New energy power station configuration optimization method based on proxy model and multi-objective optimization
By using a proxy model and multi-objective optimization method, and leveraging multilayer perceptron neural networks and Bayesian optimization, the stability characteristic values of new energy power plants can be quickly predicted. This solves the problems of large computational load and insufficient accuracy in the configuration optimization of new energy power plants, and achieves efficient global optimal solution.
Patent Information
- Application Number
- CN202511321557.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-16
- Publication Date
- 2025-12-16
AI Technical Summary
Existing technologies for optimizing the configuration of new energy power plants suffer from problems such as large computational load, long processing time, and a tendency to get trapped in local optima, making it difficult to balance efficiency and accuracy.
We employ a surrogate model-based, multi-objective optimization approach, utilizing a multilayer perceptron (MLP) neural network for offline training. By combining physical prior constraints and Bayesian optimization, we can quickly predict stability feature values and solve for the Pareto optimal solution set through multi-objective optimization.
It significantly reduces computation time, improves the optimization efficiency and accuracy of new energy power plant configuration, and can approach the global optimal solution with a limited computational budget, providing a scientific configuration scheme.
Smart Images

Figure CN121145651A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of new energy power plant capacity configuration technology, and in particular to a new energy power plant configuration optimization method based on surrogate model and multi-objective optimization. Background Technology
[0002] As the "dual carbon" goal is advanced and the penetration rate of new energy sources continues to increase, their volatility and nonlinear characteristics exacerbate the risk to system stability. Optimizing power plant configuration has become the key to ensuring system stability.
[0003] Current mainstream technologies have two main limitations: First, optimization methods based on detailed numerical simulation rely on electromagnetic / electromechanical transient simulation tools to accurately characterize the dynamic characteristics of the system. However, each simulation is time-consuming, and the optimization cycle can reach several days when there are many candidate solutions. Although traditional algorithms avoid full space traversal, the computational load is still large due to the need for simulation during iteration, and they are prone to getting trapped in local optima, resulting in insufficient global optimization quality. Second, methods based on simplified physical models improve efficiency by linearizing the state-space equations, reducing the computation cycle to hours or even minutes. However, because they ignore the nonlinearity of converters and the fluctuation of new energy output, they cannot accurately reflect the dynamics of the actual system. Optimization solutions often appear to be theoretically feasible but fail in practice, resulting in significant deviations from actual operating conditions.
[0004] Therefore, there is an urgent need in related technologies for a method that can balance the efficiency and accuracy of power plant configuration optimization. Summary of the Invention
[0005] Therefore, it is necessary to provide a new energy power plant configuration optimization method based on surrogate models and multi-objective optimization that can balance the efficiency and accuracy of power plant configuration optimization, in order to address the above-mentioned technical problems.
[0006] Firstly, this application provides a new energy power plant configuration optimization method based on a surrogate model and multi-objective optimization. The method includes: Obtain historical power system configuration parameters and corresponding historical stability characteristic values, and perform preprocessing; The preprocessed historical power system configuration parameters are input into a neural network proxy model based on a multilayer perceptron architecture. The model is then trained offline by combining the corresponding historical stability feature values and physical prior constraints, and the predicted stability feature values are output. Using the predicted stability feature value as the optimization objective, a multi-objective Bayesian optimization objective function is constructed. Based on the objective function, the Pareto optimal solution set is obtained through iterative solution to determine the optimal configuration scheme for new energy power plants.
[0007] Optionally, in one embodiment of this application, obtaining historical power system configuration parameters and corresponding stability characteristic values, and performing preprocessing, includes: Frequency response data under different new energy power plant configurations are generated based on power system simulation. Four stability feature values are extracted, and outlier removal and standardization operations are performed.
[0008] Optionally, in one embodiment of this application, the standardization operation includes: The power system configuration parameters are scaled within a range, and the stability eigenvalues are standardized using Z-score.
[0009] Optionally, in one embodiment of this application, the loss function for offline training is:
[0010] in, For stability eigenvalue weights, To predict stability eigenvalues, This represents the historical stability eigenvalue. As a priori physical constraint, For physical prior weights, For regularization weights, Refers to the first The Frobenius norm of the layer weight matrix.
[0011] Optionally, in one embodiment of this application, the physical prior constraints are determined based on physical laws of stability characteristics.
