A method and device for transient stability assessment of a multi-vsg parallel system

Abstract

The application discloses a transient stability evaluation method and device for a multi-VSG parallel system, and the method comprises the following steps: obtaining a power system model and a transient energy function according to the system; the transient energy function comprises a kinetic energy function and a potential energy function; obtaining a critical stable energy value according to the power system model and the transient energy function combined with a BCU method; constructing a first model for obtaining a dynamic coupling relationship between VSGs based on a PINN network; training the first model according to a preset data set to obtain a second model; solving the transient energy function according to a real-time state vector of the system, and solving a path-related term in the transient energy function combined with the second model to obtain a real-time energy value; if the real-time energy value is greater than the critical stable energy value, it is determined that the system is unstable, otherwise, it is determined that the system is stable.

Description

Technical Field

[0001] This invention relates to the field of electrical automation technology, and in particular to a method and apparatus for transient stability assessment of multi-VSG parallel systems. Background Technology

[0002] With the continuous and rapid growth of the installed capacity of new energy power generation, the dynamic characteristics and stability mechanism of the power system are gradually changing, causing the operating characteristics of the power system to evolve from traditional synchronous machine-dominated to power electronic.

[0003] Against this backdrop, the penetration rate of grid-connected equipment with VSG (Virtual Synchronous Generator) as the core interface (such as wind power, photovoltaic power, energy storage systems, and flexible DC transmission) in the power grid has increased dramatically. VSG devices are typically connected to the AC grid on a large scale through parallel connections, forming complex multi-VSG parallel systems. On the one hand, unlike traditional synchronous generators with large inertia and natural synchronization characteristics, the dynamic behavior of a VSG is entirely determined by its internal control strategy. It has small inertia, fast response, strong nonlinearity, and lacks the inertial support provided by physical rotating mass. When the power system encounters large disturbances, complex dynamic interactions may occur between multiple VSGs and between VSGs and the remaining synchronous machine network, leading to new types of instability problems. The transient stability mechanism of multi-VSG parallel systems is fundamentally different from that of traditional synchronous machine systems, resulting in a significant increase in the difficulty of assessment. On the other hand, when the classical transient energy function method applicable to traditional power systems is applied to multi-VSG parallel systems, due to the high nonlinearity of the system model and control loop, the path-related terms representing nonconservative forces in the energy function are difficult to express analytically and be calculated quickly. In other words, the construction of the key energy function faces fundamental difficulties, making it difficult for this type of method to be used for the actual evaluation of multi-VSG parallel systems.

[0004] Therefore, a new technical solution is urgently needed to address the technical problem of how to perform transient stability assessment on multi-VSG parallel systems. Summary of the Invention

[0005] This invention provides a method and apparatus for transient stability assessment of multiple VSG parallel systems, thereby solving the technical problem of how to perform transient stability assessment on multiple VSG parallel systems.

[0006] To achieve the above objectives, the present invention provides a transient stability evaluation method for a multi-VSG parallel system, comprising:

[0007] The system model and transient energy function are obtained based on the system. The transient energy function includes the kinetic energy function and the potential energy function. The critical stable energy value is obtained by combining the power system model and transient energy function with the BCU method. A first model is constructed based on the PINN network to obtain the dynamic coupling relationship between VSGs. The first model is trained based on a preset dataset to obtain a second model. The transient energy function is solved based on the real-time state vector of the system, and the path-related terms in the transient energy function are solved by combining the second model to obtain the real-time energy value. If the real-time energy value is greater than the critical stable energy value, the system is determined to be unstable; otherwise, it is determined to be stable.

[0008] Preferably, the power system model and transient energy function obtained from the system include:

[0009] A power system model is constructed based on a multi-VSG parallel system. The power system model includes a VSG model and a system power flow model. The VSG model includes active-frequency and reactive-voltage droop control of the VSG. The system power flow model is based on Kirchhoff's current conservation law and includes power flow equations for loads and multi-VSG parallel systems.

[0010] Ignoring damping-related terms, the transient energy function of the multi-VSG parallel system is obtained from the power system model. The transient energy function includes the kinetic energy function and the potential energy function. The kinetic energy function includes the system kinetic energy. The potential energy function includes the potential energy accumulated by the VSG, the potential energy accumulated by the load, the reactive potential energy accumulated by each VSG and the load, and the electromagnetic potential energy stored by the VSG and the line impedance on the grid side.

