CN117556706A - Wind power probability prediction method and system based on diffusion model and non-stationary signal enhancement - Google Patents
Wind power probability prediction method and system based on diffusion model and non-stationary signal enhancement Download PDFInfo
- Publication number
- CN117556706A CN117556706A CN202311581023.1A CN202311581023A CN117556706A CN 117556706 A CN117556706 A CN 117556706A CN 202311581023 A CN202311581023 A CN 202311581023A CN 117556706 A CN117556706 A CN 117556706A
- Authority
- CN
- China
- Prior art keywords
- wind power
- diffusion
- signal enhancement
- diffusion model
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000009792 diffusion process Methods 0.000 title claims abstract description 158
- 238000000034 method Methods 0.000 title claims abstract description 144
- 230000008569 process Effects 0.000 claims abstract description 52
- 238000012549 training Methods 0.000 claims abstract description 45
- 238000009826 distribution Methods 0.000 claims abstract description 21
- 238000005070 sampling Methods 0.000 claims abstract description 19
- 230000007246 mechanism Effects 0.000 claims abstract description 16
- 238000005096 rolling process Methods 0.000 claims abstract description 12
- 230000006870 function Effects 0.000 claims description 19
- 238000004590 computer program Methods 0.000 claims description 10
- 239000013598 vector Substances 0.000 claims description 9
- 238000003860 storage Methods 0.000 claims description 6
- 238000005259 measurement Methods 0.000 claims description 5
- 230000007704 transition Effects 0.000 claims description 4
- 238000013398 bayesian method Methods 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000009795 derivation Methods 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 238000004088 simulation Methods 0.000 claims description 3
- 238000012546 transfer Methods 0.000 claims description 2
- 238000012544 monitoring process Methods 0.000 abstract description 3
- 230000008901 benefit Effects 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 238000011161 development Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 108090000672 Annexin A5 Proteins 0.000 description 2
- 238000009825 accumulation Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000010248 power generation Methods 0.000 description 2
- 238000010521 absorption reaction Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000010485 coping Effects 0.000 description 1
- 230000029087 digestion Effects 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000011065 in-situ storage Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/10—Pre-processing; Data cleansing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2113/00—Details relating to the application field
- G06F2113/06—Wind turbines or wind farms
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/06—Power analysis or power optimisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Medical Informatics (AREA)
- Geometry (AREA)
- Computer Hardware Design (AREA)
- Software Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a wind power probability prediction method and system based on a diffusion model and non-stationary signal enhancement. The method comprises the following steps: modeling the uncertainty of the historical wind power by using a diffusion model, and sampling the probability distribution characteristics of the wind power through a diffusion process and a denoising process; grasping the non-stationary fluctuation trend of the future wind power by using the historical wind power sequence and the future weather forecast data, constructing a non-stationary signal enhancement method, serving as a kernel solver of a diffusion model denoising process, and rolling and training a diffusion model based on a shared parameter mechanism; based on a diffusion model which is obtained by training and takes non-stationary signal enhancement as a kernel solver, wind power multi-step probability prediction results are obtained by utilizing a double-layer sampling method of edge diffusion step length and time sequence reasoning. The method can be deployed in a power grid monitoring and dispatching department or a wind power station, provides accurate multi-time scale wind power probability prediction results, and meets the requirements of power grid monitoring and economic safety regulation.
Description
Technical Field
The invention belongs to the technical field of renewable energy development and utilization, and particularly relates to a wind power probability prediction method and system based on a diffusion model and non-stationary signal enhancement.
Background
The construction of smart grids containing new energy sources is playing an increasingly important role in coping with global climate change and realizing global economic and social sustainable development. As the installed cost of wind power is continuously reduced, the wind power generation in China is in a situation of rapid development.
However, the non-stationary nature of wind power exhibits time-series fluctuations in wind power mean and variance, which are important factors that cause fluctuations in grid frequency, and in-situ digestion to be hindered. Also, modeling of wind power uncertainty is lacking. Therefore, it is necessary to implement accurate wind power probability prediction in consideration of non-stationary fluctuation of wind power, and provide data support for safe and stable scheduling of the power grid.
Wind power probability prediction is largely divided into parametric prediction and nonparametric prediction. Although the nonparametric prediction method has the advantage of high calculation efficiency, the probability density function of the wind power is difficult to be fitted integrally, the problem of error accumulation of multi-step prediction is easy to be caused, and the improvement of the accuracy and the reliability of wind power probability prediction is limited. Therefore, grasping the time sequence non-stationary characteristic of the wind power and accurately predicting the probability distribution of the wind power is a current wind power prediction difficulty.