[0012] Optionally, in one embodiment of this application, the iterative solution to the Pareto optimal solution set based on the objective function includes: We introduce the desired hypervolume lift as a multi-objective acquisition function, and maximize the desired hypervolume lift to gradually approach the Pareto front.
[0013] Optionally, in one embodiment of this application, the offline training further includes: The model parameters are constrained by L2 weight decay, and an early stopping strategy is adopted to avoid overfitting when the performance on the validation set no longer improves.
[0014] Secondly, this application also provides a new energy power plant configuration optimization device based on a proxy model and multi-objective optimization. The device includes: The data acquisition and preprocessing module is used to acquire historical power system configuration parameters and corresponding historical stability characteristic values, and to perform preprocessing. The neural network surrogate model training module is used to input the preprocessed historical power system configuration parameters into the neural network surrogate model based on the multilayer perceptron architecture, combine the corresponding historical stability feature values and physical prior constraints for offline training, and output the predicted stability feature values. A multi-objective Bayesian optimization module is used to construct an objective function for multi-objective Bayesian optimization with the predicted stability feature value as the optimization objective. The new energy power plant configuration scheme determination module is used to iteratively solve the Pareto optimal solution set based on the objective function to determine the optimal new energy power plant configuration scheme.
[0015] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the steps of the methods described in the various embodiments above.
[0016] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the steps of the methods described in the various embodiments above.
[0017] The aforementioned method for optimizing the configuration of new energy power plants based on a surrogate model and multi-objective optimization first obtains historical power system configuration parameters and corresponding historical stability feature values, and preprocesses them. Then, the preprocessed historical power system configuration parameters are input into a neural network surrogate model based on a multilayer perceptron architecture. This model is then trained offline, combining the corresponding historical stability feature values and physical prior constraints, and outputs predicted stability feature values. Next, using these predicted stability feature values as the optimization objective, a multi-objective Bayesian optimization objective function is constructed. Finally, based on the objective function, the Pareto optimal solution set is iteratively solved to determine the optimal new energy power plant configuration scheme. In other words, by introducing a neural network surrogate model and using historical simulation data for offline training, the stability feature value prediction for power plant configuration can be quickly completed after model convergence, significantly reducing the time required for a single evaluation and overcoming the technical bottleneck of traditional simulation methods, which suffer from excessively long single-evaluation times and difficulty in supporting large-scale global searches. The surrogate model adopts a multilayer perceptron (MLP) structure, whose fully connected characteristics can efficiently capture the complex nonlinear mapping relationship between input parameters (photovoltaic capacity, wind power capacity, and energy storage capacity) and output stability feature values. Compared to convolutional neural networks or recurrent neural networks, MLP is better suited for processing tabular structured data, significantly reducing network complexity and training difficulty while maintaining modeling accuracy. A Bayesian optimization method is introduced in the optimization stage, establishing a probabilistic representation of stability feature values based on Gaussian processes, and using a sampling function to dynamically balance global exploration and local mining. This mechanism can gradually converge to the global optimum under limited computational budget, effectively avoiding the tendency of traditional heuristic algorithms to get trapped in local optima, while significantly improving information utilization efficiency through uncertainty modeling. By combining multi-objective optimization with Bayesian optimization, a multi-objective sampling function is used to approximate the Pareto front between stability feature values, thus obtaining a set of optimal solutions with different trade-offs. This approach balances accuracy and multi-objective trade-offs, providing a more flexible and scientific configuration scheme for the construction of new energy power plants. Attached Figure Description
[0018] Figure 1 This is a flowchart illustrating a new energy power plant configuration optimization method based on a proxy model and multi-objective optimization in one embodiment. Figure 2 This is a schematic diagram of the power grid frequency fluctuation curve in one embodiment; Figure 3 This is a structural block diagram of a new energy power plant configuration optimization device based on a proxy model and multi-objective optimization in one embodiment; Figure 4 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation
[0019] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.
[0020] In one embodiment, such as Figure 1 As shown, a new energy power plant configuration optimization method based on a proxy model and multi-objective optimization is provided, including the following steps: S101: Obtain historical power system configuration parameters and corresponding historical stability characteristic values, and perform preprocessing.