[0011] Preferably, the VSG model includes:

[0012] ;

[0013] in, , and They represent the first The power angle, angular velocity, and terminal voltage amplitude of the VSG; , , , and They represent the first The actual active power, active power reference value, actual reactive power, reactive power reference value, and reference voltage value of each VSG; , , and They represent the first The virtual moment of inertia, active-frequency damping coefficient, voltage regulation inertia coefficient, and reactive-voltage droop coefficient of the VSG; This represents the reference angular frequency of the power grid.

[0014] Preferably, the system power flow model includes:

[0015] ;

[0016] ;

[0017] ;

[0018] ;

[0019] ;

[0020] in, and These are the voltage and phase at the point of common coupling, respectively; Main grid voltage; and The first The equivalent connection reactance from the VSG output terminal to the point of common coupling and the equivalent connection reactance from the point of common coupling to the main grid; and These are the active power and reactive power of the load, respectively. and These represent the active and reactive power inputs from the main power grid to the point of common coupling, respectively. Indicates the total number of VSG units; It represents the imaginary unit.

[0021] Preferably, the kinetic energy function includes the system kinetic energy; the potential energy function includes the potential energy accumulated by the VSG, the potential energy accumulated by the load, the reactive potential energy accumulated by each VSG and load, and the electromagnetic potential energy stored by the VSG and the line impedance on the grid side.

[0022] The system's kinetic energy is determined by the kinetic energy function. The manifestations include:

[0023] ;

[0024] The potential energy function includes a first sub-potential energy function, a second sub-potential energy function, a third sub-potential energy function, and a fourth sub-potential energy function, including:

[0025] First sub-potential function The potential energy accumulated by VSG is represented by:

[0026] ;

[0027] Second Sub-potential Function The potential energy accumulated by a load includes:

[0028] ;

[0029] Third Sub-potential Function Used to represent the reactive potential energy accumulated by the VSG and the load, including:

[0030] ;

[0031] Fourth Sub-potential Function The electromagnetic potential energy stored in the VSG and the line impedance on the grid side is used to represent the following:

[0032] ;

[0033] in, and They represent the first The current state power angle variable of the VSG and the power angle at the stable equilibrium point after fault clearance; and They represent the first The current state voltage variables of the VSG and the voltage at the stable equilibrium point after fault clearance; The voltage at the point of common coupling, representing the stable equilibrium point after a fault; , , , , as well as They represent the first The power angle difference variable between the inverter and the point of common coupling, and the current state. The power angle difference between the inverter and the point of common coupling, and the stable equilibrium point after fault clearance. The variables include the power angle difference between the inverter and the point of common coupling, the power angle difference between the main grid and the point of common coupling, the power angle difference between the main grid and the point of common coupling in the current state, and the power angle difference between the main grid and the point of common coupling at the stable equilibrium point after fault clearance.

[0034] Preferably, the critical stable energy value obtained by combining the power system model and transient energy function with the BCU method includes:

[0035] The power system model is integrated using the state vector of the multi-VSG parallel system before the fault as the initial point. The potential energy function is calculated simultaneously, and the first data point of each integration is recorded. Integration ends when the potential energy function first reaches a local maximum, resulting in the first dataset. The first dataset includes data pairs of reactive potential energy and state vector. , This represents the state vector at the k-th integration. Indicates the k-th integration. The third sub-potential function of the VSG ;

[0036] When the potential energy function reaches a local maximum for the first time, it is determined that the system trajectory has reached the boundary of the stable region, and the state point at this time is recorded as the escape point;

[0037] By setting the inertial terms in the motion equations of each VSG rotor in the VSG model to zero, we obtain the dimensionless model:

[0038] ;

[0039] The system power flow model is recalculated based on the multi-VSG parallel system after the fault to obtain the stable equilibrium point after the fault. Using the stable equilibrium point as the search starting point and reference point, and the escape point as the initial iteration value, the dimensionality-reduced model is numerically solved based on the Newton-Raphson method. The iteration format includes:

[0040] ;

[0041] in, The equations are for a reduced-order system. Its Jacobian matrix; For the first The state vector of the next iteration; For the first The state vector of the next iteration;

[0042] When the iteration converges, the equilibrium point is obtained. The dominant disequilibrium point is identified; substituting the dominant disequilibrium point into the transient energy function yields the critical stable energy value. .

[0043] Preferably, the first model for obtaining the dynamic coupling relationship between VSGs based on the PINN network includes:

[0044] The first model adopts a graph physical information neural network architecture, which includes a graph encoder, global pooling, and regression layers. The graph encoder performs message passing and aggregation of node and neighbor information through a preset number of graph convolution operations. The global pooling and regression layers perform global summation pooling on the final layer node features of the graph encoder to obtain a graph-level representation vector, and then regress the target value through a fully connected network.

