Physically embedded artificial intelligence differential equation dynamic element modeling method and device
By constructing differential equation models and training neural networks, the problems of insufficient speed, accuracy, and generalization ability in modeling dynamic components in power systems are solved, achieving efficient and accurate power system simulation and fault analysis.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- TSINGHUA UNIVERSITY
- Filing Date
- 2026-03-25
- Publication Date
- 2026-06-26
AI Technical Summary
Existing technologies struggle to effectively balance speed, accuracy, and generalization ability in dynamic component modeling within power systems. In particular, under scenarios involving multi-parameter coupling, multi-condition switching, and extreme fault disturbances, there is a trade-off between model robustness and computational efficiency, and a lack of physical consistency and interpretability.
A physical model of dynamic components of a power system described by differential equations is constructed. Dynamic trajectory data is generated by numerical integration and then randomly sliced as supervised learning samples to train a neural network model, ensuring that the model follows physical constraints and has strong generalization ability.
It achieves high-precision and high-speed dynamic component modeling, improves the calculation speed by two orders of magnitude, has strong generalization ability and physical consistency, and supports power grid simulation and rapid fault handling.
Smart Images
Figure CN122286151A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the fields of power system, new energy grid connection, and power electronics simulation technology, and relates to the direction of machine learning with physical constraints and efficient differential equation solving. In particular, it relates to a method and device for modeling dynamic components of artificial intelligence differential equations with physical embedding. Background Technology
[0002] In fields such as power systems and power electronics, dynamic component modeling is a core component of simulation analysis and system design. Although related research has been continuously advancing both domestically and internationally, it still faces key bottlenecks. Internationally, methods such as Physical Information Neural Networks (PINN) and deep learning-based surrogate models attempt to integrate data-driven approaches with physical priors. However, these methods often apply soft constraints through loss functions, neglecting the explicit embedding of the underlying physical structure, resulting in weak model interpretability and limited generalization ability. Domestically, while progress has been made in lightweight model deployment and the integration of physical constraints, the core contradiction of balancing speed, accuracy, and generalization ability in traditional methods has not yet been resolved.
[0003] Traditional numerical integrators require iterative solutions to differential equations at each time step, resulting in low computational efficiency and making it difficult to meet the massive computing power demands of rapid simulation of provincial power grids. Conventional data-driven models rely on external data fitting, lack the inherent constraints of physical equations, are prone to physical distortion, and have poor adaptability to untrained operating conditions. They cannot meet the stringent requirements of power systems for power supply continuity and operational stability, and are insufficient to support practical applications in critical scenarios such as new energy grid connection safety verification and rapid grid fault handling.
[0004] Current research, both domestically and internationally, has failed to effectively balance modeling speed, accuracy, and generalization ability. This is particularly evident in complex scenarios such as multi-parameter coupling, multi-condition switching, and extreme fault disturbances, where the contradiction between model robustness and computational efficiency becomes prominent. Furthermore, there are technical challenges such as the difficulty in balancing physical consistency and data utilization, and the mutual constraints between model interpretability and lightweight deployment. With the accelerated construction of new power systems, scenarios such as large-scale renewable energy grid integration and AC / DC hybrid power grid operation place increasingly stringent demands on the efficiency, accuracy, and robustness of modeling. Existing technologies are no longer adequate to meet industry development needs, necessitating a novel modeling method that breaks through traditional bottlenecks. Summary of the Invention
[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and propose a physically embedded artificial intelligence differential equation dynamic element modeling method and device. This invention is applied to scenarios for high-precision, high-speed modeling and simulation of dynamic elements in power systems, and can break through the bottlenecks of traditional dynamic element modeling in terms of speed, accuracy, and generalization ability.
[0006] A first aspect of this invention proposes a method for modeling dynamic elements of physically embedded artificial intelligence differential equations, comprising:
[0007] Construct a physical model of dynamic components of a power system described by a set of differential equations;
[0008] Define the range or probability distribution of the physical parameters, initial state variables, and input variables in the differential equation to establish a sampling space covering all operating conditions of the component;
[0009] Within the sampling space, the combination of the physical parameters, the initial state variables, and the input variables is sampled; for each combination, the differential equation is solved using a numerical integrator to generate a set of dynamic trajectory data reflecting the evolution of the state variables over time, ultimately resulting in multiple sets of dynamic trajectory data covering the sampling space.