Disclosure of Invention
The invention aims to: aiming at the difficulty of wind power probability prediction, the invention provides a wind power probability prediction method and a wind power probability prediction system based on a diffusion model and non-stationary signal enhancement, and the performance of the hour-level wind power probability prediction is effectively improved.
The technical scheme is as follows: in order to achieve the above object, the wind power probability prediction method based on diffusion model and non-stationary signal enhancement provided by the invention comprises the following steps:
modeling the uncertainty of the historical wind power by using a diffusion model, wherein the diffusion model comprises a diffusion process and a denoising process, the diffusion process is used for obtaining a historical wind power probabilistic modeling result by continuously adding Gaussian noise to the original wind power until the characteristics of the diffusion process are completely destroyed, and the denoising process is used for fitting Gaussian distribution parameters of each step of diffusion process by using a kernel solver;
grasping the non-stationary fluctuation trend of the wind power without coming by utilizing the key parameters of the historical wind power stability and the future wind speed forecast information, constructing a non-stationary signal enhancement method, taking the non-stationary signal enhancement method as a kernel solver of a diffusion model denoising process, and rolling and training a diffusion model based on a shared parameter mechanism;
based on a diffusion model which is obtained by training and takes non-stationary signal enhancement as a kernel solver, wind power multi-step probability prediction results are obtained by utilizing a double-layer sampling method of edge diffusion step length and time sequence reasoning.
Further, the diffusion process is expressed as:
wherein q (x n |x n-1 ) Representing the Gaussian distribution, x, that the n-1 process needs to add n Features representing the nth step of the diffusion process;mean value of +.>Standard deviation is beta n A gaussian distribution of I; beta n The adding-to-noise ratio example of the n-th diffusion process is shown, and I represents an identity matrix;
by setting alpha n ＝1-β n Obtaining a direct analytic solution method of the diffusion process from the 0 th step to the n th step:
wherein the method comprises the steps ofThe backward propagation probability of the diffusion model is obtained by a Bayesian method:
wherein,and->Respectively the backward directions of the diffusion modelsThe expected and variance of propagation probabilities, q (x n |x n-1 )，q(x n ) And q (x) n-1 ) Are calculated by the diffusion process.
Further, the noise ratio β of the diffusion process n Expressed as:
where s is a positive real number and N represents the step size of the diffusion process.
Further, the denoising process of the diffusion model is to make the denoising probability p needing fitting θ (x n-1 |x n ) As close to the backward propagation probability q (x n-1 |x n ,x 0 ) The denoising probability is expressed as variance μ θ Standard deviation of sigma θ Gaussian distribution of (c)KL divergence is applied to establish a likelihood function to measure the difference between the learnable denoising probability and the actual diffusion probability:
wherein C is a constant and is irrelevant to the model parameter theta;representing a variable delimiter operator;
the mean value of the denoising probabilities is:
wherein, E is θ (x n N) is a learning diffusion model solver, and the diffusion model is trained by the solver E θ (x n N) realizing end-to-end training, and obtaining wind power probability prediction results by Gaussian noise sampling after the training is completed.
Further, the non-stationary signal enhancement method specifically includes:
characterizing the non-stationary characteristics of wind power fluctuations using the expectations and variances of the time series profiles, generalizing the historical wind power sequence into stationary sequences:
wherein Q' is a smoothed historical wind power sequence challenge vector, mu Q Sum sigma Q Respectively the expected and variance of the historical wind power sequence;
and calculating the similarity relation between the stabilized wind power sequences by using an improved multi-head attention mechanism to obtain a calculation method of a de-stabilized attention mechanism:
wherein f MLP Is a learnable linear layer d k The dimension of Q' is indicated and,the wind speed data representing weather forecast, K 'and V' are key vectors and value vectors of the smoothed historical wind power sequence, tau and delta represent key parameters for measuring the wind power stability, and the key parameters are represented by the historical wind power sequence and the weatherAnd forecasting the air forecast wind speed data.