[0045] The input to the first model is a state vector and a boundary information set. The state vector includes pre-selected key state parameters of the multi-VSG parallel system. The boundary information set includes data used to construct adaptive graph edge weights based on the real-time power angle difference and voltage difference between VSGs. The data in the boundary information set is obtained based on the state vector. The output of the first model is the path-dependent term in the potential energy function. , No. The third sub-potential function of the VSG Represented as .

[0046] Preferably, the second model is obtained by training the first model based on a preset dataset, including:

[0047] The first dataset is standardized and divided into a training set, a validation set, and a test set. The training set covers preset typical operating modes and fault types. The validation set is used for hyperparameter tuning. The test set includes preset extreme scenarios to assess the model's generalization ability.

[0048] The first model is trained using the training set, validation set, and test set combined with the first loss function to obtain the second model; the first loss function considers data fitting loss, physical equation fitting loss, and knowledge fitting loss.

[0049] Preferably, the transient energy function is solved based on the system's real-time state vector, and the path-dependent terms in the transient energy function are solved using the second model to obtain the real-time energy values, including:

[0050] Obtain the real-time state vector of the multi-VSG parallel system; combine the real-time state vector with the first sub-potential function. Second sub-potential function Fourth sub-potential function and kinetic energy function The non-path-dependent term of the transient energy function is calculated; the path-dependent term of the transient energy function is calculated using a neural network based on the real-time state vector and the second model; the non-path-dependent term and the path-dependent term are summed to obtain the real-time energy value.

[0051] The present invention also provides a transient stability evaluation method and apparatus for a multi-VSG parallel system. The apparatus for the method of the present invention includes a first module, a second module, a third module and a fourth module.

[0052] The first module is used to obtain the power system model and transient energy function based on the system; the transient energy function includes the kinetic energy function and the potential energy function; the second module is used to obtain the critical stable energy value based on the power system model and transient energy function combined with the BCU method; the third module is used to construct a first model based on the PINN network to obtain the dynamic coupling relationship between VSGs; the first model is trained based on the preset dataset to obtain the second model;

[0053] The fourth module is used to solve the transient energy function based on the system's real-time state vector, and to solve the path-dependent terms in the transient energy function in combination with the second model to obtain the real-time energy value. If the real-time energy value is greater than the critical stable energy value, the system is determined to be unstable; otherwise, it is determined to be stable.

[0054] The present invention has the following beneficial effects:

[0055] This invention presents a transient stability assessment method for multi-VSG parallel systems. Based on a neural network-assisted energy function, it assesses the transient stability of such systems. Specifically, it solves for the path-dependent terms in the potential energy function using a neural network, enabling rapid transient stability assessment of complex power electronic systems and effectively improving the transient stability analysis and control capabilities of high-proportion renewable energy power systems. This method compresses the computation time of traditional time-domain simulations (which takes several minutes) to the millisecond level, achieving real-time, continuous quantitative perception of transient stability and providing a crucial time window for emergency control decisions.

[0056] The transient stability evaluation device for a multi-VSG parallel system of the present invention, used in the method of the present invention, has the same beneficial effects as the method of the present invention.

[0057] In addition to the objectives, features, and advantages described above, the present invention has other objectives, features, and advantages. The invention will now be described in further detail with reference to the accompanying drawings. Attached Figure Description

[0058] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0059] Figure 1 This is a schematic diagram of the method flow of a preferred embodiment of the present invention. Detailed Implementation

[0060] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings, but the present invention can be implemented in many different ways as defined and covered by the claims.

[0061] See Figure 1 In a preferred embodiment of the present invention, a transient stability evaluation method for a multi-VSG parallel system is provided, comprising:

[0062] S1. Obtain the power system model and transient energy function based on the system; the transient energy function includes the kinetic energy function and the potential energy function; obtain the critical stable energy value based on the power system model and transient energy function combined with the BCU method.

[0063] In a preferred embodiment of the present invention, obtaining the power system model and transient energy function based on the system includes:

[0064] A power system model is constructed based on a multi-VSG parallel system. The power system model includes a VSG model and a system power flow model. The VSG model includes active-frequency and reactive-voltage droop control of the VSG. The system power flow model is based on Kirchhoff's current conservation law and includes power flow equations for loads and multi-VSG parallel systems.