[0010] By randomly slicing the dynamic trajectory data, the dynamic trajectory data is transformed into one-step inference training samples for supervised learning and formed into a training dataset.
[0011] An artificial intelligence model with a neural network as its core is constructed, and the artificial intelligence model is trained using the training dataset to obtain the final dynamic element artificial intelligence model.
[0012] In one specific embodiment of the present invention, it further includes:
[0013] The dynamic element is any one of generator-type elements, load-type elements, or power electronic interface elements;
[0014] For the aforementioned generator-type components, the differential equation is used to characterize the changes in power angle, rotational speed, and internal transient electromotive force.
[0015] For the aforementioned load-type components, the differential equation is used to describe the dynamic response characteristics of the load as voltage and frequency fluctuate;
[0016] For the power electronic interface element, the differential equation is used to reflect transient ride-through and control mode switching behavior.
[0017] In one specific embodiment of the present invention, it further includes:
[0018] The total differential equation corresponding to the physical model The expression is as follows:
[0019]
[0020] in, Represents state variables The derivative over time; For the linear part of the model, This represents the nonlinear part of the model. For the parameter part, For the input part;
[0021] The linear part consists of matrix-vector multiplication. Define, describe the linear dynamic characteristics of the system, where For one The state transition matrix, where n is dimensionality;
[0022] The nonlinear part contains nonlinear coupling terms between state variables, describing the nonlinear evolution of the system;
[0023] The parameter section describes the selected variable physical parameters. With state variables The coupling relationship reflects the influence of changes in physical parameters on system behavior;
[0024] The input section describes the external input variables. Its effect on system state.
[0025] In one specific embodiment of the present invention, it further includes:
[0026] The training samples consist of an input feature set and a corresponding output label set;
[0027] The input feature set consists of state variables, input variables, and corresponding key physical parameters at any discrete time point t.
[0028] The output label set consists of the state variables at a time step t+Δt after time point t, where Δt is the length of the slice.
[0029] The key physical parameters are one or more subsets of parameters selected from the set of physical parameters of the differential equation, or all parameters of the set of physical parameters.
[0030] A second aspect of the present invention provides a physically embedded artificial intelligence differential equation dynamic element modeling device, comprising:
[0031] The physical model building module is used to build physical models of dynamic components of a power system described by a set of differential equations;
[0032] The sampling space construction module is used to set the value range or probability distribution of physical parameters, initial state variables, and input variables in the differential equation, so as to establish a sampling space covering the full operating conditions of the component.
[0033] The dynamic trajectory acquisition module is used to sample the combination of the physical parameters, the initial state variables, and the input variables within the sampling space; for each combination, the differential equation is solved using a numerical integrator to generate a set of dynamic trajectory data reflecting the evolution of the state variables over time, and finally obtain multiple sets of dynamic trajectory data covering the sampling space.
[0034] The training dataset construction module is used to transform the dynamic trajectory data into one-step inference training samples for supervised learning and form a training dataset by randomly slicing the dynamic trajectory data.
[0035] The artificial intelligence model training module is used to construct an artificial intelligence model with a neural network as its core, and to train the artificial intelligence model using the training dataset to obtain the final dynamic element artificial intelligence model.
[0036] In one specific embodiment of the present invention, it further includes:
[0037] The dynamic element is any one of generator-type elements, load-type elements, or power electronic interface elements;
[0038] For the aforementioned generator-type components, the differential equation is used to characterize the changes in power angle, rotational speed, and internal transient electromotive force.
[0039] For the aforementioned load-type components, the differential equation is used to describe the dynamic response characteristics of the load as voltage and frequency fluctuate;
[0040] For the power electronic interface element, the differential equation is used to reflect transient ride-through and control mode switching behavior.
[0041] In one specific embodiment of the present invention, it further includes:
[0042] The total differential equation corresponding to the physical model The expression is as follows:
[0043]
[0044] in, Represents state variables The derivative over time; For the linear part of the model, This represents the nonlinear part of the model. For the parameter part, For the input part;
[0045] The linear part consists of matrix-vector multiplication. Define, describe the linear dynamic characteristics of the system, where For one The state transition matrix, where n is dimensionality;
[0046] The nonlinear part contains nonlinear coupling terms between state variables, describing the nonlinear evolution of the system;
[0047] The parameter section describes the selected variable physical parameters. With state variables The coupling relationship reflects the influence of changes in physical parameters on system behavior;
[0048] The input section describes the external input variables. Its effect on system state.