Further, the kernel solver taking the non-stationary signal enhancement method as the diffusion model denoising process, the rolling training diffusion model based on the shared parameter mechanism comprises the following steps:
time sequence connection for representing wind power fluctuation by integrating hidden state h t The state h is hidden from the last moment t -1 Historical wind power sequence after denoisingWeather forecast information at time t>The derivation is expressed as:
wherein H is Datt () A non-stationary signal enhancement method is represented,a value of n representing a diffusion step at time t, f FFN,θ Representing a linear forward propagation function, ">A transition variable representing a hidden state update process;
the non-stationary enhancement method is used as a kernel solver of a denoising process of the diffusion model, a minimum negative log likelihood function is constructed by a transfer variable of a hidden state to be used as a loss function to train the diffusion model fused with the non-stationary signal enhancement method, training parameters are shared among the denoising processes during training, and the training process of the diffusion model fused with the non-stationary signal enhancement method is expressed as follows:
where e represents a standard gaussian distribution.
Further, the double-layer sampling method along the diffusion step length and the time sequence reasoning specifically comprises the following steps:
the denoising characteristics of the denoising process at the same moment are determined by the final value of the sampling diffusion processObtaining the predicted result of the n-1 th step denoising at the t moment +.>Expressed as:
wherein,representing the model obtained by training;
iterating from the n-1 st step to the 0 th step to obtain a wind power probability prediction result at the moment t
Based on hidden state h t Wind power measurement sequenceWeather forecast information at time t+1>Updating hidden state h of diffusion model obtained by training and fused with non-stationary signal enhancement method t+1 Further realizing rolling solution of the model to obtain wind power multi-step probability prediction result +.>Where T is the time step that needs to be predicted.
The invention also provides a wind power probability prediction system based on a diffusion model and non-stationary signal enhancement, which comprises:
the historical data simulation module is used for modeling uncertainty of historical wind power by using a diffusion model, the diffusion model comprises a diffusion process and a denoising process, the diffusion process is used for obtaining a historical wind power probabilistic modeling result by continuously adding Gaussian noise to the original wind power until the characteristics of the original wind power are completely destroyed, and the denoising process is used for fitting Gaussian distribution parameters of each step of diffusion process by using a kernel solver;
the model training module is used for grasping the non-stationary fluctuation trend of the future wind power by utilizing the key parameters of the historical wind power stability and the future wind speed forecast information, constructing a non-stationary signal enhancement method, taking the non-stationary signal enhancement method as a kernel solver in the denoising process of the diffusion model, and rolling and training the diffusion model based on a shared parameter mechanism;
the model prediction module is used for obtaining a wind power multi-step probability prediction result by utilizing a double-layer sampling method based on a diffusion model which is obtained through training and takes non-stationary signal enhancement as a kernel solver and by utilizing edge diffusion step length and time sequence reasoning.
The present invention also provides a computer device comprising: one or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, which when executed by the processors implement the steps of the wind power probability prediction method based on a diffusion model and non-stationary signal enhancement as described above.
The present invention also provides a computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the steps of a wind power probability prediction method based on a diffusion model and non-stationary signal enhancement as described above.
The beneficial effects are that: the invention aims at two major limitations faced by the related research surrounding wind power probability prediction at present: the uncertainty of wind power is difficult to accurately model, and the problem that prediction is generally stable is solved, and the wind power probability prediction method based on a diffusion model and non-stable signal enhancement is provided, wherein the uncertainty of wind power is modeled by using the diffusion model, so that the model does not depend on priori knowledge of wind power probability distribution, and an accurate wind power probability density function is obtained through progressive diffusion and denoising processes; and a non-stationary signal enhancement is designed as a kernel solver of a diffusion model denoising process to capture the non-stationary characteristic of wind power time sequence fluctuation, solve the problem of overstable ubiquitous in wind power prediction, introduce weather forecast data and non-stationary parameters to correct prediction results in real time, and can effectively improve the fitting capacity of wind power fluctuation and reduce error accumulation. The wind power probability prediction result which is 10 minutes to 2.5 hours in advance is provided by the invention, and the wind power probability prediction method can be applied to a power grid centralized prediction platform or a wind power plant on-site prediction system, so that the non-stationary fluctuation of wind power can be effectively predicted, the on-site wind power absorption capacity is improved, the requirements of power grid economic safety regulation are further met, and good social benefit and economic benefit are realized.