[0065] Ignoring damping-related terms, the transient energy function of the multi-VSG parallel system is obtained from the power system model. The transient energy function includes the kinetic energy function and the potential energy function. The kinetic energy function includes the system kinetic energy. The potential energy function includes the potential energy accumulated by the VSG, the potential energy accumulated by the load, the reactive potential energy accumulated by each VSG and the load, and the electromagnetic potential energy stored by the VSG and the line impedance on the grid side.

[0066] To ensure the model accurately reflects the core control characteristics of grid-connected inverters, a VSG model is established, including active-frequency (Pf) and reactive-voltage (QV) droop control, comprising:

[0067] ;

[0068] in, , and They represent the first The power angle, angular velocity, and terminal voltage amplitude of the VSG; , , , and They represent the first The actual active power, active power reference value, actual reactive power, reactive power reference value, and reference voltage value of each VSG; , , and They represent the first The virtual moment of inertia, active-frequency damping coefficient, voltage regulation inertia coefficient, and reactive-voltage droop coefficient of the VSG; This represents the reference angular frequency of the power grid, typically 100π.

[0069] The system power flow model includes:

[0070] ;

[0071] ;

[0072] ;

[0073] ;

[0074] ;

[0075] in, and These are the voltage and phase at the point of common coupling, respectively; Main grid voltage; and The first The equivalent connection reactance from the VSG output terminal to the point of common coupling and the equivalent connection reactance from the point of common coupling to the main grid; and These are the active power and reactive power of the load, respectively. and These represent the active and reactive power inputs from the main power grid to the point of common coupling, respectively. Indicates the total number of VSG units; It represents the imaginary unit.

[0076] Drawing upon and extending the concept of constructing transient energy functions for classical synchronous machine systems, an energy function is constructed for multi-VSG systems. To maintain the conservatism of the system evaluation, damping-related terms are neglected. The transient energy function consists of two parts: a kinetic energy function and a potential energy function.

[0077] Kinetic energy function include:

[0078] ;

[0079] Potential energy function include:

[0080] ;

[0081] in, , They represent the first The current power angle variable of the VSG and the power angle at the stable equilibrium point after the fault.

[0082] Due to the potential energy function It cannot be calculated directly. The integral based on the outputs of each VSG can be converted into an integral based on the power flow variables of the transmission line. Potential energy function The process of simplifying the formula is expressed as follows:

[0083] ;

[0084] ;

[0085] ;

[0086] ;

[0087] in, and These represent the current state power angle variable at the point of common connection and the power angle at the stable equilibrium point after the fault, respectively. and They represent the first The current state voltage variables of the VSG and the voltage at the stable equilibrium point after a fault; and These represent the current state voltage variable at the point of common coupling and the voltage at the stable equilibrium point after a fault, respectively. This indicates the total load in the system.

[0088] The following terms, which cannot be directly calculated, were identified during the simplification of the above formulas:

[0089] ;

[0090] ;

[0091] Summarizing the above, we obtain the first sub-potential function, the second sub-potential function, the third sub-potential function, and the fourth sub-potential function:

[0092] First sub-potential function The potential energy accumulated by VSG is represented by:

[0093] ;

[0094] Second Sub-potential Function The potential energy accumulated by a load includes:

[0095] ;

[0096] Third Sub-potential Function Used to represent the reactive potential energy accumulated by the VSG and the load, including:

[0097] ;

[0098] Fourth Sub-potential Function The electromagnetic potential energy stored in the VSG and the line impedance on the grid side is used to represent the following:

[0099] ;

[0100] in, and They represent the first The current state power angle variable of the VSG and the power angle at the stable equilibrium point after fault clearance; and They represent the first The current state voltage variables of the VSG and the voltage at the stable equilibrium point after fault clearance; The voltage at the point of common coupling, representing the stable equilibrium point after a fault; , , , , as well as They represent the first The power angle difference variable between the inverter and the point of common coupling, and the current state. The power angle difference between the inverter and the point of common coupling, and the stable equilibrium point after fault clearance. The variables include the power angle difference between the inverter and the point of common coupling, the power angle difference between the main grid and the point of common coupling, the power angle difference between the main grid and the point of common coupling in the current state, and the power angle difference between the main grid and the point of common coupling at the stable equilibrium point after fault clearance.

