Method for constructing power system simulation operator based on physical embedded neural network
By decomposing the high-dimensional nonlinear equations of the power system into multiple submodules and using physical embedded neural networks for training and optimization, the computing bottleneck in large-scale power system simulation is solved, and efficient and real-time simulation analysis is achieved.
Patent Information
- Application Number
- CN202510476573.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-16
- Publication Date
- 2025-08-15
AI Technical Summary
The existing physical embedded neural network methods are inefficient in processing large-scale power systems and are difficult to meet the needs of real-time simulation. Especially when dealing with high-dimensional nonlinear complex systems, traditional overall modeling methods cannot fully utilize the advantages of parallel computing, resulting in too long calculations.
The high-dimensional nonlinear dynamic equation system of the power system is decomposed to form multiple interrelated submodules, and each submodule is trained and optimized through physically embedded neural networks, fused the output of the submodule to obtain a global solution, and adaptive optimization methods are used to dynamically adjust the network parameters, and the parallel computing and distributed computing frameworks are used to improve simulation efficiency.
It significantly improves the solution speed and accuracy of power system simulation, and can effectively solve the calculation bottleneck problem of traditional methods in large-scale power system simulation, maintaining high accuracy and real-time performance.
Smart Images
Figure CN120493688A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of power systems, and in particular to a method for constructing a power system simulation operator based on a physical embedded neural network. Background Art
[0002] As power systems continue to develop and become increasingly complex in the context of large-scale renewable energy integration, highly automated dispatching, and distributed energy management, dynamic process simulation and analysis of power systems face increasing challenges. While traditional numerical simulation methods (such as those based on differential equations) are widely used in many scenarios, their computational complexity and time increase exponentially when solving large-scale, high-dimensional nonlinear systems, making them unable to meet the real-time and accuracy requirements of modern power systems in practical applications. The limitations of these traditional methods, particularly the nonlinear dynamic coupling issues inherent in large-scale power networks, make it difficult for them to meet the efficiency and accuracy requirements for rapid simulation and real-time decision-making when dealing with complex systems.
[0003] In recent years, artificial intelligence-based neural network methods have gradually attracted the attention of researchers, especially the physical embedding neural network (PINN) technology, which can improve the training efficiency and accuracy of the model while ensuring physical consistency by embedding physical laws into the loss function of the neural network. However, when dealing with high-dimensional nonlinear complex systems such as power systems, the existing physical embedding neural network methods still face the challenges of low computational efficiency and difficulty in effectively decomposing the system model. Especially when dealing with the numerous mutually coupled dynamic processes in the power system, the traditional overall modeling method often cannot fully utilize the parallel computing advantages of the network, resulting in excessively long computing time and unable to meet the needs of real-time simulation of large-scale power systems. The existing power system dynamic process simulation analysis has the problems of high computational complexity and slow solution speed. Summary of the Invention
[0004] The technical problem to be solved by the present invention is to overcome the shortcomings of the existing technology and provide a method for constructing a power system simulation operator based on a physical embedded neural network. The present invention significantly improves the simulation accuracy and solution speed, and can effectively solve the computational bottleneck problem of traditional methods in large-scale power system simulation.
[0005] The present invention adopts the following technical solutions to solve the above technical problems:
[0006] A method for constructing a power system simulation operator based on a physical embedded neural network proposed in the present invention includes:
[0007] Modeling the high-dimensional nonlinear dynamic equations of the power system, we can obtain the power system state equation:
[0008] Decomposing the power system state equation based on computing resource conditions to form a plurality of interrelated power system state equation submodules;
[0009] Each power system state equation submodule is trained and optimized through physical embedding neural network;
[0010] The outputs of the physical embedded neural network of each power system state equation sub-module are integrated to obtain the global solution of the power system.
[0011] As a further optimization scheme of the construction method of the power system simulation operator based on physical embedded neural network described in the present invention, the power system state equation is:
[0012]
[0013] Among them, u(x,t) is the physical equation of the relationship between the state variables of the power system, It is a nonlinear operator used to represent the dynamic process of the power system. p(x,t) is the physical parameter of the power system, x is the state variable of the power system, and t is time.