[0049] In one specific embodiment of the present invention, it further includes:
[0050] The training samples consist of an input feature set and a corresponding output label set;
[0051] The input feature set consists of state variables, input variables, and corresponding key physical parameters at any discrete time point t.
[0052] The output label set consists of the state variables at a time step t+Δt after time point t, where Δt is the length of the slice.
[0053] The key physical parameters are one or more subsets of parameters selected from the set of physical parameters of the differential equation, or all parameters of the set of physical parameters.
[0054] A third aspect of the present invention provides an electronic device comprising:
[0055] At least one processor; and a memory communicatively connected to said at least one processor;
[0056] The memory stores instructions that can be executed by the at least one processor, and the instructions are configured to execute the above-described method for modeling dynamic elements of physical embedded artificial intelligence differential equations.
[0057] A fourth aspect of the present invention provides a computer-readable storage medium storing computer instructions for causing the computer to execute the above-described method for modeling dynamic elements of physically embedded artificial intelligence differential equations.
[0058] The features and beneficial effects of this invention are as follows:
[0059] 1) High accuracy and strong physical consistency: The core of this invention lies in fitting the state transition mapping of the exact solution of the physical equation, rather than simply fitting external data. The training data is generated entirely from the physical equation, which fundamentally ensures that the model's output can closely approximate the results of traditional numerical simulations, fully comply with the inherent constraints of the physical equation, and avoid the physical distortion problem of conventional AI models.
[0060] 2) High computation speed: Once the model training is completed, the inference of its dynamic behavior can be completed with only one efficient neural network forward computation. Compared with the traditional numerical integral solver, which requires repeated iterations at each time step, the computation speed can be improved by more than two orders of magnitude.
[0061] 3) Strong generalization ability: This invention generates massive training data covering a combination of various physical parameters, initial states and input signals, and uses key parameters as explicit inputs to the model, so that the final model has strong generalization ability and robust adaptability to working conditions that do not appear directly in the training. Attached Figure Description
[0062] Figure 1 This is a flowchart of a physical embedded artificial intelligence differential equation dynamic element modeling method according to an embodiment of the present invention.
[0063] Figure 2 This is a schematic diagram of the structure of an artificial intelligence model in a specific embodiment of the present invention. Detailed Implementation
[0064] This invention proposes a method and apparatus for modeling dynamic elements of artificial intelligence differential equations with physical embedding, which will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0065] A first aspect of this invention proposes a method for modeling dynamic elements of physically embedded artificial intelligence differential equations, comprising:
[0066] Construct a physical model of dynamic components of a power system described by a set of differential equations;
[0067] Define the range or probability distribution of the physical parameters, initial state variables, and input variables in the differential equation to establish a sampling space covering all operating conditions of the component;
[0068] Within the sampling space, the combination of the physical parameters, the initial state variables, and the input variables is sampled; for each combination, the differential equation is solved using a numerical integrator to generate a set of dynamic trajectory data reflecting the evolution of the state variables over time, ultimately resulting in multiple sets of dynamic trajectory data covering the sampling space.
[0069] By randomly slicing the dynamic trajectory data, the dynamic trajectory data is transformed into one-step inference training samples for supervised learning and formed into a training dataset.
[0070] An artificial intelligence model with a neural network as its core is constructed, and the artificial intelligence model is trained using the training dataset to obtain the final dynamic element artificial intelligence model.
[0071] In one specific embodiment of the present invention, the overall process of the physically embedded artificial intelligence differential equation dynamic element modeling method is as follows: Figure 1 As shown, the steps include:
[0072] 1) Construct a physical model of the dynamic components of the power system that is precisely described by a set of differential equations.
[0073] In this embodiment, the dynamic element is defined as a physical entity with time-varying dynamic characteristics in a power system, and is the core unit in power system stability analysis; in some specific embodiments of the present invention, it includes at least one of the following: synchronous generator, asynchronous motor, integrated load (including dynamic load and static load), high-voltage DC converter station or new energy converter.