[0021] In this embodiment, firstly, frequency response data under different new energy configurations are generated based on power system simulation, and four stability feature values are extracted, preprocessed, and a training dataset is constructed. The dataset is divided into a training set, a validation set, and a test set in a ratio of approximately 8:1:1 to ensure that the model has independent samples in the training, tuning, and performance evaluation stages.
[0022] Specifically, in one embodiment of this application, the step of obtaining historical power system configuration parameters and corresponding stability characteristic values, and performing preprocessing, includes: Frequency response data under different new energy power plant configurations are generated based on power system simulation. Four stability feature values are extracted, and outlier removal and standardization operations are performed.
[0023] In one embodiment of this application, a conventional simulation method is used to generate, such as Figure 2 The power grid frequency fluctuation curve shown is used to extract the maximum frequency deviation. Adjusting time Frequency deviation exponential change rate α, energy deficit Four-dimensional features are used to construct a dataset that strictly follows the dynamic laws of the power system. Among these, the maximum frequency deviation refers to the lowest frequency (Nadir) after the disturbance relative to the rated value. The maximum decline is calculated using the following formula:
[0024] in, As the starting point of the disturbance, See the boundary between the end of Rebound and the beginning of Recovery in the diagram.
[0025] The settling time refers to the frequency entering and maintaining a steady-state bandwidth. The time required within ) is determined by the following criteria: The first time entering the interval ,
[0026] Set the tolerance bandwidth to a fixed value. ,
[0027] The exponential rate of change of frequency deviation refers to the exponential rate of change of frequency deviation after a disturbance occurs, before reaching Nadir. It is obtained by least squares fitting, specifically in... Take 5-10 points evenly on the top:
[0028] Energy deficit refers to the amount of kinetic energy lost by a system due to a disturbance. Specifically, at the rated frequency... The starting point of the disturbance is When the frequency is lower than the steady-state bandwidth The accumulated area over time is defined as the energy deficit:
[0029] in For first entry The time within ( Set to 0.1Hz).
[0030] To ensure data validity, the raw data needs to be preprocessed, including outlier removal and standardization. Specifically, although the dataset is generated by the simulation platform, and theoretically each configuration should yield a corresponding frequency response curve, numerical anomalies may still occur under extreme configurations or insufficient simulation convergence. If these outliers are not handled, they will affect the training stability and prediction accuracy of the surrogate model. The distribution of each feature value in the training set is statistically analyzed; if a value deviates from the mean by more than 3σ, it is considered an anomaly and the data set should be removed.
[0031] In the optimization problem of new energy power plant configuration, the input parameters and output stability indices have different physical dimensions. For example, the capacity of photovoltaic, wind power, and energy storage is expressed in MW or MWh, while the four output stability characteristics are expressed in Hz, second, dimensionless decay rate, and Hz·s, respectively. If the raw values are used directly for neural network training, the difference in dimensions between different dimensions will be too large, resulting in gradient imbalance. Therefore, the data must be standardized before training.
[0032] Specifically, in one embodiment of this application, the standardized operation includes: The power system configuration parameters are scaled within a range, and the stability eigenvalues are standardized using Z-score.
[0033] In one embodiment of this application, the input variables include photovoltaic capacity, wind power capacity, and energy storage capacity, which typically have significantly different value ranges. To avoid the large range of features having an excessive impact on the training process, the input parameters are range-scaled, linearly mapping the data to the [0,1] interval:
[0034] The output variables are maximum frequency deviation, settling time, exponential rate of change of frequency deviation, and energy deficit. Due to their greater unit differences, Z-score standardization is uniformly adopted.
[0035] This ensures that the four targets are at similar numerical scales during training, preventing one target from dominating gradient updates due to excessively large values.
[0036] S102: Input the preprocessed historical power system configuration parameters into the neural network proxy model based on the multilayer perceptron architecture, combine the corresponding historical stability feature values and physical prior constraints for offline training, and output the predicted stability feature values.
[0037] In this embodiment, the neural network proxy model uses a multilayer perceptron (MLP) as its architecture in the new energy power plant optimization scenario. This choice is based on the structured characteristics of the input data: the optimization variables are tabular data consisting of photovoltaic capacity, wind power capacity, and energy storage capacity, eliminating the need to process spatial or temporal correlation features. The fully connected nature of the MLP can efficiently capture the global nonlinear mapping between parameters and stability feature values, making it more efficient than convolutional neural networks (which require image or grid data) or recurrent neural networks (which require time series data).