[0101] In a preferred embodiment of the present invention, the critical stability energy value is obtained by combining the power system model and transient energy function with the BCU (the boundary of stability region based controlling UEP method) method, including:

[0102] The power system model is integrated using the state vector of the multi-VSG parallel system before the fault as the initial point. The potential energy function is calculated simultaneously, and the first data point of each integration is recorded. Integration ends when the potential energy function first reaches a local maximum, resulting in the first dataset. The first dataset includes data pairs of reactive potential energy and state vector. , This represents the state vector at the k-th integration. Indicates the k-th integration. The third sub-potential function of the VSG ;

[0103] When the potential energy function reaches a local maximum for the first time, it is determined that the system trajectory has reached the boundary of the stable region, and the state point at this time is recorded as the escape point;

[0104] Set the inertial terms in the motion equations of each VSG rotor in the VSG model to zero (specifically, set the inertial terms in the VSG model to zero). , and (Setting to zero), we obtain the dimensionality reduction model:

[0105] ;

[0106] The system power flow model is recalculated based on the multi-VSG parallel system after the fault to obtain the stable equilibrium point after the fault. Using the stable equilibrium point as the search starting point and reference point, and the escape point as the initial iteration value, the dimensionality-reduced model is numerically solved based on the Newton-Raphson method. The iteration format includes:

[0107] ;

[0108] in, The equations are for a reduced-order system. Its Jacobian matrix; For the first The state vector of the next iteration; For the first The state vector of the next iteration;

[0109] When the iteration converges, the equilibrium point is obtained. The dominant imbalance point.

[0110] Substituting the dominant imbalance point into the transient energy function yields the critical stable energy value. Its core lies in calculating the potential energy, specifically the energy distribution from the stable equilibrium point after the fault to the dominant disequilibrium point. The integral of this equation yields the energy in that state, i.e., the critical stable energy value. .

[0111] S2. Construct a first model based on the PINN network to obtain the dynamic coupling relationship between VSGs; train the first model according to the preset dataset to obtain the second model.

[0112] In a preferred embodiment of the present invention, constructing a first model based on PINN (Physics-Informed Neural Networks) for obtaining the dynamic coupling relationship between VSGs includes:

[0113] This paper addresses the problem of analytically calculating path-dependent potential terms in the transient energy function by introducing a PINN neural network. A surrogate model integrating data-driven approaches and physical laws is constructed to directly learn the complex mapping from the system's high-dimensional states to the complete transient energy function, thereby achieving rapid online stability assessment. Specifically, this includes:

[0114] The first model adopts a graph physical information neural network architecture, which includes a graph encoder, global pooling, and regression layers.

[0115] The graph encoder performs message passing and aggregation of node and neighbor information through a preset number of layers of graph convolution operations. Layer nodes The features are updated as follows:

[0116] ;

[0117] in For nodes Adjacent nodes; and These are learnable weights; For activation functions; For nodes In the Layer feature representation; For nodes In the Layer feature representation; For nodes neighboring nodes In the The feature representation of the layer, where .

[0118] Global pooling and regression layers perform global summation pooling on the final layer node features of the graph encoder to obtain the graph-level representation vector. Then, the target value is regressed through a fully connected network, which is expressed as:

[0119] ;

[0120] in, These are the output layer weights, the fully connected layer weight matrix, the bias term, and the output bias, respectively. This structure ensures that the model can explicitly perceive and process the system topology and interactions.

[0121] The input to the first model is a state vector and a boundary information set; the state vector includes pre-selected key state parameters of the multi-VSG parallel system; the boundary information set includes data used to construct adaptive graph edge weights based on the real-time power angle difference and voltage difference between VSGs, and the data in the boundary information set is obtained based on the state vector; the output of the first model is the path-dependent term in the potential energy function. , No. The third sub-potential function of the VSG Represented as .

[0122] State vector Composed of all the key state variables of the VSG in an ordered manner, represented as:

[0123] ;

[0124] in, This indicates the transpose; Represents a matrix, The superscript 3m indicates the dimension of the matrix.

[0125] The essence of boundary information set is to dynamically quantify and embed the nonlinear coupling strength between multiple converters, thereby providing graph neural networks with topologically guided boundary information that directly reflects the key physical essence of the system's transient interactions; boundary information set elements in The calculation method is as follows:

[0126] ;

[0127] in, and Both represent the sequence index of VSG.

[0128] Neural network via the first model The output of the first The third sub-potential function of the VSG Represented as:

[0129] ;

[0130] Neural network of the first model The learning objective is to establish the path-related integral values ​​from the state vector X to the unanalyzable path. The direct functional relationship is obtained, thus bypassing the dependence on dynamic trajectory integrals.

[0131] In a preferred embodiment of the present invention, training a first model based on a preset dataset to obtain a second model includes:

[0132] The first dataset is standardized and divided into a training set, a validation set, and a test set. The training set covers preset typical operating modes and fault types. The validation set is used for hyperparameter tuning. The test set includes preset extreme scenarios to evaluate the model's generalization ability.