[0074] Among them, generator components: their physical model follows the rotor kinematic equation (oscillation equation) and the law of electromagnetic induction, and the differential equation is used to describe the changes in their power angle, speed and internal transient electromotive force.
[0075] Load-type components (such as dynamic or static loads): their physical models are based on induction motor models or polynomial models (ZIP models), and differential equations describe the dynamic response characteristics of the load as voltage and frequency fluctuate.
[0076] Power electronic interface components: Their physical models describe the dynamic interaction between the internal control logic of the converter station or inverter and the external grid, and their differential equations reflect their transient ride-through and control mode switching behavior.
[0077] In this embodiment, the physical model of the dynamic element to be modeled is precisely described by a set of differential equations, the total differential equation being... The expression is as follows:
[0078]
[0079] in, Represents state variables The derivative over time (i.e.) ); For the linear part of the model, This represents the nonlinear part of the model. For the parameter part, This is the input section.
[0080] In a specific embodiment of the present invention, the detailed definitions of each part in the above expression are as follows:
[0081] Linear part ( ): Matrix-vector product Defined, describing the inherent linear dynamic characteristics of a system, where, For one The state transition matrix, Let n be the state variable. The dimension of.
[0082] Nonlinear part ( ): Contains nonlinear coupling terms between state variables (such as trigonometric functions, multiplicative terms, etc.), describing the complex nonlinear evolution of the system.
[0083] Parameter section ( ): Describes the selected variable physical parameters With state variables The coupling relationship reflects the influence of changes in physical parameters on system behavior.
[0084] Input section ( ): Describes external input variables How (such as control commands, external disturbances) affect the system state.
[0085] 2) Based on the results of step 1), set the physical parameters (i.e., those in step 1). The initial state variables and the range or probability distribution of the input variables are used to establish a sampling space covering all operating conditions of the components.
[0086] The physical parameters in this step are those in step 1). The initial state variable is The initial value, the input variable is the one in step 1). .
[0087] 3) By performing diverse combination sampling within the sampling space established in step 2), the differential equations of step 1) are solved using a high-precision numerical integrator for each combination of physical parameters, initial state, and input variables obtained from each sampling, generating a set of high-dimensional dynamic trajectory data that reflects the evolution of state variables over time, and finally obtaining multiple sets of dynamic trajectory data covering the entire sampling space.
[0088] 4) Perform random slicing on the multiple sets of dynamic trajectory data generated in step 3) to transform the long-term trajectory data into massive one-step inference training samples for supervised learning and form a training dataset; wherein each training sample consists of an input feature set and a corresponding output label set.
[0089] In this embodiment, the input feature set is a set consisting of state variables, input variables, and corresponding key physical parameters at any discrete time point t.
[0090] The output label set is defined as the state variable at a time step t+Δt after time point t. Δt is the length of the slice.
[0091] Furthermore, the random slicing process decomposes the long-term dynamic trajectory into a large number of independent time segments with a single discrete time step interval. Its core purpose is to transform the time-dependent trajectory fitting problem into a supervised learning regression problem that can be parallelized on a large scale and trained with high efficiency.
[0092] Furthermore, in this embodiment, key physical parameters composed of one or more variable physical parameters are explicitly used as one of the input features of the subsequent artificial intelligence model. The key physical parameters are one or more subsets of parameters selected from the set of physical parameters of the differential equation in step 1) according to the modeling requirements of the artificial intelligence model, or all parameters of the set of physical parameters, so that the trained artificial intelligence model can directly deduce the dynamic behavior under different physical parameter configurations, thereby having a strong parameter generalization ability.
[0093] 5) Construct an artificial intelligence model with a neural network as its core. The dimension of the input layer of this model matches the dimension of the input feature set defined in step 4), and the dimension of the output layer matches the dimension of the output label set defined in step 4). The structure of this model is designed to learn and fit the state transition mapping from input features to output labels defined by the differential equation.
[0094] Furthermore, the internal network structure of the artificial intelligence model in this embodiment can be modularly designed according to the physical structure of the differential equation. For example, the equations corresponding to slow mechanical dynamics and fast electrical dynamics can be fitted by different network modules.
[0095] 6) Use the training dataset constructed in step 4) to train the artificial intelligence model constructed in step 5).
[0096] In this embodiment, the internal weights of the model are iteratively adjusted using optimization algorithms such as gradient descent to minimize the error between the model's predicted output label and the actual output label. Training is stopped when the error meets the preset accuracy requirements, and the final dynamic element artificial intelligence model is obtained.