[0038] A multilayer perceptron (MLP) is a feedforward fully connected neural network. Its fundamental principle lies in constructing a nonlinear mapping relationship between the input and output through layers of linear transformations and nonlinear activations. Let the input vector be... , No. The output of the layer is The calculation process can be expressed as follows:
[0039] in, and These are the weight matrix and bias vector for this layer, respectively. This is a non-linear activation function (ReLU). Through multiple nested layers, MLPs can approximate arbitrarily complex continuous functions, thereby achieving a high-dimensional non-linear mapping from input variables to output feature values. The final output layer is:
[0040] In this technique, the input vector Corresponding photovoltaic capacity Wind power capacity Energy storage capacity Output vector The key stability characteristics of the corresponding system include the number of corresponding output stability characteristics: maximum frequency deviation. Adjusting time Frequency deviation exponential change rate α, energy deficit .
[0041] The specific architecture of the multilayer perceptron (MLP) is shown in Table 1 below.
[0042] Table 1
[0043]
[0044]
[0045]
[0046] During model training, batch gradient descent is used for iterative optimization. Each training epoch includes forward propagation, loss function calculation, backpropagation, and parameter update. Mean squared error (MSE) is chosen as the loss function, and engineered weights can be introduced to highlight the importance of different stability features. The optimizer uses Adam, which improves convergence stability while ensuring computational efficiency. Its update rule is based on an adaptive adjustment mechanism of first- and second-order momentum, automatically adjusting the learning rate across different parameter dimensions, thus achieving stable and fast convergence without the need for complex scheduling strategies. The learning rate is set to... As a commonly used empirical value, this ensures an effective learning speed in the early stages of training while avoiding oscillations caused by excessively large update steps. The training batch size is typically set between 32 and 128 to balance computational efficiency and the stability of gradient estimation.
[0047] After training, the model performance is evaluated using an independent test set, and the mean squared error (MSE), mean absolute error (MAE), and coefficient of determination are calculated. Feature values such as (e.g., learning rate, regularization strength) are used to verify the model's prediction accuracy and generalization ability. If the evaluation results do not meet expectations, the model can be optimized by adjusting the network structure, learning rate, or regularization strength. The final optimal model is used as a surrogate model during the deployment phase to quickly complete the stability assessment of the new energy power plant configuration.
[0048] Multilayer perceptrons (MLPs) serve as surrogate models to replace time-consuming dynamic simulations of power systems. By training offline on historical simulation data, the MLP establishes an approximate functional mapping from configuration parameters to stability eigenvalues. During optimization, the surrogate model can quickly complete a single stability assessment, significantly reducing computational costs. This not only ensures the accuracy of input-output relationship modeling but also significantly improves the computational efficiency of the optimization process, enabling the optimization of new energy power plant configurations to balance accuracy and efficiency.
[0049] In one embodiment of this application, the loss function for offline training is:
[0050] in, For stability eigenvalue weights, To predict stability eigenvalues, This represents the historical stability eigenvalue. As a priori physical constraint, For physical prior weights, For regularization weights, Refers to the first The Frobenius norm of the layer weight matrix.
[0051] In one embodiment of this application, the physical prior constraints are determined based on physical laws with stability characteristics.
[0052] In one embodiment of this application, in traditional neural network agent modeling, the network relies entirely on data to learn mapping relationships. If the training data is limited or noisy, the model prediction may produce physically unreasonable results, such as: the predicted maximum frequency deviation being negative (which does not conform to the definition, the deviation should be a non-negative amount); the predicted oscillation decay rate α being positive (corresponding to system instability, but in this scenario, the power system should show a decay trend after disturbance); the predicted settling time being negative or too small (which does not conform to the physical meaning of time); and the predicted energy deficit being negative (which has no physical meaning).