[0014] As a further optimization scheme of the construction method of a power system simulation operator based on a physical embedded neural network according to the present invention, the output of the physical embedded neural network is:
[0015]
[0016] in, is the physical equation fitting result of the power system state variable relationship, A computational model for embedding physics into neural networks.
[0017] As a further optimization scheme of the construction method of a power system simulation operator based on a physical embedded neural network described in the present invention, the global solution of the power system is:
[0018]
[0019] Among them, x(t) is the value of all x at time t, x i (t) is the value of x at time t included in the i-th power system state equation submodule, M is the number of power system state equation submodules, W i is the weight matrix of the i-th power system state equation submodule.
[0020] As a further optimization scheme of the construction method of a power system simulation operator based on physical embedded neural network described in the present invention, the power system state variables include voltage, current, power and power angle.
[0021] As a further optimization scheme of the construction method of a power system simulation operator based on physical embedded neural network described in the present invention, by minimizing the loss function To optimize the parameters θ of the physical embedding neural network, the loss function is:
[0022]
[0023] in, is the overall error value, is the data error term, is the physical constraint error term; λ is the weight parameter of the physical constraint error;
[0024] Data error term:
[0025] Among them, N is the training sample data, is the physical equation fitting result of the power system state variable relationship of the e-th sample, u e is the actual result of the physical equation of the power system state variable relationship of the e-th sample, x e is the power system state variable of the e-th sample;
[0026] Physical constraint error term:
[0027]
[0028] in, is the physical equation fitting result of the power system state variable relationship of the jth sample, x j is the power system state variable of the jth sample, is the result of the nonlinear operator calculation of the jth sample through the dynamic process of the power system, p(x j ,t) is the calculation result of the system physical parameters of the jth sample.
[0029] As a further optimization scheme of the construction method of the power system simulation operator based on physical embedded neural network described in the present invention, the operator of each power system state equation submodule is expressed as:
[0030] L i =A i ·B i (x i )+C i (p i )
[0031] Among them, L i is the operator of the ith power system state equation submodule; A i is the coefficient matrix of the ith power system state equation submodule; B i (xi ) is the nonlinear part of the ith power system state equation submodule, which depends on the state variable x in the ith power system state equation submodule. i ; C i (p i ) is the parameter item of the ith power system state equation submodule, which depends on the parameter p of the ith power system state equation submodule i .
[0032] As a further optimization scheme of the construction method of the power system simulation operator based on the physical embedding neural network described in the present invention, the parameters of the physical embedding neural network are dynamically adjusted by an adaptive optimization method. The optimization process is as follows:
[0033]
[0034] Among them, θ k+1 is the physical embedding neural network parameter after the k+1th training, θ k is the physical embedding neural network parameter after the kth training, η is the learning rate, is the loss function relative to θ k gradient.
[0035] A computer device includes a memory, a processor, and a computer program stored in the memory and capable of running on the processor. When the processor executes the computer program, the steps of the method for constructing a power system simulation operator based on a physical embedded neural network as described above are implemented.
[0036] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method for constructing a power system simulation operator based on a physical embedded neural network as described above.
[0037] Compared with the prior art, the present invention adopts the above technical solution and has the following technical effects:
[0038] This method mathematically models the high-dimensional nonlinear equations of the power system and uses a physically embedded neural network to perform operator decomposition on the power system simulation calculations, generating multiple interconnected submodules of the power system state equation. Each submodule is trained and optimized using the physically embedded neural network, enabling efficient simulation of the complex dynamic processes of the power system. This improves solution efficiency and effectively addresses the computational bottlenecks of traditional methods in large-scale power system simulation, ensuring high accuracy and real-time performance in large-scale power system simulations. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] Figure 1The present invention provides a power system simulation calculation operator decomposition and simulation calculation process using the method proposed in the present invention;
[0040] Figure 2 The operator training effect after adopting the method proposed by the present invention;
[0041] Figure 3 It is the error statistics between the operator fusion model and the benchmark model using the method proposed in this invention. DETAILED DESCRIPTION
[0042] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.
