following formula:

[0121]

[0122] In the formula, To optimize the objective function globally, The set of all nodes in the virtual power plant. Let be the actual voltage value of the i-th node at time t. This is the voltage reference value for the i-th node. , , These are the weighting coefficients, Let be the power loss of the i-th node at time t. Let be the economic performance index of the i-th node at time t.

[0123] The global optimization problem is broken down into local optimization subproblems for each node, resulting in the local objective function for each node. and neighbor node constraints;

[0124] The local objective function is calculated using the following formula:

[0125]

[0126] A distributed optimization algorithm is used, where each node iteratively optimizes based on the local objective function and the constraints of its neighboring nodes to obtain the local reference value for the current iteration. ;

[0127] Each node sends the calculated local reference value to its neighboring nodes through the communication network;

[0128] The local constraints are updated based on the local reference values, and local optimization is performed again to obtain a new round of reference values ​​until the preset global convergence condition is met. The final node reference values ​​of each node are then output, including node voltage reference values, current reference values, or power reference values.

[0129] The global convergence condition is as follows:

[0130]

[0131] In the formula, This is the preset convergence threshold.

[0132] The constraints are as follows:

[0133] Node voltage constraints:

[0134]

[0135] In the formula, Minimum allowable voltage, Maximum allowable voltage;

[0136] Energy storage SOC constraints:

[0137]

[0138] In the formula, For minimum energy storage SOC, For energy storage SOC, Maximum energy storage SOC;

[0139] Inverter power upper and lower limits:

[0140]

[0141] In the formula, For minimum inverter power, For inverter power, This represents the maximum inverter power.

[0142] It should be noted that through distributed iterative optimization, the virtual power plant achieves global voltage stability, minimum power loss, and economic optimization on a slow time scale. The distributed strategy reduces dependence on the central controller and improves system scalability and robustness.

[0143] S6 sends the node reference value to the local inverter and generates a control input signal in conjunction with the fast loop control law.

[0144] In this embodiment, the node reference value is sent to the local inverter and combined with the fast-loop control law to generate a control input signal, specifically:

[0145] The node reference values ​​are smoothed to generate smoothed reference values. ;

[0146] The smoothing process is specifically calculated using the following formula:

[0147]

[0148] In the formula, This is the smoothing coefficient.

[0149] The smoothing reference value is sent to the local inverter controller via the control communication interface;

[0150] After receiving the smoothed reference value, the local inverter performs data verification, which includes data integrity checks and data range validity checks to ensure that the reference value is within the inverter's executable range.

[0151] The control input signal is calculated based on the fast-loop control law and the verified smoothed reference value. ;

[0152]

[0153] In the formula, For the fast loop feedback gain matrix, This is the fast loop feedforward control variable.

[0154] Among them, the fast-loop control law compares the smooth reference value with the actual node state and generates inverter control commands within a time scale of milliseconds to seconds, thereby realizing rapid adjustment of node voltage or power.

[0155] After receiving the control input signal, the local inverter adjusts the output power, current, or voltage.

[0156] It should be noted that the smoothed reference value reduces the impact of sudden changes on the fast-loop control, improves system stability, and the fast-loop control law combined with the smoothed reference value enables rapid response of node voltage or power. Combined with slow dynamic optimization, the virtual power plant can take into account both fast and slow dynamic coordination and global optimization.

[0157] Reference Figure 2 As shown in the diagram, the hierarchical voltage coordination control system for a virtual power plant based on multi-agent technology provided by this invention includes a model building module, a model decomposition module, a fast-loop control law generation and model correction module, a slow-dynamic optimization module, and a control execution module. These modules are interconnected.

[0158] The model building module is used to acquire the raw operating data of distributed energy units in the virtual power plant and build a dynamic mapping model that represents the relationship between node voltage and control input.