[0053] To ensure that the prediction results conform to the physical laws of the power system, physical prior constraints are introduced into the surrogate model. Known physical laws from power system stability analysis are incorporated into the network training objective, ensuring that the model predictions satisfy both numerical accuracy and physical plausibility. A regularization term is added to the loss function. Specifically: For the exponential rate of change of frequency deviation α, if α ≤ 0, the regularization term is:
[0054] For the maximum frequency deviation ,Require ≥0, the regularization term is:
[0055] Adjustment time ,Require ≥0, the regularization term is:
[0056] energy deficit ,Require ≥0, the regularization term is:
[0057] After synthesis, the physical constraint regularization can be written as:
[0058] In one embodiment of this application, the offline training further includes: The model parameters are constrained by L2 weight decay, and an early stopping strategy is adopted to avoid overfitting when the performance on the validation set no longer improves.
[0059] In one embodiment of this application, L2 weight decay is used to constrain the model parameters by adding a penalty of the squared weight term to the loss function. This limits the model parameters to a large size, prevents the model from overfitting on the training data, and improves generalization ability.
[0060] S103: Using the predicted stability feature value as the optimization objective, construct a multi-objective Bayesian optimization objective function.
[0061] In this embodiment, a trained surrogate model is combined with a multi-objective Bayesian optimization algorithm to dynamically balance global exploration and local mining, continuously selecting new candidate configurations and updating the model. Bayesian optimization (BO) is a class of methods specifically designed to solve global optimization problems of high-cost black-box functions. Its basic idea is to maximize information utilization efficiency by combining probabilistic modeling and sequential decision-making. In this technique, multiple stability features need to be optimized simultaneously. The four stability features predicted by the neural network surrogate model are used as optimization objectives, denoted as:
[0062] Where x represents the power station configuration parameters: photovoltaic capacity, wind power capacity, and energy storage capacity.
[0063] In multi-objective Bayesian optimization, each objective function It can be modeled as a Gaussian process:
[0064] Through a limited sample dataset Obtain the posterior distribution for each target:
[0065] Since there is no single global optimum in multi-objective optimization, the goal of Bayesian optimization is to gradually approach the Pareto front.
[0066] S104: Based on the objective function, iteratively solve for the Pareto optimal solution set to determine the optimal configuration scheme for new energy power plants.
[0067] In this embodiment, the non-dominated solution set is obtained through iterative convergence, forming a Pareto front among various stability eigenvalues, thereby quickly obtaining a set of optimal configuration schemes with different trade-offs.
[0068] Specifically, in one embodiment of this application, the iterative solution to the Pareto optimal solution set based on the objective function includes: We introduce the desired hypervolume lift as a multi-objective acquisition function, and maximize the desired hypervolume lift to gradually approach the Pareto front.
[0069] In one embodiment of this application, Expected Hypervolume Improvement (EHVI) is introduced as a multi-target acquisition function, which is defined as follows:
[0070] in, Indicates supervolume. This represents the currently obtained non-dominated solution set. Maximizing EHVI expands the non-dominated solution set while exploring unknown regions, gradually converging the search process to the Pareto optimal solution set.
[0071] First, design variables. Target vector (Maximum frequency deviation, settling time, exponential rate of change of frequency deviation, and energy deficit are all "the smaller the better"). The distribution of each objective function value can be obtained by predicting it through a surrogate model. Worse point Each component is slightly worse than the current worst nondominated solution.
[0072] In iteration number At turn 1, there is already a non-dominated solution set. The corresponding reference point is The current definition of hypervolume is:
[0073] Hypervolume is composed of non-dominated solution sets Dominate, and do not exceed the reference point The volume of all regions, when considering a new candidate point x, its predicted result is: If it is added to the existing solution set, the hypervolume is increased to
[0074] Since y is a random variable, its expected increase is the same as its expected hypervolume increase:
[0075] As the iteration continues, new candidate points are constantly added, and the non-dominated solution set... The coverage area gradually expands, and the hypervolume monotonically increases. When the Pareto front is sufficiently approximated, all candidate points... If all values approach zero, it means that the search process naturally converges to the Pareto optimal solution set.
[0076] In the scenario of optimizing the configuration of new energy power plants, the neural network surrogate model can quickly output four stability feature values under different configurations. Bayesian optimization then uses these predictions and their uncertainties to construct a multi-objective probabilistic model and actively selects the most valuable new configuration point through the EHVI multi-objective acquisition function. This not only approximates the complete Pareto front with a limited computational budget, but also reveals the trade-off between maximum frequency deviation, settling time, oscillation decay rate, and energy deficit.