[0043] This invention effectively decomposes operators used in power system simulation calculations, transforming complex system dynamics into multiple submodules with lower dimensions and relative independence, thereby improving the efficiency and accuracy of simulation analysis. This new approach should be able to decompose complex system dynamics into multiple subproblems through an efficient physical embedding neural network architecture. By independently solving each subproblem, computational complexity can be reduced, enabling efficient real-time dynamic simulation of large-scale power systems.
[0044] This paper proposes a method for constructing power system operators based on a physical embedding neural network (PINN). This approach aims to solve high-dimensional nonlinear dynamic systems with low computational complexity while maintaining high-precision simulation results by rationally decomposing them. This method mathematically models the high-dimensional nonlinear equations of the power system and decomposes the model into operators using a physical embedding neural network, generating multiple interconnected submodules. Each submodule is trained and optimized using the physical embedding neural network, improving solution efficiency and effectively addressing the computational bottlenecks encountered by traditional methods in large-scale power system simulations.
[0045] The core idea of this method is to decompose the dynamic equations of the power system into multiple relatively independent submodules based on computing resource constraints. While ensuring physical consistency, the operators of each submodule are learned through a neural network. This method allows independent calculations to be performed in each submodule, and ultimately, through fusion, a global solution for the entire system is obtained, thus achieving efficient simulation of the complex dynamic processes of the power system.
[0046] The technical solution of the present invention includes the following key steps:
[0047] 1. Power system dynamic modeling
[0048] By modeling the state variables and nonlinear dynamic relationships of the power system, a set of equations for a high-dimensional nonlinear system is obtained. The dynamic process of the power system can be expressed as:
[0049]
[0050] Among them, u(x,t) is the physical equation of the relationship between the state variables of the power system, It is a nonlinear operator used to represent the dynamic process of the power system, p(x,t) is the physical parameter of the power system, x is the power system state variable, and t is time;
[0051] 2. System model decomposition
[0052] The power system model decomposition needs to be carried out according to the physical structure and dynamic characteristics of the system. Generally, the power system can be divided into the following main modules:
[0053] Power generation module: includes all generators and power generation units, describing the power production process. The characteristics of traditional thermal power units and new energy units are quite different and need to be further subdivided and considered.
[0054] Transmission module: This module includes substations, transmission lines, circuit breakers, etc., and describes the power transmission process. The AC transmission and DC transmission models are quite different and require further consideration.
[0055] Load module: includes various loads, consumer demand, etc., and describes the electricity demand process.
[0056] Control and scheduling module: includes automated scheduling system, protection system, etc., and describes the system's control process.
[0057] 3. Physical Embedded Neural Network Training
[0058] The operators of each submodule are optimized using a physical embedding neural network framework. The loss function includes data error terms and physical constraint error terms:
[0059]
[0060] in, is the overall error value, is the data error term, is the physical constraint error term; λ is the weight parameter of the physical constraint error.
[0061] 4. Submodule fusion
[0062] After all submodules are solved and optimized independently, the results of each submodule are aggregated through the fusion method to obtain the global solution of the power system. The expression of the global solution is:
[0063]
[0064] Among them, x(t) is the value of all x at time t, xi (t) is the value of x at time t included in the i-th power system state equation submodule, M is the number of power system state equation submodules, W i is the weight matrix of the i-th power system state equation submodule.
[0065] 5. Efficient computing and parallelization
[0066] In terms of efficient computing and parallelization, the present invention significantly improves the solution efficiency by decomposing the high-dimensional nonlinear model of the power system into multiple relatively independent sub-modules and adopting a parallel computing strategy. Each sub-module is trained and optimized through a physical embedded neural network (PINN) and can be executed in parallel on multiple computing units (such as multi-core CPUs or GPUs), thereby accelerating the training process of the model. Through data parallelism, the training data is distributed to multiple computing devices, and a GPU-accelerated deep learning framework (such as TensorFlow or PyTorch) is used for efficient calculations. In order to handle the coupling relationship between sub-modules, a synchronization mechanism is used to ensure that the output of each sub-module meets physical constraints (such as power flow and load balance). In addition, through a distributed computing framework such as MapReduce, the computing tasks are distributed to multiple nodes to achieve collaborative processing of large-scale computing. This modular and parallel computing strategy not only improves computing efficiency, but also ensures high precision and real-time performance in large-scale power system simulations.
[0067] Figure 1 A power system operator decomposition and fusion method based on physical embedded neural network, including:
[0068] 1. System model decomposition
[0069] First, the power system is physically divided into multiple submodules. Each submodule represents an important part of the power system, such as generators, transmission lines, loads, and dispatch control modules. The dynamic equations of each submodule can be expressed as:
[0070]
[0071] Among them, u(x,t) is the physical equation of the relationship between the state variables of the power system, This is a nonlinear operator used to represent the dynamic processes of power systems. p(x, t) represents the physical parameters of the power system, x represents the power system state variable, and t represents time. By breaking down the complex power system equations into multiple submodules, each of which can be solved independently, the computational complexity is reduced.
[0072] 2. Physical Embedded Neural Network (PINN) Training and Optimization
[0073] Each submodule is trained and optimized using a physical embedding neural network (PINN). PINN combines traditional physical models with data-driven neural networks to improve the model's fit while ensuring physical consistency. During training, PINN optimizes the model by minimizing a weighted loss function that combines data error and physical error:
[0074]
[0075] in, is the overall error value, is the data error term, is the physical constraint error term; λ is the weight parameter for the physical constraint error. Each submodule is trained using parallel computing to accelerate the training process. During training, computing resources such as GPUs are used to optimize the neural network parameters.
[0076] 3. Parallelization and distributed computing
[0077] To improve computational efficiency, the present invention employs a parallel computing strategy to train and solve each submodule. Through data parallelization, the data of each submodule is distributed to multiple computing units (such as multi-core CPUs or GPUs) for parallel computing. Each computing unit independently performs forward and backward propagation calculations, and finally, a full-reduce operation is performed to aggregate the gradients and update the network parameters.
[0078] When the computational tasks of a submodule are large, a distributed computing architecture is used to divide the model into multiple parts and assign them to different computing nodes for parallel computing. For example, using a distributed framework based on MapReduce, the computational tasks of each submodule can be distributed to multiple nodes, greatly improving the computing speed.
[0079] 4. Synchronization and coupling between submodules
[0080] Since there are usually certain coupling relationships between the modules of the power system (such as power flow, load balance, etc.), synchronization between modules is required. At the end of each training cycle, the calculation results of each sub-module are merged through the synchronization mechanism to ensure the physical consistency of the entire system. For example, the power balance constraints between the sub-modules can be checked after each update:
[0081]
[0082] If the power balance is inconsistent, the module output is adjusted to ensure the consistency of the global solution. In this way, the solution of each submodule is guaranteed to be feasible in the global system.
[0083] 5. Fusion of global solutions
[0084] After training, the results of each submodule are combined to obtain a global solution. Specifically, the global state of the power system is obtained by weighted summing the output of each submodule with the corresponding weight:
[0085] Where W i is the weight matrix of each submodule.
[0086] 6. Real-time simulation and scheduling optimization
[0087] Figure 2 To show the operator training effect after using the method proposed in this invention, Figure 3 The error statistics of the operator fusion model and the benchmark model using the method proposed in the present invention. The method of the present invention is suitable for real-time simulation and scheduling optimization of the dynamic process of the power system. By adopting a physical embedded neural network, the scheduling plan of the power system can be quickly optimized under the premise of ensuring physical consistency. When faced with real-time data, the model can be trained and updated online according to the new input information, thereby providing more accurate simulation analysis and scheduling decisions. The present method is applied to the scheduling, optimization and real-time monitoring of the power system, and performs real-time dynamic analysis and prediction of the power system through physical embedded neural networks. The present method decomposes the multiple sub-modules of the power system into operators, and trains the operators of each sub-module separately through neural networks, thereby achieving efficient solution to the overall dynamic process of the system.
