Weighted prediction-decision integration method, system, device and medium considering uncertainty impact

Abstract

This invention discloses a weighted prediction-decision integration method, system, device, and medium that considers the impact of uncertainty. Belonging to the interdisciplinary field of power system optimization and artificial intelligence, it includes generating a weighted dataset, training prediction models under different weight parameters and accumulating data; training the prediction model for each weight to obtain the prediction task and calculating the decision loss; constructing and training a fitted surrogate model, and applying the surrogate model to optimize the weight parameters to obtain the optimal value. This invention breaks the traditional paradigm of separating prediction and optimization decision-making. Through a problem-driven weighted mechanism, it directly links the training objective of the prediction model with the quality of downstream decision-making. It achieves the alignment of prediction serving decision-making objectives, thereby automatically identifying and accurately predicting the key uncertainty factors that have the greatest impact on decision-making outcomes in complex power system operation scenarios. Under the same prediction resources, it significantly improves the safety and economy of system operation.

Description

Technical Field

[0001] This invention relates to the field of power system optimization operation and artificial intelligence, specifically to a weighted prediction-decision integrated method, system, device and medium that takes into account the impact of uncertainty. Background Technology

[0002] The uncertainties brought about by the rapid integration of renewable energy and new loads have become a major challenge to the safe operation of the power system.

[0003] Uncertainty management typically involves two key processes: uncertainty quantification and decision-making under uncertainty.

[0004] Forecasting is an effective and widely adopted method for quantifying future uncertainties based on observable characteristics. These forecasts further serve as a key reference for informed decision-making under conditions of uncertainty, forming a prevalent "predict first, optimize later" paradigm that has been widely applied. The accuracy of forecasts has a significant impact on optimization results. Extensive research has made significant contributions to the development and improvement of forecasting methods (e.g., model-based and data-based methods) and optimization methods (e.g., chance-constrained optimization and partial bar optimization) to improve forecast accuracy and optimization quality. Traditional "predict first, optimize later" methods typically treat forecasting and optimization as two independent and separate steps. However, in actual power system operation, forecasting and optimization are intrinsically linked, especially in the presence of multiple uncertainties. Different optimization tasks emphasize different characteristics of the forecasting objectives, and the impact of forecasting errors on optimization results exhibits nonlinear, unbalanced, and problem-specific behavior. For example, the same forecasting error in load forecasting can have different effects on voltage control and economic dispatch problems. In other words, existing "predict first, optimize later" methods often ignore the specific characteristics of downstream decision problems, resulting in a serious drawback: they cannot generate forecasts tailored to specific decision characteristics, leading to poor decision performance. Summary of the Invention

[0005] In view of the above-mentioned problems, the present invention is proposed.

[0006] Therefore, the technical problem that this invention aims to solve is that traditional prediction models take average statistical error as the optimization objective, which is out of touch with the actual needs of downstream decision-making tasks. They cannot automatically identify and prioritize the prediction accuracy of key uncertainty factors that have a significant impact on decision results, thus leading to suboptimal decision-making.

[0007] To address the aforementioned technical problems, this invention provides the following technical solution: a weighted prediction-decision integration method considering the impact of uncertainty, comprising, Generate a weighted dataset, train a prediction model under different weight parameters and accumulate data; train a prediction model for each weight and obtain the prediction task, calculate the decision loss; construct and train a fitted surrogate model, and apply the surrogate model to optimize the weight parameters to obtain the optimal value.

[0008] As a preferred embodiment of the weighted prediction-decision integration method considering the impact of uncertainty described in this invention, the prediction model includes: constructing a parameterized weighted prediction model for a prediction task containing two or more uncertain variables, wherein the corresponding loss function contains adjustable weight parameters that correspond one-to-one with the variables.

[0009] As a preferred embodiment of the weighted prediction-decision integration method considering the impact of uncertainty described in this invention, the prediction model further includes defining a problem-oriented prediction loss function based on the optimization objective and decision variables of the decision task, and quantifying the degree of suboptimality of the decision scheme derived from the prediction results of the weighted prediction model relative to the optimal decision scheme derived from the true value of uncertainty.