Drawings
FIG. 1 is a schematic flow chart of a prediction method of the present invention;
FIG. 2 is a schematic diagram of a non-stationary signal enhancement method according to the present invention;
FIG. 3 is a schematic diagram of a parameter sharing training method according to the present invention;
FIG. 4 is a graph showing the effect of wind power probability prediction in accordance with an embodiment of the present invention;
FIG. 5 is a graph showing the effect of multi-step prediction of stroke power in an embodiment of the present invention.
Detailed Description
The technical scheme of the invention is further described below with reference to the accompanying drawings.
Referring to fig. 1, the wind power probability prediction method based on diffusion model and non-stationary signal enhancement provided by the invention comprises the following steps:
step S1, modeling the uncertainty of the historical wind power by using a diffusion model, wherein the diffusion model comprises a diffusion process and a denoising process, the diffusion process is used for obtaining a historical wind power probabilistic modeling result by continuously adding Gaussian noise to the original wind power until the characteristics of the diffusion process are completely destroyed, and the denoising process is used for fitting Gaussian distribution parameters of each diffusion process by using a kernel solver.
In the invention, the diffusion model is also called a generated diffusion model, and in order to accurately model the uncertainty of wind power, the generated diffusion model samples the probability distribution characteristics of wind power through two stages of diffusion and denoising.
(1) The diffusion process can be expressed by continuously adding Gaussian noise to the original wind power until the characteristics are completely destroyed
The method is shown as follows:
wherein q (x n |x n-1 ) Representing the gaussian profile that the n-1 process needs to incorporate. X is x n Characterizing the nth step of the diffusion process.Mean value of +.>Standard deviation is beta n Gaussian distribution of I. Beta n The added noise ratio example of the n-th diffusion process is shown, and I represents the identity matrix.
Thus, by setting alpha n ＝1-β n The direct analytic solving method from the 0 th step to the n th step in the diffusion process can be obtained:
wherein,
then, the backward propagation probability of the diffusion model can also be obtained by a bayesian method:
wherein,and->The expected and variance, q (x) n |x n-1 )，q(x n ) And q (x) n-1 ) Can be calculated by the diffusion process. And through setting an N-step diffusion process, the original wind power sequence can be denoised into Gaussian noise.
Considering that the degree of noise added in the diffusion process increases with the diffusion step length, the noise adding proportion beta of the diffusion process n Can be expressed as:
where s is a positive real number and N represents the step size of the diffusion process.
(2) The denoising process of the diffusion model is to make the denoising probability p needing fitting θ (x n-1 |x n ) As close to the backward propagation probability q (x n-1 |x n ,x 0 ) The denoising probability can be expressed as a variance μ θ Standard deviation is sigma θ Gaussian distribution of (c)In order to accurately measure the difference between the learnable denoising probability and the actual diffusion probability, a KL divergence is applied to establish a likelihood function:
where C is a constant and is independent of the model parameter θ.Representing a variable delimiter operator.
Thus, the mean value of the denoising probability can be derived as:
wherein, E is θ (x n N) is a learnable diffusion model solver. The diffusion model goes through training solver epsilon θ (x n N) realizing end-to-end training, and obtaining wind power probability prediction results by Gaussian noise sampling after the training is completed.
And S2, grasping the non-stationary fluctuation trend of the future wind power by utilizing the key parameters of the historical wind power stability and the future wind speed forecast information, constructing a non-stationary signal enhancement method, taking the non-stationary signal enhancement method as a kernel solver of a diffusion model denoising process, and rolling and training a diffusion model based on a shared parameter mechanism.
As shown in fig. 2, the non-stationary signal enhancing method specifically includes:
first, consider that the non-stationary nature of wind power fluctuations can be characterized by the expectations and variances of the timing profile, generalizing the historical wind power sequence challenge vector into stationary sequences:
wherein Q' is the stabilized wind power sequence, mu x Sum sigma x The expected and variance of the historical wind power sequence, respectively. The same applies to key vectors and value vectors:
then, because the nonstationary characteristic of the future wind power is difficult to be completely inferred from the historical wind power sequence, the capability of the predictive model for grasping the nonstationary characteristic of the wind power can be increased by integrating the wind speed information of the weather forecast, so that the similarity relation between the stabilized wind power sequences can be calculated by an improved multi-head attention mechanism to obtain the unsteady attention mechanism calculating method H Datt ()：
Wherein f MLP Is a learnable linear layer d k For the dimension of the plateau-feature Q',and the wind speed data of the weather forecast are represented, and tau and delta represent key parameters for measuring the stability of wind power, and are predicted by the historical wind power sequence and the wind speed data of the weather forecast.