[0159] The model decomposition module is used to decompose the dynamic mapping model into fast dynamic sub-models and slow dynamic sub-models;

[0160] The fast-loop control law generation and model correction module is used to generate fast-loop control laws based on fast-dynamic sub-models, and to perform residual calculation and online correction on slow-dynamic sub-models based on fast-loop control laws.

[0161] The slow-dynamic optimization module is used to establish a global optimization objective function based on the corrected slow-dynamic sub-model and solve for the reference values ​​of each node through a distributed algorithm.

[0162] The control execution module is used to send node reference values ​​to the local inverter and generate control input signals in conjunction with the fast-loop control law.

[0163] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0164] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0165] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0166] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0167] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0168] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A hierarchical voltage coordination control method for virtual power plants based on multi-agent systems, characterized in that, include: Obtain the raw operating data of distributed energy units in a virtual power plant and construct a dynamic mapping model representing the relationship between node voltage and control input; The dynamic mapping model is decomposed into a fast dynamic sub-model and a slow dynamic sub-model. Based on the fast dynamic sub-model, a fast-loop control law is generated, specifically as follows: Based on the fast dynamic sub-model, obtain the fast dynamic variables and discretized scheduling variables of the fast dynamic subsystem; Based on the local linearization of the fast dynamic subsystem using fast dynamic variables and scheduling variables, an LPV state-space model is established. For each LPV state-space model, the closed-loop pole locations are set based on the fast dynamic response performance; Based on the pole placement method, the corresponding state feedback gain matrix is ​​solved by combining the closed-loop pole location, and each group of gain matrices is integrated to form a gain library. The real-time control gain matrix is ​​obtained by selecting or multi-point interpolating the scheduling variables in the gain library, and the fast-loop control law is calculated by combining the fast dynamic variables. The residuals of the slow dynamic sub-model are calculated and corrected online based on the fast-loop control law, specifically as follows: Based on the difference between the actual node voltage under the action of the fast loop control law and the predicted output value of the slow dynamic sub-model, the slow dynamic prediction residual signal is obtained. The preliminary residual signal is obtained by performing preliminary screening of the predicted residual signal through wavelet transformation; The initial residual signal is input into a low-pass filter for processing to obtain the low-pass residual signal; The slow-dynamic sub-model is corrected based on the low-pass residual signal. The residual is added as a feedback correction term to the state equation of the slow-dynamic sub-model to obtain the corrected slow-dynamic sub-model. Based on the revised slow-dynamic sub-model, a global optimization objective function is established, and the reference values ​​of each node are solved using a distributed algorithm. The node reference value is sent to the local inverter and combined with the fast loop control law to generate the control input signal.

2. The hierarchical voltage coordination control method for virtual power plants based on multi-agent systems according to claim 1, characterized in that, The construction of the dynamic mapping model representing the relationship between node voltage and control input specifically involves: The raw operational data is preprocessed, and time series analysis is performed on the preprocessed data to distinguish system dynamic variables with different time scales and extract key feature vectors. Based on key feature vectors, the model structure is determined, which includes an autoregressive part related to historical outputs and an exogenous input part related to control inputs. Based on the determined model structure, the node voltage output time series and the input signal are used to construct a regression model; The model structure parameters of the regression model are solved by a parameter identification algorithm on the preprocessed data. Based on the solved parameters, a dynamic mapping model between the node voltage and the control input is generated.

3. The hierarchical voltage coordination control method for virtual power plants based on multi-agent systems according to claim 2, characterized in that, The decomposition of the dynamic mapping model into a fast dynamic sub-model and a slow dynamic sub-model is specifically as follows: Adaptive adjustment of singular perturbation small parameters based on Lyapunov adaptive law; The mapping model is scaled according to the adjusted singular perturbation small parameters to obtain a fast dynamic subsystem containing fast dynamic variables and a slow dynamic subsystem containing slow dynamic variables. By freezing variables in the fast dynamic subsystem, a fast dynamic sub-model that evolves with a fast time scale is obtained. By performing asymptotic expansion on the slow-dynamic subsystem, a slow-dynamic sub-model that evolves with a slow time scale is obtained.