[0010] As a preferred embodiment of the weighted prediction-decision integration method considering the impact of uncertainty described in this invention, the prediction model further includes introducing variable-specific weights into the loss function, and reformulating the traditional loss function into a weighted loss function. The weighted prediction model is trained under the weighted loss function to obtain the optimal parameters of the prediction model based on the weighted loss function.

[0011] As a preferred embodiment of the weighted prediction-decision integration method considering the impact of uncertainty described in this invention, wherein: the calculation of decision loss includes, based on a given prediction result ,Depend on Derived optimal decision Will be realized After being revealed, it is applied to ; for quantification Compared to Derived optimal decision Due to the suboptimal nature of the decision, a decision loss is introduced: in, The objective function is... When the prediction is perfect... hour, ; Based on the prediction model The prediction loss for a problem-oriented dataset is defined as follows: The expected decision loss under all uncertainties is: in, As the expected value, the current prediction loss directly correlates prediction performance with optimization quality.

[0012] As a preferred embodiment of the weighted prediction-decision integration method considering the impact of uncertainty described in this invention, the step of optimizing the weight parameters using a proxy model to obtain the optimal value includes, to achieve joint learning of prediction and optimization, minimizing the prediction loss, adjusting the weights... Optimize: in, Indicates will The expected decision loss obtained after substitution; To give the first An uncertain weight; through optimization To minimize prediction loss, obtain weights that reflect the relative importance of uncertainty in decision-making tasks, and thus generate prediction results oriented towards decision-making; Building a differentiable agent model using a data-driven approach approximation and Relationship between them: Proxy model The fitting performance is evaluated by the following formula: in, The dataset used to train the proxy model contains For the sample .

[0013] As a preferred embodiment of the weighted prediction-decision integration method considering the impact of uncertainty described in this invention, the method of optimizing the weight parameters using a surrogate model to obtain the optimal value further includes: a surrogate model... Optimize by minimization: After optimization After that, it will be realized from arrive A precise approximation, and further, through Perform gradient descent on top, for Optimize to minimize PDPL: in, The number of iterations. The learning rate; Repeat the iteration until The optimal weight setting can be obtained by converging to the minimum value. .

[0014] Another objective of this invention is to provide a weighted prediction-decision integrated system that takes into account the effects of uncertainty.

[0015] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a weighted prediction-decision integrated system considering the impact of uncertainty, comprising: a prediction model construction module, a prediction task generation module, and an optimization module; The prediction model building module generates a weighted dataset, trains prediction models with different weight parameters, and accumulates data. The prediction task generation module trains a prediction model and obtains a prediction task for each weight, and calculates the decision loss. The optimization module constructs and trains a fitted surrogate model, and applies the surrogate model to optimize the weight parameters to obtain the optimal values.

[0016] The present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, characterized in that the processor executes the computer program to implement the steps of the weighted prediction-decision integration method considering the influence of uncertainty.

[0017] The present invention provides 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 weighted prediction-decision integration method that takes into account the effects of uncertainty.

[0018] The beneficial effects of this invention are as follows: This invention achieves direct alignment between prediction model training and downstream decision-making objectives by constructing a closed-loop optimization framework centered on problem-oriented prediction loss (PDPL).

[0019] First, optimizable uncertainty-specific weights are introduced into the loss function of the prediction model to break the limitations of traditional equal weighting. Then, the suboptimal decision-making caused by prediction error is explicitly quantified through PDPL, transforming the prediction evaluation standard from statistical accuracy to decision utility.

[0020] To address the challenge of optimizing the highly non-convex relationship between PDPL and weights, a data-driven approach is innovatively used to train a differentiable surrogate model to approximate this complex relationship. Based on this surrogate model, efficient gradient descent is performed to automatically learn the optimal weight allocation that maximizes decision quality.