According to one embodiment, a non-stationary wind power enhancement module is established in step S2, and the non-stationary wind power enhancement module accesses the historical wind power sequence and future weather forecast data, and retains non-stationary characteristics in the historical wind power sequence through a stationary attention removing mechanism, so that the ubiquitous overstable problem is alleviated.
The method has the advantages that the diffusion model and the non-stationary signal enhancement method are integrated, and the non-stationary signal enhancement method is used as the diffusion model of the kernel solver, and the diffusion model comprises a model training process and a sampling solving process.
The training process of the dispersion model with the non-stationary signal enhancement method as the kernel solver is shown in fig. 3. First of all,to enrich the feature representation learning ability, hidden states are incorporated to characterize the timing relationship of wind power fluctuations. Hidden state h t Is to hide the state h from the last moment t-1 Denoised historical wind power sequenceWeather forecast information at time t>The derivation can be expressed as:
wherein,a value of n representing a diffusion step at time t, f FFN,θ Representing a linear forward propagation function, ">A transition variable representing a hidden state update process.
Secondly, a minimum negative log likelihood function can be constructed from the transition variables of the hidden states, and a diffusion-non-stationary signal modeling model can be trained as a loss function, and can be expressed as:
next, since the essence of the diffusion model fit is to train the kernel solver ε θ . The non-stationary enhancement method is used as a kernel solver in the denoising process of the diffusion model, so that the model can give consideration to the non-stationary fluctuation characteristic and uncertainty of wind power. During training, training parameters are shared among all denoising processes, and a model training process fused with a non-stationary signal enhancement method can be expressed as follows:
where e represents a standard gaussian distribution.
And step S3, obtaining a wind power multi-step probability prediction result by utilizing a double-layer sampling method based on a diffusion model which is obtained by training and takes non-stationary signal enhancement as a kernel solver and adopting edge diffusion step length and time sequence reasoning. This is the sampling solution process.
The double-layer sampling method based on edge diffusion step length and time sequence reasoning specifically comprises the following steps:
first, the denoising characteristics of the denoising process at the same time can be determined by the final value of the sampling diffusion processThus, the predicted result after denoising in step n-1 at time t is +.>Can be expressed as:
wherein,representing the model obtained by training.
Then, from the n-1 step iteration to the 0 step, a wind power probability prediction result at the moment t can be obtainedThen, based on the hidden state h t Wind power measurement sequence->Weather forecast information at time t+1>Updating hidden states by training-derived non-stationary signal enhancement modulesh t+1 Further realizing rolling solution of the prediction model to obtain wind power multi-step probability prediction resultWhere T is the time step that needs to be predicted.
The specific implementation process of wind power probability prediction by the method is described above, the data set provided by the American renewable energy laboratory is selected, and the historical wind power measurement information of a certain wind farm in michigan in 2010-2012 in U.S. is selected for model verification. The time resolution of wind power measurement was 10 minutes.
According to the multi-step probability prediction result of the photovoltaic power, three error analysis and evaluation indexes are adopted to evaluate the prediction performance of the model, namely a prediction interval bandwidth (PIAW), a Prediction Interval Coverage (PICP) and a comprehensive index Winkler Score (WS), and the prediction performance is expressed as follows:
wherein y is t The measured wind power value at time t is shown.And->The time t represents the lower and upper bounds of the confidence alpha probability interval. n is n s Representing the number of test set samples.
For the data set of the American renewable energy laboratory, data from 1 month in 2010 to 12 months in 2011 are selected as training samples, and data from 1 month in 2012 to 12 months in 2012 are selected as test samples. The wind power probability prediction error 1 hour in advance using the method of the present invention is shown in table 1. To verify the gain of the non-stationary signal enhancement method for wind power prediction, table 2 compares the original diffusion model with the diffusion model using a Recurrent Neural Network (RNN) as a kernel solver, and demonstrates the success of the non-stationary signal enhancement method in solving the over-stationary problem based on error analysis and a stationary-detector (ADF). In addition, in order to intuitively reflect the prediction effect, fig. 4 shows the wind power probability prediction result obtained by using the method of the present invention, and fig. 5 shows the performance of the model of the present invention on a longer prediction time scale. From tables 1, 2, 4 and 5, the invention can accurately realize the wind power probability prediction 10 minutes to 2.5 hours in advance, and meet the real-time scheduling requirement of the power grid.