[0097] Furthermore, in this embodiment, the loss function during model training may include not only the prediction error term, but also one or more physical regularization terms based on the residuals of the differential equation, in order to further enhance the physical consistency of the model during training.
[0098] Furthermore, the trained physical embedded AI model can be directly embedded into new power system electromagnetic transient or electromechanical transient simulation software as a high-precision, high-efficiency dynamic proxy model, replacing traditional numerical integration modules based on complex physical differential equations (such as Runge-Kutta solvers). In actual large-scale power grid node simulations, new energy power plant grid connection safety verification, and rapid simulations of extreme fault scenarios, this model can directly output the next-moment dynamic response of components through a single forward network inference. This not only completely avoids the computational bottleneck and convergence problems caused by repeated iterations in traditional numerical methods, increasing the single-step inference speed by more than two orders of magnitude; more importantly, because the model explicitly embeds underlying physical constraints and key physical parameters... This ensures that the model still possesses strong generalization ability and physical consistency when facing complex operating conditions and parameter fluctuations that were not directly involved in the training, effectively supporting real-time online safety and stability analysis of the digital twin power grid.
[0099] The method described in this embodiment will be further explained in detail below with reference to a specific example.
[0100] In this embodiment, for a 10th-order nonlinear generator system, the physically embedded artificial intelligence differential equation dynamic element modeling method includes the following steps:
[0101] 1) Determine the physical model of the dynamic components of the power system that is accurately described by a set of differential equations.
[0102] This embodiment selects a typical sixth-order detailed model of a synchronous generator and a fourth-order excitation regulation system in a power system to construct a set of tenth-order nonlinear differential equations to describe the electromagnetic and mechanical dynamic processes of the generator after disturbance. Its dynamic behavior is described by a set of ordinary differential equations. Precise description.
[0103] The system's variables are structured as follows:
[0104] State variables (10-dimensional): This represents the internal state of the synchronous generator and its excitation system, such as power angle, rotor speed, and transient electromotive force. Let i represent the i-th dimension of the state variable, where i = 1, 2, ..., 10.
[0105] Parameter variables (8-dimensional): This represents the physical characteristic parameters of the generator, such as the inertia constant. Damping coefficient Parameters such as reactance, among which, This represents the i-th dimension parameter in the parameter variable, where i = 1, 2, ..., 8.
[0106] Input variables at time t (2D): This represents an external signal acting on the system, such as an increase in mechanical power or an excitation reference voltage. Let represent the i-th dimension of the input variables at time t, where i = 1, 2.
[0107] Furthermore, in this embodiment, the total differential equation The expression is as follows:
[0108]
[0109] in:
[0110] Linear part ( ) by matrix-vector product Definition. Wherein, matrix It is a sparse matrix whose non-zero elements are mainly distributed on the 2×2 diagonal blocks.
[0111] Nonlinear part ( It mainly contains nonlinear coupling terms of state variables, and its vector form is:
[0112]
[0113] Parameter section ( Description of 8-dimensional variable physical parameters With state variables The coupling relationship, in vector form, is as follows:
[0114]
[0115] Input section ( Describe the external input signal How it acts on the system, its vector form is:
[0116]
[0117] In this embodiment, by adding the above four parts, the final system of differential equations is obtained as follows:
[0118]
[0119] 2) Based on the results of step 1), set the range or probability distribution of physical parameters, initial state variables and input variables to establish a sampling space covering all operating conditions of the component.
[0120] In this embodiment, the value ranges of the 10-dimensional state variables, 8-dimensional parameter variables, and 2-dimensional input variables are defined.
[0121] 3) By performing diverse combination sampling within the sampling space established in step 2), the differential equation in step 1) is solved using a high-precision numerical integrator for each combination of physical parameters, initial state variables, and input variables obtained from each sampling, generating a corresponding set of dynamic trajectory time series data, and finally obtaining multiple sets of dynamic trajectory time series data covering the entire sampling space.
[0122] 4) Randomly slice the multiple sets of dynamic trajectory time series data generated in step 3) to transform the long-term trajectory data into massive one-step inference training samples for supervised learning and form a training dataset; wherein each training sample consists of an input feature set and a corresponding output label set.