[0021] This invention enables power system operation to intelligently identify and accurately predict the most critical uncertainties affecting decision-making objectives such as safety and economy, ultimately significantly improving the overall performance and adaptability of operation optimization under limited resource conditions. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 The overall flowchart of a weighted prediction-decision integration method considering the impact of uncertainty is provided in one embodiment of the present invention. Detailed Implementation

[0024] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0025] Example 1, referring to Figure 1 This is one embodiment of the present invention, which provides a weighted prediction-decision integration method that takes into account the effects of uncertainty, including: S100: Generate a weighted dataset, train a prediction model with different weight parameters, and accumulate data; S200. Train the prediction model for each weight and obtain the prediction task, then calculate the decision loss. S300. Construct and train a fitted surrogate model, and apply the surrogate model to optimize the weight parameters to obtain the optimal value; It should be noted that forecasting is an effective method for quantifying uncertainty and an important reference for informed optimization under conditions of uncertainty. Combining forecasting with optimization can improve decision quality by generating near-optimal solutions. Given that forecasting errors associated with various uncertainties have different impacts on downstream decision-making, improving the accuracy of forecasts for key uncertainties that significantly affect decision quality can lead to better optimization results.

[0026] Therefore, to address the aforementioned problems, this invention proposes a novel Weighted Prediction and Optimization (WPO) framework for decision-making under uncertainty conditions through steps S100-S300. Specifically, this invention introduces an uncertainty-aware weighting mechanism into the prediction model to capture the relative importance of each uncertainty to a specific optimization task, and introduces a Problem-Driven Prediction Loss (PDPL) to quantify the suboptimality of weighted prediction relative to perfect prediction for downstream optimization. By optimizing the uncertainty weights to minimize PDPL, the WPO framework enables adaptive assessment of the impact of uncertainty and ensemble learning of prediction and optimization. Furthermore, to facilitate weight optimization, this invention constructs a surrogate model to establish the mapping relationship between weights and PDPL.

[0027] Example 2, refer to Figure 1 This is one embodiment of the present invention, which provides a weighted prediction-decision integration method that takes into account the effects of uncertainty, including: It should be noted that, compared to existing methods, this approach facilitates prediction and optimization by prioritizing key uncertainties that significantly impact the optimization results. In power system operations with multiple uncertainties, the impact of prediction errors varies greatly depending on the specific problem characteristics and the role of each uncertainty in the optimization process. For example, in voltage control problems, load demand prediction errors at nodes with lower voltage safety margins have a greater impact on system safety. Therefore, accurately predicting key uncertainties can mitigate the negative impact of prediction errors and improve optimization quality.

[0028] Typically, weights are introduced into the prediction loss function to indicate the relative importance of uncertainties. All other things being equal, prediction models tend to reduce prediction errors for uncertainties with higher weights. Based on this, this invention proposes a novel Weighted Prediction and Optimization (WPO) framework for uncertainty management in power system operation. The WPO framework introduces an uncertainty-aware weighting mechanism into the prediction model to capture the relative importance of each uncertainty factor. This relative importance is quantified by a Problem-Driven Prediction Loss (PDPL), which explicitly quantifies the suboptimality of weighted prediction relative to perfect prediction in specific decision-making. By optimizing the weights to minimize the PDPL, the weights act as a bridge between prediction and optimization, generating predictions for specific decision tasks, thereby improving interpretability and adaptability. To facilitate weight optimization, this invention develops a surrogate model to capture the relationship between weights and PDPL, and optimizes the weights by performing gradient descent on the surrogate model.

[0029] In this embodiment of the invention, S100 generates a weighted dataset, trains a prediction model under different weight parameters, and accumulates data, including the following steps S101-S102: S101. Generally speaking, the prediction task can be described as follows: ,in For parametric prediction models, These are model parameters; These are observable features; for A prediction vector for an uncertain variable; This is the corresponding actual implementation.

[0030] Predictive performance is measured by a pre-defined loss function to determine the overall deviation of all uncertain variables: in, For loss due to a single uncertain variable; for deterministic predictions, mean absolute error and mean squared error are commonly used as... .

[0031] Given the prediction model structure and loss function, the goal of training the prediction model is to determine the optimal parameters. This minimizes the expected loss: in, For expectation operators; This represents the training dataset used for the prediction task, containing... One sample. Then, the prediction results... Embedded optimization models to support decision-making under uncertainty: in, As decision variables, For feasible sets, The objective function is denoted as .

[0032] S102, based on the mathematical expression of S101, in the traditional paradigm, the loss function is weighted equally. When dealing with various uncertainties, we should ignore the differences in their impact on decision-making. However, if key uncertainties are predicted inaccurately, they will significantly reduce the quality of decision-making.