Table 1 test sample error indicators based on the data set of the us renewable energy laboratory
| PIAW | PICP | WS | |
| 90% confidence | 1.901±1.062 | 83.537±11.937 | 2.246±1.031 |
| 80% confidence | 1.297±0.613 | 78.636±21.476 | 1.892±0.575 |
Table 2 performance gains for non-stationary signal enhancement based on data sets of the us renewable energy laboratory
In summary, the wind power probability prediction method based on the diffusion model and the non-stationary signal enhancement can remarkably improve the wind power probability prediction precision 10 minutes to 2.5 hours in advance. The method can be applied to a new energy centralized prediction system or a wind power plant on-site prediction system, and can guide related units to adjust a power generation plan according to a photovoltaic power probability prediction result, reduce standby capacity, promote on-site consumption of new energy and meet the requirements of power grid monitoring and economic safety regulation.
Based on the same technical concept as the method embodiment, the invention also provides a wind power probability prediction system based on a diffusion model and non-stationary signal enhancement, which comprises the following steps:
the historical data simulation module is used for modeling uncertainty of historical wind power by using a diffusion model, the diffusion model comprises a diffusion process and a denoising process, the diffusion process is used for obtaining a historical wind power probabilistic modeling result by continuously adding Gaussian noise to the original wind power until the characteristics of the original wind power are completely destroyed, and the denoising process is used for fitting Gaussian distribution parameters of each step of diffusion process by using a kernel solver;
the model training module is used for grasping the non-stationary fluctuation trend of the future wind power by utilizing the key parameters of the historical wind power stability and the future wind speed forecast information, constructing a non-stationary signal enhancement method, taking the non-stationary signal enhancement method as a kernel solver in the denoising process of the diffusion model, and rolling and training the diffusion model based on a shared parameter mechanism;
the model prediction module is used for obtaining a wind power multi-step probability prediction result by utilizing a double-layer sampling method based on a diffusion model which is obtained through training and takes non-stationary signal enhancement as a kernel solver and by utilizing edge diffusion step length and time sequence reasoning.
It should be understood that, the wind power probability prediction system based on the diffusion model and the non-stationary signal enhancement in the embodiment of the present invention may implement all the technical solutions in the above method embodiments, and the functions of each functional module may be specifically implemented according to the methods in the above method embodiments, and the specific implementation process may refer to the relevant descriptions in the above embodiments, which are not repeated herein.
The present invention also provides a computer device comprising: one or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, which when executed by the processors implement the steps of the wind power probability prediction method based on a diffusion model and non-stationary signal enhancement as described above.
The present invention also provides a computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the steps of a wind power probability prediction method based on a diffusion model and non-stationary signal enhancement as described above.
It will be appreciated by those skilled in the art that embodiments of the invention may be provided as a method, apparatus (system), computer device, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may 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, and the like) having computer-usable program code embodied therein.
The invention is described with reference to flow charts of methods according to embodiments of the invention. It will be understood that each flow in the flowchart, and combinations of flows in the flowchart, can be implemented by computer program instructions. These computer program instructions may 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, create means for implementing the functions specified in the flowchart flow or flows.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows.
Claims (10)
1. The wind power probability prediction method based on the diffusion model and the non-stationary signal enhancement is characterized by comprising the following steps of:
modeling the uncertainty of the historical wind power by using a diffusion model, wherein the diffusion model comprises a diffusion process and a denoising process, the diffusion process is used for obtaining a historical wind power probabilistic modeling result by continuously adding Gaussian noise to the original wind power until the characteristics of the diffusion process are completely destroyed, and the denoising process is used for fitting Gaussian distribution parameters of each step of diffusion process by using a kernel solver;
grasping the non-stationary fluctuation trend of the wind power without coming by utilizing the key parameters of the historical wind power stability and the future wind speed forecast information, constructing a non-stationary signal enhancement method, taking the non-stationary signal enhancement method as a kernel solver of a diffusion model denoising process, and rolling and training a diffusion model based on a shared parameter mechanism;
based on a diffusion model which is obtained by training and takes non-stationary signal enhancement as a kernel solver, wind power multi-step probability prediction results are obtained by utilizing a double-layer sampling method of edge diffusion step length and time sequence reasoning.