[0123] In this embodiment, given the input variable (e.g., mechanical power increment), the 10th-order equation established in step 1) is simulated using a numerical integration algorithm to generate state variables under different initial conditions and different combinations of physical parameters. High-dimensional dynamic trajectory data (such as power angle, speed, voltage, etc.) that evolves over time; in this embodiment, it is a 10th-order dynamic trajectory.
[0124] Then, the above 10th-order dynamic trajectory is randomly sliced to construct a structure based on the current state. Physical parameters Current input Given the input feature set, the state at the next time step The training samples are used for single-step transitions to labels, and these samples are combined to form the training dataset.
[0125] 5) Build an artificial intelligence model.
[0126] In this embodiment, a multi-layer deep neural network is built as an artificial intelligence model, and its input layer dimension matches the input feature set dimension defined in step 4). Its internal network structure is modularly decoupled according to the mechanism coupling relationship of the 10th order equation, so that the network can learn state transition mapping.
[0127] Figure 2 This is a schematic diagram of the structure of an artificial intelligence model in a specific embodiment of the present invention. For example... Figure 2As shown, in this embodiment, the artificial intelligence model adopts a fully connected neural network structure. Its input layer has a dimension of 21 (including 10-dimensional state, 8-dimensional physical parameters, 2-dimensional input, and 1-dimensional time step); the hidden layer contains two layers with 64 and 128 neurons respectively, and uses the SiLU activation function for nonlinear mapping; the output layer has a dimension of 10, corresponding to 10-dimensional state, and the output layer directly outputs the state vector of the next time step.
[0128] 6) Use the training dataset constructed in step 4) to train the artificial intelligence model constructed in step 5).
[0129] In this embodiment, the training dataset from step 4) is used to iteratively train the artificial intelligence. The weights are adjusted through an optimization algorithm to minimize the error between the predicted output and the true label until the accuracy requirements are met. After training, the artificial intelligence model can quickly extrapolate the dynamic behavior of the 10th-order nonlinear system through single-step iterations under unknown physical parameters.
[0130] Model training and generalization ability testing
[0131] Furthermore, according to the method described in this embodiment, multiple sets of dynamic trajectories with a duration of 10 seconds (each set containing 500 discrete points) are randomly sliced to construct a one-step inference training dataset containing "input features at time t" and "output labels at time t+Δt", which is used for supervised learning training of the artificial intelligence model.
[0132] Generalization ability test: Select one or more sets of dynamic trajectories (each set containing 1500 discrete points) that were not used for training and lasted for 30 seconds as the test set. During the test simulation, uniformly distributed random noise in the interval [-0.5, 0.5] is superimposed on all dimensions (20 items in total) of the model's three types of input vectors—state vectors, physical parameter vectors, and external input signals (provided by input_interpolator)—to evaluate the model's generalization ability and robustness to parameter uncertainties and unseen conditions.
[0133] Test Results: The AI model constructed in this embodiment consistently maintains a mean squared error (MSE) of 10⁻⁻⁶ between its single-step prediction output and the ground truth in the aforementioned generalization test scenario. 4 The test results show that the model has strong generalization ability and robustness to operating conditions and parameter perturbations that were not included in the training.
[0134] To implement the above embodiments, a second aspect of the present invention proposes a physically embedded artificial intelligence differential equation dynamic element modeling device, comprising:
[0135] The physical model building module is used to build physical models of dynamic components of a power system described by a set of differential equations;
[0136] The sampling space construction module is used to set the value range or probability distribution of physical parameters, initial state variables, and input variables in the differential equation, so as to establish a sampling space covering the full operating conditions of the component.
[0137] The dynamic trajectory acquisition module is used to sample the combination of the physical parameters, the initial state variables, and the input variables within the sampling space; for each combination, the differential equation is solved using a numerical integrator to generate a set of dynamic trajectory data reflecting the evolution of the state variables over time, and finally obtain multiple sets of dynamic trajectory data covering the sampling space.
[0138] The training dataset construction module is used to transform the dynamic trajectory data into one-step inference training samples for supervised learning and form a training dataset by randomly slicing the dynamic trajectory data.
[0139] The artificial intelligence model training module is used to construct an artificial intelligence model with a neural network as its core, and to train the artificial intelligence model using the training dataset to obtain the final dynamic element artificial intelligence model.