[0133] The first model is trained using the training set, validation set, and test set combined with the first loss function to obtain the second model; the first loss function considers data fitting loss, physical equation fitting loss, and knowledge fitting loss.

[0134] In a preferred embodiment of the present invention, in order to make the neural network of the first model The output strictly conforms to the physical meaning of the energy function, and the first loss function is defined for training:

[0135] ;

[0136] in, , , For data fitting loss, physical equation fitting loss, and knowledge fitting loss, , These are the weights for the two types of fitting losses, respectively.

[0137] The data fitting loss is used to utilize large-scale datasets generated through offline simulation. Constrain the network output to approximate the true value:

[0138] ;

[0139] The physical equation fitting loss directly injects the physics of the system's control equations as soft constraints into the network training.

[0140] ;

[0141] For knowledge fitting loss, this loss establishes an absolute, unambiguous physical reference point for the energy function:

[0142] ;

[0143] in, This is the state vector corresponding to the stable equilibrium point (SEP) of the system; The symbol for partial derivative is , which represents the partial derivative with respect to the corresponding variable; For the first The first integral step Voltage amplitude at each node; When the k-th integration step is the first The reactive power of each node; The total number of samples in the training dataset;

[0144] The definite physical prior that "the path-dependent potential energy is zero at the stable equilibrium point" is directly applied to the network as a strong constraint. This loss term does not depend on noisy numerical labels, but rather mandates that the network output must converge to this physical truth, thereby ensuring that the entire surrogate energy function has a correct absolute benchmark and eliminating possible systematic biases.

[0145] In a preferred embodiment of the present invention, the first model is trained using an adaptive moment estimation optimizer.

[0146] S3. Solve for the transient energy function based on the system's real-time state vector, and combine this with the second model to solve for the path-dependent terms in the transient energy function, obtaining the real-time energy value. If the real-time energy value is greater than the critical stable energy value, the system is determined to be unstable; otherwise, it is determined to be stable. S3 specifically includes:

[0147] Obtain the real-time state vector of the multi-VSG parallel system; combine the real-time state vector with the first sub-potential function. Second sub-potential function Fourth sub-potential function and kinetic energy function The non-path-dependent term of the transient energy function is calculated; the path-dependent term of the transient energy function is calculated using a neural network based on the real-time state vector and the second model; the real-time energy value is obtained by summing the non-path-dependent term and the path-dependent term. The stability of the system can be determined by the relationship between the real-time energy value and the critical stable energy value. The calculation specifically includes:

[0148] Real-time energy value include:

[0149] ;

[0150] Calculate the transient stability margin of the current state. Stability margin:

[0151] ;

[0152] The system is based on margin Real-time stability assessment of size: when A value greater than 0 and having a certain margin is considered stable; when When the value approaches or falls below zero, an instability warning is immediately triggered.

[0153] This invention presents a transient stability assessment method for multi-VSG parallel systems. Based on a neural network-assisted energy function, it assesses the transient stability of such systems. Specifically, it solves for the path-dependent terms in the potential energy function using a neural network, enabling rapid transient stability assessment of complex power electronic systems and effectively improving the transient stability analysis and control capabilities of high-proportion renewable energy power systems. This method compresses the computation time of traditional time-domain simulations (which takes several minutes) to the millisecond level, achieving real-time, continuous quantitative perception of transient stability and providing a crucial time window for emergency control decisions.

[0154] In a preferred embodiment of the present invention, a transient stability evaluation method and apparatus for a multi-VSG parallel system is also provided. The apparatus for the method of the present invention includes a first module, a second module, a third module and a fourth module.

[0155] The first module is used to obtain the power system model and transient energy function based on the system; the transient energy function includes the kinetic energy function and the potential energy function; the second module is used to obtain the critical stable energy value based on the power system model and transient energy function combined with the BCU method; the third module is used to construct a first model based on the PINN network to obtain the dynamic coupling relationship between VSGs; the first model is trained based on the preset dataset to obtain the second model;

[0156] The fourth module is used to solve the transient energy function based on the system's real-time state vector, and to solve the path-dependent terms in the transient energy function in combination with the second model to obtain the real-time energy value. If the real-time energy value is greater than the critical stable energy value, the system is determined to be unstable; otherwise, it is determined to be stable.

[0157] The transient stability evaluation device for a multi-VSG parallel system of the present invention, used in the method of the present invention, has the same beneficial effects as the method of the present invention.