2. The method of claim 1, wherein the diffusion process is expressed as:
wherein q (x n |x n-1 ) Representing the Gaussian distribution, x, that the n-1 process needs to add n Features representing the nth step of the diffusion process;mean value of +.>Standard deviation is beta n A gaussian distribution of I; beta n The adding-to-noise ratio example of the n-th diffusion process is shown, and I represents an identity matrix;
by setting alpha n ＝1-β n Obtaining a direct analytic solution method of the diffusion process from the 0 th step to the n th step:
wherein the method comprises the steps ofThe backward propagation probability of the diffusion model is obtained by a Bayesian method:
wherein,and->The expected and variance, q (x) n |x n-1 )，q(x n ) And q (x) n-1 ) Are calculated by the diffusion process.
3. The method according to claim 2, characterized in that the noise ratio β of the diffusion process n Expressed as:
where s is a positive real number and N represents the step size of the diffusion process.
4. The method according to claim 2, wherein the denoising process of the diffusion model is such that a fitted denoising probability p is required θ (x n-1 |x n ) As close to the backward propagation probability q (x n-1 |x n ,x 0 ) The denoising probability is expressed as variance μ θ Standard deviation of sigma θ Gaussian distribution of (c)Applying KL divergence to build likelihood function to measure learnable denoising probability and actual diffusion probabilityVariability between rates:
wherein C is a constant and is irrelevant to the model parameter theta;representing a variable delimiter operator;
the mean value of the denoising probabilities is:
wherein, E is θ (x n N) is a learning diffusion model solver, and the diffusion model is trained by the solver E θ (x n N) realizing end-to-end training, and obtaining wind power probability prediction results by Gaussian noise sampling after the training is completed.
5. The method of claim 1, wherein the non-stationary signal enhancement method specifically comprises:
characterizing the non-stationary characteristics of wind power fluctuations using the expectations and variances of the time series profiles, generalizing the historical wind power sequence into stationary sequences:
wherein Q' is a smoothed historical wind power sequence challenge vector, mu Q Sum sigma Q Respectively the expected and variance of the historical wind power sequence;
calculating the similarity relation between the stabilized wind power sequences by using an improved multi-head attention mechanism to obtain a calculation method H for the de-stabilized attention mechanism Datt ()：
Wherein f MLP Is a learnable linear layer d k The dimension of Q' is indicated and,and the K 'and the V' are respectively key vectors and value vectors of the smoothed historical wind power sequences, tau and delta represent key parameters for measuring the wind power stability, and the key parameters are obtained by predicting the historical wind power sequences and the weather forecast wind speed data.
6. The method of claim 1, wherein using the non-stationary signal enhancement method as a kernel solver for a diffusion model denoising process, rolling the training diffusion model based on a shared parameter mechanism comprises:
time sequence connection for representing wind power fluctuation by integrating hidden state h t The state h is hidden from the last moment t-1 Historical wind power sequence after denoisingWeather forecast information at time t>The derivation is expressed as:
wherein H is Datt () A non-stationary signal enhancement method is represented,a value of n representing a diffusion step at time t, f FFN,θ Representing a linear forward propagation function, ">A transition variable representing a hidden state update process;
the non-stationary enhancement method is used as a kernel solver of a denoising process of the diffusion model, a minimum negative log likelihood function is constructed by a transfer variable of a hidden state to be used as a loss function to train the diffusion model fused with the non-stationary signal enhancement method, training parameters are shared among the denoising processes during training, and the training process of the diffusion model fused with the non-stationary signal enhancement method is expressed as follows:
where e represents a standard gaussian distribution.
7. The method according to claim 1, wherein the two-layer sampling along diffusion step length, time sequence reasoning method specifically comprises:
the denoising characteristics of the denoising process at the same moment are determined by the final value of the sampling diffusion processObtaining the predicted result of the n-1 th step denoising at the t moment +.>Expressed as:
wherein,representing the model obtained by training;
iterating from the n-1 st step to the 0 th step to obtain a wind power probability prediction result at the moment t
Based on hidden state h t Wind power measurement sequenceWeather forecast information at time t+1>Updating hidden state h of diffusion model obtained by training and fused with non-stationary signal enhancement method t+1 Further realizing rolling solution of the model to obtain wind power multi-step probability prediction result +.>Where T is the time step that needs to be predicted.