[0140] In one specific embodiment of the present invention, it further includes:
[0141] The dynamic element is any one of generator-type elements, load-type elements, or power electronic interface elements;
[0142] For the aforementioned generator-type components, the differential equation is used to characterize the changes in power angle, rotational speed, and internal transient electromotive force.
[0143] For the aforementioned load-type components, the differential equation is used to describe the dynamic response characteristics of the load as voltage and frequency fluctuate;
[0144] For the power electronic interface element, the differential equation is used to reflect transient ride-through and control mode switching behavior.
[0145] In one specific embodiment of the present invention, it further includes:
[0146] The total differential equation corresponding to the physical model The expression is as follows:
[0147]
[0148] in, Represents state variables The derivative over time; For the linear part of the model, This represents the nonlinear part of the model. For the parameter part, For the input part;
[0149] The linear part consists of matrix-vector multiplication. Define, describe the linear dynamic characteristics of the system, where For one The state transition matrix, where n is dimensionality;
[0150] The nonlinear part contains nonlinear coupling terms between state variables, describing the nonlinear evolution of the system;
[0151] The parameter section describes the selected variable physical parameters. With state variables The coupling relationship reflects the influence of changes in physical parameters on system behavior;
[0152] The input section describes the external input variables. Its effect on system state.
[0153] In one specific embodiment of the present invention, it further includes:
[0154] The training samples consist of an input feature set and a corresponding output label set;
[0155] The input feature set consists of state variables, input variables, and corresponding key physical parameters at any discrete time point t.
[0156] The output label set consists of the state variables at a time step t+Δt after time point t, where Δt is the length of the slice.
[0157] The key physical parameters are one or more subsets of parameters selected from the set of physical parameters of the differential equation, or all parameters of the set of physical parameters.
[0158] This enables high-precision, high-speed modeling and simulation of dynamic components in power systems, overcoming the bottlenecks of traditional dynamic component modeling in terms of speed, accuracy, and generalization ability.
[0159] To implement the above embodiments, a third aspect of the present invention provides an electronic device, comprising:
[0160] At least one processor; and a memory communicatively connected to said at least one processor;
[0161] The memory stores instructions that can be executed by the at least one processor, and the instructions are configured to execute the above-described method for modeling dynamic elements of physical embedded artificial intelligence differential equations.
[0162] To implement the above embodiments, a fourth aspect of the present invention provides a computer-readable storage medium storing computer instructions for causing the computer to execute the above-described method for modeling dynamic elements of physically embedded artificial intelligence differential equations.
[0163] It should be noted that the computer-readable medium described in this disclosure can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this disclosure, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this disclosure, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.
[0164] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device. The aforementioned computer-readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to perform a physically embedded artificial intelligence differential equation dynamic element modeling method according to the above embodiments.
[0165] Computer program code for performing the operations of this disclosure can be written in one or more programming languages or a combination thereof, including object-oriented programming languages such as Java, Smalltalk, and C++, and conventional procedural programming languages such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).
[0166] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.
[0167] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.
[0168] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the function involved, as will be understood by those skilled in the art to which embodiments of this application pertain.
[0169] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which programs can be printed, because programs can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.
[0170] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.
[0171] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.
[0172] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.
[0173] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.
Claims
1. A method for modeling dynamic elements of physically embedded artificial intelligence differential equations, characterized in that, include: Construct a physical model of dynamic components of a power system described by a set of differential equations; Define the range or probability distribution of the physical parameters, initial state variables, and input variables in the differential equation to establish a sampling space covering all operating conditions of the component; Within the sampling space, the combination of the physical parameters, the initial state variables, and the input variables is sampled; For each combination, the differential equation is solved using a numerical integrator to generate a set of dynamic trajectory data reflecting the evolution of state variables over time, ultimately resulting in multiple sets of dynamic trajectory data covering the sampling space. By randomly slicing the dynamic trajectory data, the dynamic trajectory data is transformed into one-step inference training samples for supervised learning and formed into a training dataset. An artificial intelligence model with a neural network as its core is constructed, and the artificial intelligence model is trained using the training dataset to obtain the final dynamic element artificial intelligence model.