[0033] To address this, this invention introduces variable-specific weights into the prediction model to highlight key uncertainties; and proposes a problem-oriented prediction loss (PDPL) to explicitly quantify the suboptimal nature of the prediction model for downstream decision-making. Furthermore, by optimizing the weights to minimize the PDPL, an adaptive fusion of prediction and optimization is achieved.

[0034] First, variable-specific weights are introduced into the loss function, reformulating the traditional loss function into a weighted loss function: in, To give the first A weight with uncertainty, satisfying and ; In particular, when At this point, the weighted loss function degenerates into the traditional loss function.

[0035] The process of training a weighted prediction model under a weighted loss function can be represented as follows: in, Indicates in weight The optimal parameters of the prediction model based on the weighted loss function are then determined.

[0036] In this embodiment of the invention, step S200 involves training a prediction model for each weight to obtain a prediction task, and calculating the decision loss, including the following steps S201-S203: S201, Given the prediction result ,Depend on Derived optimal decision Will be realized After being revealed, it was applied. For quantification Compared to Derived optimal decision Due to the suboptimal nature of the decision, a decision loss is introduced: in, The objective function is... Obviously When the prediction is perfect... hour, Based on our predictive model Problem-oriented prediction loss (PDPL) is defined as the loss of a dataset... Expected decision loss under all uncertainties: The PDPL directly correlates prediction performance with optimization quality, providing a problem-oriented metric for prediction performance.

[0037] In an embodiment of the present invention, S300 involves constructing and training a fitted surrogate model, and applying the surrogate model to optimize the weight parameters to obtain the optimal value, including the following steps S301-S302: S301. To achieve joint learning of prediction and optimization, we aim to minimize the PDPL in the weights. Optimize: in, Indicates will The expected decision loss obtained after substitution.

[0038] Weight It acts as a bridge between prediction and subsequent optimization.

[0039] Through optimization By minimizing PDPL, we can obtain weights that reflect the relative importance of each uncertainty in the decision-making task, thereby generating decision-oriented prediction results and improving decision quality.

[0040] However, predictive models Typically represented by complex neural networks, and It includes optimization terms, making the middle The optimization problem is highly nonconvex and difficult to solve directly.

[0041] Given and The relationships between them are highly non-convex. This invention uses a data-driven approach to construct a differentiable proxy model. To approximate this relationship: ; Proxy Model The fitting performance can be evaluated by the following formula: in The dataset used to train the proxy model contains For the sample ; S302, Proxy Model Optimize by minimization: Obtain the optimized After that, it can be realized from arrive A precise approximation.

[0042] Furthermore, through Performing gradient descent on top can... Optimize to minimize PDPL: in The number of iterations. The learning rate is used. Repeat the iteration until... The optimal weight setting can be obtained by converging to the minimum value. .

[0043] Example 3 is an embodiment of the present invention. The above is an illustrative scheme of the weighted prediction-decision integration method considering the impact of uncertainty. It should be noted that the technical solution of the weighted prediction-decision integration system considering the impact of uncertainty and the technical solution of the weighted prediction-decision integration method considering the impact of uncertainty described above belong to the same concept. For details not described in detail in the technical solution of the weighted prediction-decision integration system considering the impact of uncertainty in this embodiment, please refer to the description of the technical solution of the weighted prediction-decision integration method considering the impact of uncertainty described above.

[0044] This embodiment provides a weighted prediction-decision integrated system that takes into account the impact of uncertainty, including: a prediction model construction module, a prediction task generation module, and an optimization module; The prediction model building module generates a weighted dataset, trains prediction models with different weight parameters, and accumulates data. The prediction task generation module trains a prediction model and obtains a prediction task for each weight, and calculates the decision loss. The optimization module constructs and trains a fitted surrogate model, and applies the surrogate model to optimize the weight parameters to obtain the optimal values.

[0045] This embodiment also provides an electronic device applicable to the weighted prediction-decision integration method considering the impact of uncertainty, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the weighted prediction-decision integration method considering the impact of uncertainty as proposed in the above embodiment.