[0077] In the aforementioned optimization method for new energy power plant configuration based on a surrogate model and multi-objective optimization, firstly, historical power system configuration parameters and corresponding historical stability feature values are obtained and preprocessed. Then, the preprocessed historical power system configuration parameters are input into a neural network surrogate model based on a multilayer perceptron architecture. Offline training is performed using the corresponding historical stability feature values and physical prior constraints to output predicted stability feature values. Next, using the predicted stability feature values as the optimization objective, a multi-objective Bayesian optimization objective function is constructed. Finally, based on the objective function, the Pareto optimal solution set is iteratively solved to determine the optimal new energy power plant configuration scheme. In other words, by introducing a neural network surrogate model and using historical simulation data for offline training, the stability feature value prediction for power plant configuration can be quickly completed after model convergence, significantly reducing the time required for a single evaluation and overcoming the technical bottleneck of traditional simulation methods being too time-consuming and unable to support large-scale global searches. The surrogate model adopts a multilayer perceptron (MLP) structure, whose fully connected characteristics can efficiently capture the complex nonlinear mapping relationship between input parameters (photovoltaic capacity, wind power capacity, and energy storage capacity) and output stability feature values. Compared to convolutional neural networks or recurrent neural networks, MLP is better suited for processing tabular structured data, significantly reducing network complexity and training difficulty while maintaining modeling accuracy. A Bayesian optimization method is introduced in the optimization stage, establishing a probabilistic representation of stability feature values based on Gaussian processes, and using a sampling function to dynamically balance global exploration and local mining. This mechanism can gradually converge to the global optimum under limited computational budget, effectively avoiding the tendency of traditional heuristic algorithms to get trapped in local optima, while significantly improving information utilization efficiency through uncertainty modeling. By combining multi-objective optimization with Bayesian optimization, a multi-objective sampling function is used to approximate the Pareto front between stability feature values, thus obtaining a set of optimal solutions with different trade-offs. This approach balances accuracy and multi-objective trade-offs, providing a more flexible and scientific configuration scheme for the construction of new energy power plants.
[0078] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.
[0079] Based on the same inventive concept, this application also provides a new energy power plant configuration optimization device for implementing the aforementioned new energy power plant configuration optimization method based on surrogate model and multi-objective optimization. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the new energy power plant configuration optimization device based on surrogate model and multi-objective optimization provided below can be found in the limitations of the new energy power plant configuration optimization method based on surrogate model and multi-objective optimization described above, and will not be repeated here.
[0080] In one embodiment, such as Figure 3 As shown, a new energy power plant configuration optimization device 300 based on surrogate model and multi-objective optimization is provided, including: a data acquisition and preprocessing module 301, a neural network surrogate model training module 303, a multi-objective Bayesian optimization module 305, and a new energy power plant configuration scheme determination module 307, wherein: The data acquisition and preprocessing module 301 is used to acquire historical power system configuration parameters and corresponding historical stability characteristic values, and to perform preprocessing.
[0081] The neural network surrogate model training module 303 is used to input the preprocessed historical power system configuration parameters into the neural network surrogate model based on the multilayer perceptron architecture, combine the corresponding historical stability feature values and physical prior constraints for offline training, and output the predicted stability feature values.
[0082] The multi-objective Bayesian optimization module 305 is used to construct the objective function of multi-objective Bayesian optimization with the predicted stability feature value as the optimization objective.
[0083] The new energy power plant configuration scheme determination module 307 is used to iteratively solve the Pareto optimal solution set based on the objective function to determine the optimal new energy power plant configuration scheme.
[0084] In one embodiment of this application, the data acquisition and preprocessing module is further configured to: Frequency response data under different new energy power plant configurations are generated based on power system simulation. Four stability feature values are extracted, and outlier removal and standardization operations are performed.
[0085] In one embodiment of this application, the data acquisition and preprocessing module is further configured to: The power system configuration parameters are scaled within a range, and the stability eigenvalues are standardized using Z-score.
[0086] In one embodiment of this application, the loss function for offline training is:
[0087] in, For stability eigenvalue weights, To predict stability eigenvalues, This represents the historical stability eigenvalue. As a priori physical constraint, For physical prior weights, For regularization weights, Refers to the first The Frobenius norm of the layer weight matrix.