4. The hierarchical voltage coordination control method for virtual power plants based on multi-agent systems according to claim 3, characterized in that, The adaptive adjustment of singular perturbation small parameters based on the Lyapunov adaptive law is specifically as follows: Obtain the system control error and construct the Lyapunov function; The derivative of the Lyapunov function is calculated, and the result is used as a stability constraint. An adaptive law is set based on stability constraints, and the values ​​of singular perturbation parameters are updated in real time according to the adaptive law for dynamic application.

5. The hierarchical voltage coordination control method for virtual power plants based on multi-agent systems according to claim 1, characterized in that, The preliminary residual signal is obtained by performing preliminary screening of the predicted residual signal through wavelet transform, specifically as follows: Obtain the sampling frequency and typical fast dynamic frequency, and calculate the wavelet decomposition level; Based on the wavelet decomposition level, the prediction residual signal is decomposed into multi-scale wavelet coefficients to obtain wavelet coefficients for each level, including high-frequency coefficients and low-frequency coefficients. Based on the characteristics of the fast and slow systems of the virtual power plant, the high-frequency coefficient is regarded as a fast dynamic disturbance, and the low-frequency coefficient is regarded as a slow dynamic component. High-frequency coefficients are removed, and low-frequency coefficients are retained as the initial residual signal.

6. The hierarchical voltage coordination control method for virtual power plants based on multi-agent systems according to claim 1, characterized in that, The process involves establishing a global optimization objective function based on the modified slow-dynamic sub-model, and solving for the reference values ​​of each node using a distributed algorithm. Specifically: Based on the state vector output by the corrected slow-dynamic sub-model and the voltage measurement value, a global optimization objective function is constructed and constraints are set. The global optimization objective function includes node voltage deviation, power loss and economic indicators, and the constraints include node voltage constraints, energy storage SOC constraints and inverter power upper and lower limits. The global optimization problem is broken down into local optimization subproblems for each node, resulting in the local objective function of each node and the constraints of its neighboring nodes. A distributed optimization algorithm is used to iteratively optimize each node based on the local objective function and the constraints of neighboring nodes, so as to obtain the local reference value of the current iteration. Each node sends its local reference value to its neighboring nodes via the communication network; The local constraints are updated based on the local reference values, and local optimization is performed again to obtain a new round of reference values ​​until the preset global convergence condition is met. The final node reference values ​​of each node are then output, including node voltage reference values, current reference values, or power reference values.

7. The hierarchical voltage coordination control method for virtual power plants based on multi-agent systems according to claim 6, characterized in that, The step of sending the node reference value to the local inverter and generating a control input signal in conjunction with the fast-loop control law is as follows: The node reference values ​​are smoothed to generate smooth reference values; The smoothing reference value is sent to the local inverter controller via the control communication interface; The local inverter performs data verification after receiving the smoothed reference value; The control input signal is calculated based on the fast-loop control law and the verified smooth reference value. After receiving the control input signal, the local inverter adjusts the output power, current, or voltage.

8. A system using the multi-agent-based virtual power plant hierarchical voltage coordination control method as described in any one of claims 1-7, characterized in that, include: The model building module is used to acquire the raw operating data of distributed energy units in the virtual power plant and build a dynamic mapping model that represents the relationship between node voltage and control input. The model decomposition module is used to decompose the dynamic mapping model into fast dynamic sub-models and slow dynamic sub-models; The fast-loop control law generation and model correction module is used to generate fast-loop control laws based on fast-dynamic sub-models, and to perform residual calculation and online correction on slow-dynamic sub-models based on fast-loop control laws. The slow-dynamic optimization module is used to establish a global optimization objective function based on the corrected slow-dynamic sub-model and solve for the reference values ​​of each node through a distributed algorithm. The control execution module is used to send node reference values ​​to the local inverter and generate control input signals in conjunction with the fast-loop control law.

Images