[0158] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for evaluating the transient stability of a multi-VSG parallel system, characterized in that, include: Based on the system described, a power system model and transient energy function are obtained; The transient energy function includes a kinetic energy function and a potential energy function; The critical stable energy value is obtained by combining the power system model and the transient energy function with the BCU method; a first model for obtaining the dynamic coupling relationship between VSGs is constructed based on the PINN network. The first model is trained according to a preset dataset to obtain the second model; the transient energy function is solved according to the real-time state vector of the system, and the path-related term in the transient energy function is solved in combination with the second model to obtain the real-time energy value; if the real-time energy value is greater than the critical stable energy value, the system is determined to be unstable, otherwise it is determined to be stable. The critical stable energy value obtained based on the power system model and the transient energy function combined with the BCU method includes: The power system model is integrated using the state vector of the multi-VSG parallel system before the fault as the initial point. The potential energy function is calculated synchronously, and the first data point of each integration is recorded. Integration ends when the potential energy function first reaches a local maximum, yielding the first dataset. The first dataset includes a pair of reactive potential energy and state vector data. , This represents the state vector during the k-th integration. Indicates the k-th integration. The third sub-potential function of the VSG The third sub-potential function W 23 Used to represent the reactive potential energy accumulated by the VSG and the load; When the potential energy function reaches a local maximum for the first time, it is determined that the system trajectory has reached the boundary of the stable region, and the state point at this time is recorded as the escape point; By setting the inertial terms in the motion equations of each VSG rotor in the VSG model to zero, we obtain the dimensionless model: ； in, and They represent the first The actual active power, active power reference value, actual reactive power, reactive power reference value, reference voltage value, terminal voltage amplitude, and reactive-voltage droop coefficient of each VSG; Based on the post-fault multi-VSG parallel system, the system power flow model is recalculated to obtain the stable equilibrium point after the fault. Using the stable equilibrium point as the search starting point and reference point, and the escape point as the initial iteration value, the dimensionality-reduced model is numerically solved based on the Newton-Raphson method. The iteration format includes: ； in, The equations are for a reduced-order system. Its Jacobian matrix; For the first The state vector of the next iteration; For the first The state vector of the next iteration; When the iteration converges, the equilibrium point is obtained. The dominant imbalance point is identified; substituting the dominant imbalance point into the transient energy function yields the critical stable energy value. .

2. The transient stability evaluation method for a multi-VSG parallel system according to claim 1, characterized in that, The power system model and transient energy function obtained from the system include: The power system model is constructed based on the multi-VSG parallel system; the power system model includes a VSG model and a system power flow model; the VSG model includes active-frequency and reactive-voltage droop control of the VSG; the system power flow model is based on Kirchhoff's current conservation law and includes power flow equations for loads and the multi-VSG parallel system. Ignoring damping-related terms, the transient energy function of the multi-VSG parallel system is obtained based on the power system model. The transient energy function includes a kinetic energy function and a potential energy function. The kinetic energy function includes the system kinetic energy. The potential energy function includes the potential energy accumulated by the VSGs, the potential energy accumulated by the loads, the reactive potential energy accumulated by each VSG and the loads, and the electromagnetic potential energy stored by the VSGs and the line impedance on the grid side.

3. The transient stability evaluation method for a multi-VSG parallel system according to claim 2, characterized in that, The VSG model includes: ； in, , and They represent the first The power angle, angular velocity, and terminal voltage amplitude of the VSG; , , , and They represent the first The actual active power, active power reference value, actual reactive power, reactive power reference value, and reference voltage value of each VSG; , , and They represent the first The virtual moment of inertia, active-frequency damping coefficient, voltage regulation inertia coefficient, and reactive-voltage droop coefficient of the VSG; This represents the reference angular frequency of the power grid.

4. The transient stability evaluation method for a multi-VSG parallel system according to claim 3, characterized in that, The system power flow model includes: ； ； ； ； ； in, and These are the voltage and phase at the point of common coupling, respectively; Main grid voltage; and The first The equivalent connection reactance from the VSG output terminal to the point of common coupling and the equivalent connection reactance from the point of common coupling to the main grid; and These are the active power and reactive power of the load, respectively. and These represent the active and reactive power inputs from the main power grid to the point of common coupling, respectively. Indicates the total number of VSG units; It represents the imaginary unit.