[0088] A computer device includes a memory, a processor, and a computer program stored in the memory and capable of running on the processor. When the processor executes the computer program, the steps of the method for constructing a power system simulation operator based on a physical embedded neural network as described above are implemented.
[0089] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method for constructing a power system simulation operator based on a physical embedded neural network as described above.
[0090] It will be understood by those skilled in the art that the embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention may be implemented in various computer languages, for example, the object-oriented programming language Java and the interpreted scripting language JavaScript.
[0091] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0092] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.
[0093] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.
[0094] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.
[0095] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.
Claims
1. A method for constructing a power system simulation operator based on physical embedded neural network, characterized in that: include: Model the high-dimensional nonlinear dynamic equations of the power system and obtain the power system state equation: Decomposing the power system state equation based on computing resource conditions to form a plurality of interrelated power system state equation submodules; Each power system state equation submodule is trained and optimized through physical embedding neural network; The outputs of the physical embedded neural network of each power system state equation sub-module are integrated to obtain the global solution of the power system.
2. The method for constructing a power system simulation operator based on physical embedded neural network according to claim 1, characterized in that: The power system state equation is: Among them, u(x,t) is the physical equation of the relationship between the state variables of the power system, It is a nonlinear operator used to represent the dynamic process of the power system. p(x,t) is the physical parameter of the power system, x is the state variable of the power system, and t is time.
3. The method for constructing a power system simulation operator based on physical embedded neural network according to claim 1, characterized in that: The output of the physical embedding neural network is: in, is the physical equation fitting result of the power system state variable relationship, A computational model for embedding physics into neural networks.
4. The method for constructing a power system simulation operator based on physical embedded neural network according to claim 1, characterized in that: The global solution of the power system is: Among them, x(t) is the value of all x at time t, x i (t) is the value of x at time t included in the i-th power system state equation submodule, M is the number of power system state equation submodules, W i is the weight matrix of the i-th power system state equation submodule.
5. The method for constructing a power system simulation operator based on physical embedded neural network according to claim 2, characterized in that: The power system state variables include voltage, current, power and power angle.
6. The method for constructing a power system simulation operator based on physical embedded neural network according to claim 1, characterized in that: By minimizing the loss function To optimize the parameters θ of the physical embedding neural network, the loss function is: in, is the overall error value, is the data error term, is the physical constraint error term; λ is the weight parameter of the physical constraint error; Data error term: Among them, N is the training sample data, is the physical equation fitting result of the power system state variable relationship of the e-th sample, u e is the actual result of the physical equation of the power system state variable relationship of the e-th sample, x e is the power system state variable of the e-th sample; Physical constraint error term: in, is the physical equation fitting result of the power system state variable relationship of the jth sample, x j is the power system state variable of the jth sample, is the result of the nonlinear operator calculation of the jth sample through the dynamic process of the power system, p(x j ,t) is the calculation result of the system physical parameters of the jth sample.
7. The method for constructing a power system simulation operator based on physical embedded neural network according to claim 1, characterized in that: The operator of each power system state equation submodule is expressed as: L i =A i ·B i (x i )+C i (p i ) Among them, L i is the operator of the ith power system state equation submodule; A i is the coefficient matrix of the ith power system state equation submodule; B i (x i ) is the nonlinear part of the ith power system state equation submodule, which depends on the state variable x in the ith power system state equation submodule. i ; C i (p i ) is the parameter item of the ith power system state equation submodule, which depends on the parameter p of the ith power system state equation submodule i .
8. The method for constructing a power system simulation operator based on physical embedded neural network according to claim 1, characterized in that: The parameters of the physical embedding neural network are dynamically adjusted through adaptive optimization methods. The optimization process is as follows: Among them, θ k+1 is the physical embedding neural network parameter after the k+1th training, θ k is the physical embedding neural network parameter after the kth training, η is the learning rate, is the loss function relative to θ k gradient.
9. A computer device comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that: When the processor executes the computer program, the steps of the method for constructing a power system simulation operator based on a physical embedded neural network as described in any one of claims 1 to 8 are implemented.
10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method for constructing a power system simulation operator based on a physical embedded neural network as described in any one of claims 1 to 8 are implemented.