[0088] In one embodiment of this application, the physical prior constraints are determined based on physical laws with stability characteristics.
[0089] In one embodiment of this application, the new energy power station configuration scheme determination is further used for: We introduce the desired hypervolume lift as a multi-objective acquisition function, and maximize the desired hypervolume lift to gradually approach the Pareto front.
[0090] In one embodiment of this application, the neural network proxy model training module is further configured to: The model parameters are constrained by L2 weight decay, and an early stopping strategy is adopted to avoid overfitting when the performance on the validation set no longer improves.
[0091] The modules in the aforementioned new energy power plant configuration optimization device based on the surrogate model and multi-objective optimization can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.
[0092] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 4As shown, the computer device includes a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When executed by the processor, the computer program implements a new energy power plant configuration optimization method based on a proxy model and multi-objective optimization. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.
[0093] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.
[0094] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.
[0095] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.
[0096] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.
[0097] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.
[0098] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.
[0099] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0100] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.
Claims
1. A new energy power plant configuration optimization method based on surrogate model and multi-objective optimization, characterized in that, The method includes: Obtain historical power system configuration parameters and corresponding historical stability characteristic values, and perform preprocessing; The preprocessed historical power system configuration parameters are input into a neural network proxy model based on a multilayer perceptron architecture. The model is then trained offline by combining the corresponding historical stability feature values and physical prior constraints, and the predicted stability feature values are output. Using the predicted stability feature value as the optimization objective, a multi-objective Bayesian optimization objective function is constructed. Based on the objective function, the Pareto optimal solution set is obtained through iterative solution to determine the optimal configuration scheme for new energy power plants.
2. The new energy power plant configuration optimization method based on surrogate model and multi-objective optimization according to claim 1, characterized in that, The process of obtaining historical power system configuration parameters and corresponding stability characteristic values, and performing preprocessing, includes: Frequency response data under different new energy power plant configurations are generated based on power system simulation. Four stability feature values are extracted, and outlier removal and standardization operations are performed.
3. The new energy power plant configuration optimization method based on surrogate model and multi-objective optimization according to claim 2, characterized in that, The standardized operations include: The power system configuration parameters are scaled within a range, and the stability eigenvalues are standardized using Z-score.
4. The new energy power plant configuration optimization method based on surrogate model and multi-objective optimization according to claim 1, characterized in that, The loss function for offline training is: in, For stability eigenvalue weights, To predict stability eigenvalues, For historical stability eigenvalues, As a priori physical constraint, For physical prior weights, For regularization weights, Refers to the first The Frobenius norm of the layer weight matrix.
5. The new energy power plant configuration optimization method based on surrogate model and multi-objective optimization according to claim 4, characterized in that, The physical prior constraints are determined based on physical laws with stability characteristics.
6. The new energy power plant configuration optimization method based on surrogate model and multi-objective optimization according to claim 1, characterized in that, The iterative solution to the Pareto optimal solution set based on the objective function includes: We introduce the desired hypervolume lift as a multi-objective acquisition function, and maximize the desired hypervolume lift to gradually approach the Pareto front.
7. The new energy power plant configuration optimization method based on surrogate model and multi-objective optimization according to claim 1, characterized in that, The offline training also includes: The model parameters are constrained by L2 weight decay, and an early stopping strategy is adopted to avoid overfitting when the performance on the validation set no longer improves.
8. A new energy power plant configuration optimization device based on surrogate model and multi-objective optimization, characterized in that, The device includes: The data acquisition and preprocessing module is used to acquire historical power system configuration parameters and corresponding historical stability characteristic values, and to perform preprocessing. The neural network surrogate model training module is used to input the preprocessed historical power system configuration parameters into the neural network surrogate model based on the multilayer perceptron architecture, combine the corresponding historical stability feature values and physical prior constraints for offline training, and output the predicted stability feature values. A multi-objective Bayesian optimization module is used to construct an objective function for multi-objective Bayesian optimization with the predicted stability feature value as the optimization objective. The new energy power plant configuration scheme determination module is used to iteratively solve the Pareto optimal solution set based on the objective function to determine the optimal new energy power plant configuration scheme.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.
Citation Information
Cited By
Casting shrinkage cavity process parameter optimization method and system based on deep learning
CN121413453A