[0046] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, it implements the weighted prediction-decision integration method considering the impact of uncertainty as proposed in the above embodiments.

[0047] The storage medium proposed in this embodiment and the weighted prediction-decision integration method considering the impact of uncertainty proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0048] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0049] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A weighted forecasting-decision integration method considering the impact of uncertainty, characterized by: include, Generate a weighted dataset, train prediction models with different weight parameters, and accumulate data. For each weight, train the prediction model and obtain the prediction task, then calculate the decision loss; Construct and train a fitted surrogate model, and apply the surrogate model to optimize the weight parameters to obtain the optimal values.

2. The weighted prediction-decision integration method considering the impact of uncertainty as described in claim 1, characterized in that: The prediction model includes constructing a parameterized weighted prediction model for a prediction task containing two or more uncertain variables, wherein the corresponding loss function contains adjustable weight parameters that correspond one-to-one with each other.

3. The weighted prediction-decision integration method considering the impact of uncertainty as described in claim 2, characterized in that: The prediction model also includes defining a problem-oriented prediction loss function based on the optimization objective and decision variables of the decision task, and quantifying the degree of suboptimality of the decision scheme derived from the prediction results of the weighted prediction model relative to the optimal decision scheme derived from the true value of uncertainty.

4. The weighted prediction-decision integration method considering the impact of uncertainty as described in claim 3, characterized in that: The prediction model also includes introducing variable-specific weights into the loss function, and reformulating the traditional loss function into a weighted loss function; The weighted prediction model is trained under the weighted loss function to obtain the optimal parameters of the prediction model based on the weighted loss function.

5. The weighted prediction-decision integration method considering the impact of uncertainty as described in claim 4, characterized in that: The calculation of decision loss includes, based on a given prediction result ,Depend on Derived optimal decision Will be realized After being revealed, it is applied to ; for quantification Compared to Derived optimal decision Due to the suboptimal nature of the decision, a decision loss is introduced: in, The objective function is... When the prediction is perfect... hour, ; Based on the prediction model The prediction loss for a problem-oriented dataset is defined as follows: The expected decision loss under all uncertainties is: in, As the expected value, the current prediction loss directly correlates prediction performance with optimization quality.

6. The weighted prediction-decision integration method considering the impact of uncertainty as described in claim 5, characterized in that: The application of the proxy model to optimize the weight parameters to obtain optimal values ​​includes, to achieve joint learning of prediction and optimization, minimizing the prediction loss, adjusting the weights... Optimize: in, Indicates will The expected decision loss obtained after substitution; To give the first An uncertain weight; through optimization To minimize prediction loss, obtain weights that reflect the relative importance of uncertainty in decision-making tasks, and thus generate prediction results oriented towards decision-making; Building a differentiable agent model using a data-driven approach approximation and Relationship between them: Proxy model The fitting performance is evaluated by the following formula: in, The dataset used to train the proxy model contains For the sample .

7. The weighted prediction-decision integration method considering the impact of uncertainty as described in claim 6, characterized in that: The optimization of weight parameters using the proxy model to obtain optimal values ​​also includes the proxy model. Optimize by minimization: After optimization After that, it will be realized from arrive A precise approximation, and further, through Perform gradient descent on top, for Optimize to minimize PDPL: in, The number of iterations. The learning rate; Repeat the iteration until The optimal weight setting can be obtained by converging to the minimum value. .

8. A weighted prediction-decision integration system considering the impact of uncertainty, employing the weighted prediction-decision integration method considering the impact of uncertainty as described in any one of claims 1 to 7, characterized in that, include: Prediction model building module, prediction task generation module, optimization module; The prediction model building module generates a weighted dataset, trains prediction models with different weight parameters, and accumulates data. The prediction task generation module trains a prediction model and obtains a prediction task for each weight, and calculates the decision loss. The optimization module constructs and trains a fitted surrogate model, and applies the surrogate model to optimize the weight parameters to obtain the optimal values.

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 weighted prediction-decision integration method considering the effects of uncertainty as described in 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 the processor, it implements the steps of the weighted prediction-decision integration method considering the effects of uncertainty as described in any one of claims 1 to 7.

Images