5. The transient stability evaluation method for a multi-VSG parallel system according to claim 4, characterized in that, The kinetic energy function includes the system kinetic energy; the potential energy function includes the potential energy accumulated by the VSG, the potential energy accumulated by the load, the reactive potential energy accumulated by each VSG and the load, and the electromagnetic potential energy stored by the VSG and the line impedance on the grid side. The kinetic energy of the system is given by the kinetic energy function. The manifestations include: ； The potential energy function includes a first sub-potential energy function, a second sub-potential energy function, a third sub-potential energy function, and a fourth sub-potential energy function, including: First sub-potential function The potential energy accumulated by VSG is represented by: ； Second sub-potential function The potential energy accumulated by a load includes: ； The third sub-potential function Used to represent the reactive potential energy accumulated by the VSG and the load, including: ； The fourth sub-potential function The electromagnetic potential energy stored in the VSG and the line impedance on the grid side is used to represent the following: ； in, and They represent the first The current state power angle variable of the VSG and the power angle at the stable equilibrium point after fault clearance; and They represent the first The current state voltage variables of the VSG and the voltage at the stable equilibrium point after fault clearance; The voltage at the point of common coupling, representing the stable equilibrium point after a fault; , , , , as well as They represent the first The power angle difference variable between the inverter and the point of common coupling, and the current state. The power angle difference between the inverter and the point of common coupling, and the stable equilibrium point after fault clearance. The variables include the power angle difference between the inverter and the point of common coupling, the power angle difference between the main grid and the point of common coupling, the power angle difference between the main grid and the point of common coupling in the current state, and the power angle difference between the main grid and the point of common coupling at the stable equilibrium point after fault clearance.

6. The transient stability evaluation method for a multi-VSG parallel system according to claim 5, characterized in that, The first model for obtaining the dynamic coupling relationship between VSGs based on the PINN network includes: The first model adopts a graph physical information neural network architecture, which includes a graph encoder, a global pooling layer, and a regression layer. The graph encoder performs message passing and aggregation of node and neighbor information through a preset number of graph convolution operations. The global pooling and regression layers perform global summation pooling on the final layer node features of the graph encoder to obtain a graph-level representation vector, and then regress the target value through a fully connected network. The input to the first model is a state vector and a boundary information set; the state vector includes pre-selected key state parameters of the multi-VSG parallel system; the boundary information set includes data used to construct adaptive graph edge weights based on the real-time power angle difference and voltage difference between VSGs, and the data in the boundary information set is obtained based on the state vector; the output of the first model is the path-related term in the potential energy function, i.e., the third sub-potential energy function. , No. The third sub-potential function of the VSG Represented as .

7. The transient stability assessment method for a multi-VSG parallel system according to claim 6, characterized in that, The step of training the first model based on a preset dataset to obtain the second model includes: The first dataset is standardized and divided into a training set, a validation set, and a test set. The training set covers preset typical operating modes and fault types. The validation set is used for hyperparameter tuning. The test set includes preset extreme scenarios to assess the model's generalization ability. The first model is trained using the training set, validation set, and test set combined with the first loss function to obtain the second model; the first loss function considers data fitting loss, physical equation fitting loss, and knowledge fitting loss.

8. The transient stability evaluation method for a multi-VSG parallel system according to claim 7, characterized in that, The transient energy function is solved based on the real-time state vector of the system, and the path-dependent terms in the transient energy function are solved using the second model to obtain the real-time energy values, including: Obtain the real-time state vector of the multi-VSG parallel system; combine the real-time state vector with the first sub-potential function The second sub-potential function The fourth sub-potential function and the kinetic energy function The non-path-dependent term of the transient energy function is calculated; the path-dependent term of the transient energy function is calculated using a neural network based on the real-time state vector and the second model; the non-path-dependent term and the path-dependent term are summed to obtain the real-time energy value.

9. An apparatus for transient stability evaluation of a multi-VSG parallel system, used in the method described in any one of claims 1 to 8, characterized in that, The device includes a first module, a second module, a third module, and a fourth module; The first module is used to obtain a power system model and a transient energy function based on the system; the transient energy function includes a kinetic energy function and a potential energy function; the second module is used to obtain the critical stable energy value based on the power system model and the transient energy function combined with the BCU method; the third module is used to construct a first model based on the PINN network to obtain the dynamic coupling relationship between VSGs; and to train the first model according to a preset dataset to obtain a second model; The fourth module is used to solve the transient energy function based on the real-time state vector of the system, and to solve the path-dependent term in the transient energy function in combination with the second model to obtain the real-time energy value; if the real-time energy value is greater than the critical stable energy value, the system is determined to be unstable, otherwise it is determined to be stable.

Images