8. A wind power probability prediction system based on a diffusion model and non-stationary signal enhancement, comprising:
the historical data simulation module is used for modeling uncertainty of historical wind power by using a diffusion model, the diffusion model comprises a diffusion process and a denoising process, the diffusion process is used for obtaining a historical wind power probabilistic modeling result by continuously adding Gaussian noise to the original wind power until the characteristics of the original wind power are completely destroyed, and the denoising process is used for fitting Gaussian distribution parameters of each step of diffusion process by using a kernel solver;
the model training module is used for grasping the non-stationary fluctuation trend of the future wind power by utilizing the key parameters of the historical wind power stability and the future wind speed forecast information, constructing a non-stationary signal enhancement method, taking the non-stationary signal enhancement method as a kernel solver in the denoising process of the diffusion model, and rolling and training the diffusion model based on a shared parameter mechanism;
the model prediction module is used for obtaining a wind power multi-step probability prediction result by utilizing a double-layer sampling method based on a diffusion model which is obtained through training and takes non-stationary signal enhancement as a kernel solver and by utilizing edge diffusion step length and time sequence reasoning.
9. A computer device, comprising:
one or more processors;
a memory; and
one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, which when executed by the processors implement the steps of the wind power probability prediction method based on a diffusion model and non-stationary signal enhancement as claimed in any one of claims 1-7.
10. A computer readable storage medium, on which a computer program is stored, which computer program, when being executed by a processor, implements the steps of a wind power probability prediction method based on a diffusion model and non-stationary signal enhancement as claimed in any of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311581023.1A CN117556706A (en) | 2023-11-24 | 2023-11-24 | Wind power probability prediction method and system based on diffusion model and non-stationary signal enhancement |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311581023.1A CN117556706A (en) | 2023-11-24 | 2023-11-24 | Wind power probability prediction method and system based on diffusion model and non-stationary signal enhancement |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117556706A true CN117556706A (en) | 2024-02-13 |
Family
ID=89810682
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311581023.1A Pending CN117556706A (en) | 2023-11-24 | 2023-11-24 | Wind power probability prediction method and system based on diffusion model and non-stationary signal enhancement |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117556706A (en) |
-
2023
- 2023-11-24 CN CN202311581023.1A patent/CN117556706A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110298501B (en) | Electrical load prediction method based on long-time and short-time memory neural network | |
| Sun et al. | Using Bayesian deep learning to capture uncertainty for residential net load forecasting | |
| CN111144286A (en) | Urban PM2.5 concentration prediction method fusing EMD and LSTM | |
| CN110942194A (en) | Wind power prediction error interval evaluation method based on TCN | |
| CN112686464A (en) | Short-term wind power prediction method and device | |
| CN111695724B (en) | Wind speed prediction method based on hybrid neural network model | |
| CN111461463A (en) | Short-term load prediction method, system and equipment based on TCN-BP | |
| CN111950805B (en) | Medium-and-long-term power load prediction method and device, computer equipment and storage medium | |
| CN111241755A (en) | Power load prediction method | |
| CN112100911B (en) | Solar radiation prediction method based on depth BILSTM | |
| CN111814956A (en) | Multi-task learning air quality prediction method based on multi-dimensional secondary feature extraction | |
| CN113240153A (en) | Photovoltaic power generation data prediction method and device, computing equipment and storage medium | |
| Li et al. | GMM-HMM-based medium-and long-term multi-wind farm correlated power output time series generation method | |
| CN115034485A (en) | Wind power interval prediction method and device based on data space | |
| CN115796004A (en) | Photovoltaic power station ultra-short term power intelligent prediction method based on SLSTM and MLSTNet models | |
| CN113111592A (en) | Short-term wind power prediction method based on EMD-LSTM | |
| CN117060401A (en) | New energy power prediction method, device, equipment and computer readable storage medium | |
| CN117556706A (en) | Wind power probability prediction method and system based on diffusion model and non-stationary signal enhancement | |
| CN111709585A (en) | Air conditioner load prediction method and device and storage medium | |
| CN116316600A (en) | Wind power day-ahead prediction self-adaptive filtering method based on seasonal typing | |
| CN112446550B (en) | Short-term building load probability density prediction method | |
| CN115936062A (en) | Photovoltaic power day-ahead interval prediction method based on scene generation and deep learning | |
| CN111027680B (en) | Monitoring quantity uncertainty prediction method and system based on variational self-encoder | |
| CN114462684A (en) | Wind speed multipoint synchronous prediction method for coupling numerical weather forecast and measured data | |
| CN112766535A (en) | Building load prediction method and system considering load curve characteristics |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |