Power prediction model evaluation method based on Bayesian model averaging and related device

Abstract

The invention discloses a Bayesian model averaging-based power prediction model evaluation method and a related device, and belongs to the technical field of power system prediction. The method comprises the following steps: constructing a candidate model set comprising a plurality of power prediction models; secondly, historical data are used for training all the models, the posterior model probability of each model is calculated based on the Bayesian theorem, the posterior model probability serves as the scientific weight of the model, and the goodness of fit and complexity of the model are considered in the weight at the same time; and finally, for a new prediction input, performing weighted average on the prediction distribution of each candidate model by taking the posterior probability as the weight to generate comprehensive probability prediction distribution. According to the method, the advantages and disadvantages of each candidate model are scientifically evaluated through the posterior probability, and a comprehensive and probabilistic prediction result is finally generated, so that the robustness and reliability of prediction are improved, and richer decision information is provided for power grid dispatching.

Description

Technical Field

[0001] This invention belongs to the field of power system prediction technology, and relates to a method and related apparatus for evaluating power prediction models based on Bayesian model averaging. Background Technology

[0002] With the acceleration of the global energy transition, the penetration rate of renewable energy sources such as wind power and solar power in the power system continues to rise. However, the inherent intermittency, randomness, and volatility of renewable energy output pose serious challenges to the real-time balance, safe and stable operation, and efficient absorption of the power grid. Against this backdrop, high-precision power forecasting technology has become a key technology for alleviating these contradictions and ensuring the safe and economical operation of the power grid. Accurate power forecasting can provide forward-looking decision support for grid dispatching departments and is an important cornerstone for improving the flexibility and reliability of the power system.

[0003] Currently, in the construction and application of power prediction models, how to select the optimal model from numerous candidate models and make reliable predictions is a core issue facing the industry. Existing model evaluation and selection methods mainly fall into two categories. The first is the "single model selection method," which selects only the single model with the best performance on the test set from multiple candidate models for final prediction. This method has inherent flaws such as "ignoring model uncertainty" and "insufficient information utilization." Its single-model selection mechanism leads to a lack of robustness in the prediction results and fails to provide the probabilistic prediction information necessary for risk assessment. The second is the "simple model averaging method." To overcome the limitations of a single model, this method attempts to comprehensively utilize information from multiple models, performing a simple arithmetic average of the prediction results or a weighted average based on historical performance (such as the inverse of the error on the validation set) with fixed weights. Although this method improves the robustness of predictions to some extent, its weight determination method is often subjective and lacks solid statistical theoretical support. Simple weighting strategies cannot scientifically measure the relative reliability of each model and are difficult to adapt to the dynamic characteristics of the prediction object. More importantly, the prediction intervals generated by such methods often have poor calibration, meaning that the predicted probability distribution is not highly consistent with the distribution of the actual observed values, and cannot truly reflect the level of uncertainty in the prediction, thus weakening its reference value in actual decision-making.

[0004] In summary, there is an urgent need in this field for a new method for model evaluation and combination that can quantify model uncertainty, comprehensively utilize information from multiple models, and provide highly reliable probabilistic predictions. This would provide stronger technical support for the refined scheduling and risk management of smart grids and the energy internet. Summary of the Invention

[0005] The purpose of this invention is to provide a power prediction model evaluation method and related apparatus based on Bayesian model averaging, so as to solve the technical problems of low accuracy and poor reliability of existing model evaluation methods.

[0006] To achieve the above objectives, the present invention employs the following technical solution: In a first aspect, the present invention provides a method for evaluating a power prediction model based on Bayesian model averaging, comprising the following steps: Acquire historical power data and corresponding meteorological data, and construct K different power prediction models as a candidate model set; Each power prediction model in the candidate model set is trained using historical data, and the posterior model probability (PMP) of each model is calculated based on Bayes' theorem. The power prediction models in the candidate model set are evaluated and ranked based on the posterior model probability (PMP).

[0007] Furthermore, the method also includes: The new prediction input point X_new is input into the candidate model set to generate the final probability prediction distribution; the probability prediction distribution is a weighted mixture distribution of the prediction distributions of each candidate model with their posterior model probabilities as weights. Based on the probability prediction distribution, point prediction values ​​and interval prediction values ​​are extracted to obtain the probability prediction results.

[0008] Furthermore, the specific calculation formula for the probability prediction distribution is as follows:

[0009] In the formula, It is the power prediction model M k The probability density function for predicting the value y given the input X_new; The weights of the model are equal to the posterior model probability PMP.

[0010] Furthermore, the formula for calculating the posterior model probability PMP is as follows:

[0011] In the formula, P(M) represents the posterior model probability PMP; k ) is model M k The prior probability; P(D | M k ) is model M k The marginal likelihood.

[0012] Furthermore, the prior probability P(M) k P(M) is set as a uniform prior. k ) = 1 / K.

[0013] Furthermore, the marginal likelihood P(D|M) k The calculation is performed using the Bayesian Information Criterion (BIC), the Akaike Information Criterion (AIC), or the Markov Chain Monte Carlo (MCMC) sampling method; P(D | M) is calculated using the Bayesian Information Criterion (BIC). k The specific calculation formula is as follows:

[0014] In the formula, LLk is model M k Maximum likelihood value, p k Model M k The number of parameters, N is the sample size.

[0015] Furthermore, the candidate model set includes one or more combinations of physical models, statistical models, and machine learning models.

[0016] Secondly, the present invention provides a power prediction model evaluation system based on Bayesian model averaging, comprising: The model set construction module is used to acquire historical power data and corresponding meteorological data, and construct K different power prediction models as a candidate model set. The PMP calculation module is used to train each power prediction model in the candidate model set using historical data and to calculate the posterior model probability PMP for each model based on Bayes' theorem. The evaluation module is used to evaluate and rank the power prediction models in the candidate model set based on the posterior model probability (PMP).

[0017] Thirdly, the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the power prediction model evaluation method based on Bayesian model averaging as described above.

[0018] Fourthly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the power prediction model evaluation method based on Bayesian model averaging as described above.

[0019] Compared with the prior art, the present invention has the following beneficial effects: This invention discloses a method and related apparatus for evaluating power prediction models based on Bayesian model averaging. By establishing a candidate model set and calculating the posterior model probability, a quantitative and statistically significant evaluation index based on Bayesian theory is provided for each candidate model. This index simultaneously considers the model's fitting accuracy and complexity, avoiding overfitting models from receiving excessive weights, resulting in more scientific and reliable evaluation results. In the field of power prediction, "model selection uncertainty" is quantified. The distribution of the weights of each model directly reflects this uncertainty. Dispersed weights indicate high model uncertainty, while concentrated weights indicate low model uncertainty. This provides a crucial basis for judging the reliability of prediction results.

[0020] Furthermore, this invention effectively integrates the advantages of multiple models through BMA model averaging, avoiding the risk of single model failure and significantly improving the robustness and accuracy of the prediction results. Simultaneously, the final output probability prediction incorporates both "model uncertainty" and "random uncertainty," resulting in better calibration of the prediction interval and a more realistic reflection of prediction risk.

[0021] Furthermore, this invention provides power grid dispatching departments with probabilistic prediction results containing uncertainty information, enabling them to make risk-aware decisions, such as formulating more conservative or more aggressive strategies in power trading, thereby improving the economy and security of power grid operation. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a schematic diagram of the system of the present invention; Figure 3 This is a flowchart illustrating the technical route for evaluating the power prediction model based on BMA in an embodiment of the present invention. Figure 4 This is a schematic diagram of the posterior model probability (PMP) for each model in the embodiments of the present invention; Figure 5 This is a graph showing the RMSE comparison results of various models in the embodiments of the present invention; Figure 6 This is a schematic diagram of the prediction results of each model in the embodiments of the present invention. Detailed Implementation

[0024] The present invention will now be described in detail with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0025] The following detailed description is exemplary and intended to provide further detailed explanation of the invention. Unless otherwise specified, all technical terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terminology used in this invention is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention.

[0026] See Figure 1 and Figure 3 This invention discloses an evaluation method for power prediction models based on Bayesian model averaging, comprising the following steps: S1. Obtain historical power data and corresponding meteorological data, and construct K different power prediction models as a candidate model set; Obtain historical power data and corresponding meteorological data, and construct K different power prediction models as a candidate model set M = {M1, M2, ..., M}. k The candidate models may include physical models, statistical models, machine learning models, or any combination thereof. Furthermore, the candidate model set M can be dynamically changed, for example, by periodically introducing new models or eliminating models that consistently perform poorly, enabling the system to have self-evolution capabilities.

[0027] S2, train each power prediction model in the candidate model set using historical data, and calculate the posterior model probability PMP for each model based on Bayes' theorem; Each candidate model is trained using historical data D, and the posterior model probability (PMP) of each model is calculated based on Bayes' theorem, which is then used as the weights of that model. w k .

[0028] The posterior model probability P(M) k The formula for calculating |D) is:

[0029] Among them, P(M k ) is model M k The prior probability is set to a uniform prior, typically 1 / K by default, indicating no preference for any model before seeing the data. P(D | M) k ) is model M k The marginal likelihood represents the model's ability to fit the observed data D. Preferably, in the prior probability P(Mk In terms of setting, uniform priors can be avoided, and different prior probabilities can be set for different models based on domain knowledge or historical experience.

[0030] To simplify the calculation, the Bayesian Information Criterion (BIC) can be used for approximate calculation:

[0031] in, LL k Model M k Maximum likelihood value, p k Model M k The number of parameters, N, is the sample size. Preferably, in calculating the marginal likelihood P(D|M)... k In addition to using the BIC approximation, more accurate estimates can be obtained by employing the Akaike Information Criterion (AIC), Markov Chain Monte Carlo (MCMC) sampling method, or other Bayesian computational methods.

[0032] S3. Evaluate and rank the power prediction models in the candidate model set based on the posterior model probability (PMP).

[0033] The models are evaluated and ranked based on the posterior probabilities calculated in step S2. The higher the posterior probability, the better the model is, and the greater its contribution to the final prediction result.

[0034] S4, input the new prediction input point X_new into the candidate model set to generate the final probability prediction distribution; the probability prediction distribution is a weighted mixture distribution of the prediction distributions of each candidate model with their posterior model probabilities as weights; For a new prediction input point X_new, the final prediction output of this invention is not a single value, but a probability distribution. This probability distribution is a weighted mixture of the prediction distributions of each candidate model, with their posterior probabilities as weights.

[0035] The BMA prediction probability density function is:

[0036] in, Model M k The probability density function for predicting the value y given the input X_new.

[0037] S5. Based on the probability prediction distribution, extract the point prediction value and the interval prediction value to obtain the probability prediction result.

[0038] Based on the BMA prediction probability distribution generated in step S4, point prediction values ​​(such as the mean) and interval predictions (such as the 95% confidence interval) can be extracted to form a complete probability prediction result.

[0039] See Figure 2 This invention discloses a power prediction model evaluation system based on Bayesian model averaging, comprising a model set construction module, a PMP calculation module, and an evaluation module. The model set construction module acquires historical power data and corresponding meteorological data to construct K different power prediction models as a candidate model set. The PMP calculation module trains each power prediction model in the candidate model set using historical data and calculates the posterior model probability (PMP) of each model based on Bayes' theorem. The evaluation module evaluates and ranks the power prediction models in the candidate model set based on the posterior model probability (PMP). This invention aims to quantify the uncertainty of model selection, scientifically evaluate the merits of each candidate model through posterior probability, and ultimately generate a comprehensive, probabilistic prediction result, thereby improving the robustness and reliability of prediction and providing richer decision-making information for power grid dispatching. See also... Figure 4 , Figure 5 and Figure 6 By comparing traditional single-model predictions with the power prediction model evaluation method based on Bayesian model averaging proposed in this invention, it can be seen that the BMA fusion prediction method of this invention is closer to the true value. No single model's prediction line can maintain the same high accuracy as the BMA mean curve. The prediction interval generated by this invention is based on scientifically calculated weights (PMP), resulting in better calibration. This provides a reliable risk quantification tool for power grid dispatching. In contrast, the intervals generated by simple model averaging methods are often too wide or too narrow, resulting in poor calibration.

[0040] In one embodiment of the present invention, a computer device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions from the computer storage medium to achieve a corresponding method flow or corresponding function. The processor described in this embodiment of the present invention can be used in the operation of a power prediction model evaluation method based on Bayesian model averaging.

[0041] This invention also provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the power prediction model evaluation method based on Bayesian model averaging in the above embodiments.

[0042] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied 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.

[0043] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations 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, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0044] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0045] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0046] Finally, 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 the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A Bayesian model averaging based power prediction model evaluation method, characterized in that, The method comprises the following steps: obtaining historical power data and corresponding meteorological data, and constructing K different power prediction models as a candidate model set; training each power prediction model in the candidate model set using historical data, and calculating the posterior model probability PMP of each model based on Bayes' theorem; evaluating and ranking the power prediction models in the candidate model set based on the posterior model probability PMP.

2. The method of claim 1, wherein the Bayesian model averaging-based power prediction model evaluation method is characterized by, Further comprising: inputting a new prediction input point X_new into the candidate model set to generate a final probability prediction distribution; the probability prediction distribution is a weighted mixture distribution of each candidate model prediction distribution weighted by its posterior model probability; extracting point prediction values and interval prediction values based on the probability prediction distribution to obtain a probability prediction result.

3. The method of claim 2, wherein the Bayesian model averaging-based power prediction model evaluation method is characterized by, The specific calculation formula of the probability prediction distribution is: wherein is the power prediction model M k the probability density function of the predicted value y at a given input X_new; are the weights of the model, equal to the posterior model probability PMP.

4. The method of claim 1, wherein the Bayesian model averaging-based power prediction model evaluation method is characterized by, The calculation formula of the posterior model probability PMP is: wherein denotes the posterior model probability PMP; P(M k ) is the prior probability of the model M k ; P(D | M k ) is the marginal likelihood of the model M k .

5. The Bayesian model averaging based power prediction model evaluation method of claim 4, wherein, The prior probability P(M k ) is set as a uniform prior, P(M k ) = 1 / K.

6. The Bayesian model averaging based power prediction model evaluation method of claim 4, wherein, The marginal likelihood P(D | M k ) is calculated by using the Bayesian information criterion BIC, the Akaike information criterion AIC, or the Markov chain Monte Carlo MCMC sampling method; and the specific calculation formula of P(D | M k ) calculated by using the Bayesian information criterion BIC is: where, LLkis the maximum likelihood value of the model M k p k is the number of parameters of the model M k and N is the sample size.

7. The Bayesian model averaging based power prediction model evaluation method of claim 4, wherein, The candidate model set comprises one or a combination of physical models, statistical models and machine learning models.

8. A Bayesian model averaging based power prediction model evaluation system, characterized by, It comprises: a model set construction module for obtaining historical power data and corresponding meteorological data, and constructing K different power prediction models as a candidate model set; a PMP calculation module for training each power prediction model in the candidate model set using historical data, and calculating the posterior model probability PMP of each model based on Bayes' theorem; an evaluation module for evaluating and ranking the power prediction models in the candidate model set based on the posterior model probability PMP.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the computer program to realize the steps of the power prediction model evaluation method based on Bayesian model averaging according to any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1-9. The computer program is executed by the processor to realize the steps of the power prediction model evaluation method based on Bayesian model averaging according to any one of claims 1-7.

Images