2. The method according to claim 1, characterized in that, Also includes: The dynamic element is any one of generator-type elements, load-type elements, or power electronic interface elements; For the aforementioned generator-type components, the differential equation is used to characterize the changes in power angle, rotational speed, and internal transient electromotive force. For the aforementioned load-type components, the differential equation is used to describe the dynamic response characteristics of the load as voltage and frequency fluctuate; For the power electronic interface element, the differential equation is used to reflect transient ride-through and control mode switching behavior.
3. The method according to claim 1, characterized in that, Also includes: The total differential equation corresponding to the physical model The expression is as follows: in, Represents state variables The derivative over time; For the linear part of the model, This represents the nonlinear part of the model. For the parameter part, For the input part; The linear part consists of matrix-vector multiplication. Define, describe the linear dynamic characteristics of the system, where For one The state transition matrix, where n is dimensionality; The nonlinear part contains nonlinear coupling terms between state variables, describing the nonlinear evolution of the system; The parameter section describes the selected variable physical parameters. With state variables The coupling relationship reflects the influence of changes in physical parameters on system behavior; The input section describes the external input variables. Its effect on system state.
4. The method according to claim 3, characterized in that, Also includes: The training samples consist of an input feature set and a corresponding output label set; The input feature set consists of state variables, input variables, and corresponding key physical parameters at any discrete time point t. The output label set consists of the state variables at a time step t+Δt after time point t, where Δt is the length of the slice. The key physical parameters are one or more subsets of parameters selected from the set of physical parameters of the differential equation, or all parameters of the set of physical parameters.
5. A physically embedded artificial intelligence differential equation dynamic element modeling device, characterized in that, include: The physical model building module is used to build physical models of dynamic components of a power system described by a set of differential equations; The sampling space construction module is used to set the value range or probability distribution of physical parameters, initial state variables, and input variables in the differential equation, so as to establish a sampling space covering the full operating conditions of the component. The dynamic trajectory acquisition module is used to sample the combination of the physical parameters, the initial state variables, and the input variables within the sampling space; For each combination, the differential equation is solved using a numerical integrator to generate a set of dynamic trajectory data reflecting the evolution of state variables over time, ultimately resulting in multiple sets of dynamic trajectory data covering the sampling space. The training dataset construction module is used to transform the dynamic trajectory data into one-step inference training samples for supervised learning and form a training dataset by randomly slicing the dynamic trajectory data. The artificial intelligence model training module is used to construct an artificial intelligence model with a neural network as its core, and to train the artificial intelligence model using the training dataset to obtain the final dynamic element artificial intelligence model.
6. The apparatus according to claim 5, characterized in that, Also includes: The dynamic element is any one of generator-type elements, load-type elements, or power electronic interface elements; For the aforementioned generator-type components, the differential equation is used to characterize the changes in power angle, rotational speed, and internal transient electromotive force. For the aforementioned load-type components, the differential equation is used to describe the dynamic response characteristics of the load as voltage and frequency fluctuate; For the power electronic interface element, the differential equation is used to reflect transient ride-through and control mode switching behavior.
7. The apparatus according to claim 5, characterized in that, Also includes: The total differential equation corresponding to the physical model The expression is as follows: in, Represents state variables The derivative over time; For the linear part of the model, This represents the nonlinear part of the model. For the parameter part, For the input part; The linear part consists of matrix-vector multiplication. Define, describe the linear dynamic characteristics of the system, where For one The state transition matrix, where n is dimensionality; The nonlinear part contains nonlinear coupling terms between state variables, describing the nonlinear evolution of the system; The parameter section describes the selected variable physical parameters. With state variables The coupling relationship reflects the influence of changes in physical parameters on system behavior; The input section describes the external input variables. Its effect on system state.
8. The apparatus according to claim 7, characterized in that, Also includes: The training samples consist of an input feature set and a corresponding output label set; The input feature set consists of state variables, input variables, and corresponding key physical parameters at any discrete time point t. The output label set consists of the state variables at a time step t+Δt after time point t, where Δt is the length of the slice. The key physical parameters are one or more subsets of parameters selected from the set of physical parameters of the differential equation, or all parameters of the set of physical parameters.
9. An electronic device, characterized in that, include: At least one processor; And, a memory communicatively connected to the at least one processor; The memory stores instructions executable by the at least one processor, the instructions being configured to perform the method described in any one of claims 1-4.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to perform the method according to any one of claims 1-4.