CN115423175A - Wind power prediction error decomposition method - Google Patents
Wind power prediction error decomposition method Download PDFInfo
- Publication number
- CN115423175A CN115423175A CN202211045066.3A CN202211045066A CN115423175A CN 115423175 A CN115423175 A CN 115423175A CN 202211045066 A CN202211045066 A CN 202211045066A CN 115423175 A CN115423175 A CN 115423175A
- Authority
- CN
- China
- Prior art keywords
- error
- prediction
- longicorn
- noise
- power
- 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
- 238000000034 method Methods 0.000 title claims abstract description 35
- 238000000354 decomposition reaction Methods 0.000 title claims abstract description 10
- 238000005457 optimization Methods 0.000 claims abstract description 23
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 21
- 238000001914 filtration Methods 0.000 claims abstract description 14
- 238000004458 analytical method Methods 0.000 claims abstract description 10
- 238000004364 calculation method Methods 0.000 claims abstract description 10
- 238000012545 processing Methods 0.000 claims abstract description 7
- 238000004088 simulation Methods 0.000 claims abstract description 7
- 241001481710 Cerambycidae Species 0.000 claims description 38
- 238000013507 mapping Methods 0.000 claims description 27
- 238000013528 artificial neural network Methods 0.000 claims description 26
- 238000005070 sampling Methods 0.000 claims description 18
- 230000006870 function Effects 0.000 claims description 15
- 230000008569 process Effects 0.000 claims description 14
- 238000005259 measurement Methods 0.000 claims description 12
- 239000011159 matrix material Substances 0.000 claims description 9
- 230000007704 transition Effects 0.000 claims description 9
- 238000011156 evaluation Methods 0.000 claims description 8
- 230000007246 mechanism Effects 0.000 claims description 8
- 230000035945 sensitivity Effects 0.000 claims description 6
- 238000012360 testing method Methods 0.000 claims description 6
- 238000012549 training Methods 0.000 claims description 6
- 238000011160 research Methods 0.000 claims description 4
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 claims description 3
- 241001609692 Anoplophora chinensis Species 0.000 claims description 3
- 230000009471 action Effects 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 238000010845 search algorithm Methods 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 239000000470 constituent Substances 0.000 claims description 2
- 229910052731 fluorine Inorganic materials 0.000 claims description 2
- 125000001153 fluoro group Chemical group F* 0.000 claims description 2
- 230000014509 gene expression Effects 0.000 claims description 2
- 230000001537 neural effect Effects 0.000 claims 1
- 238000010586 diagram Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 2
- 230000002431 foraging effect Effects 0.000 description 2
- 241000254173 Coleoptera Species 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000010835 comparative analysis Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000010248 power generation Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/003—Load forecast, e.g. methods or systems for forecasting future load demand
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2113/00—Details relating to the application field
- G06F2113/04—Power grid distribution networks
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- Strategic Management (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Human Resources & Organizations (AREA)
- Software Systems (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Primary Health Care (AREA)
- Public Health (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Power Engineering (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Water Supply & Treatment (AREA)
- Mathematical Physics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to the technical field of wind power, in particular to a method for decomposing a wind power prediction error, which is characterized by comprising the following steps of: the method comprises the steps of wind power prediction error decomposition, data processing based on Kalman filtering, model parameter optimization based on BAS optimization algorithm, simulation calculation and the like. Compared with the existing error analysis method only considering the error distribution characteristics, the method can decompose the predicted errors of all links and analyze how the predicted errors of all links are improved, and has the advantages of clear physical significance, strong interpretability and the like.
Description
Technical Field
The invention relates to the technical field of wind power prediction, in particular to a method for decomposing a wind power prediction error.
Background
Wind energy is an advanced green energy source, and after more than a century of technological innovation, wind power generation reaches high levels in the aspects of efficiency, cost and the like. However, due to the nature of wind power, in order to ensure the power balance of a power grid, effective planning and scheduling must be performed on the intake and output of electric energy, and while normal power supply is ensured, the operation cost needs to be considered, so that effective prediction and analysis must be performed on the wind power, which becomes the most important ring in the research of wind power grid connection.
With the application of advanced artificial intelligence algorithm in the field of wind power prediction and the proposal of more perfect prediction strategy, the current wind power prediction precision is greatly improved. However, the prediction error exists objectively all the time, and the current related field is lack of research on wind power prediction error decomposition.
In order to analyze the source and the composition of the prediction error, the actual prediction error is decomposed into a predictor error, an input disturbance error and an observation error, so that a Kalman filtering algorithm and a longicorn searching algorithm are introduced according to the adaptation condition of the model to different types of errors, and the input and parameter setting of the prediction model are optimally decided.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, and provides a wind power prediction error decomposition method with clear physical significance by adopting a technical means.
The technical scheme adopted for realizing the purpose of the invention is as follows: a wind power prediction error decomposition method is characterized by comprising the following steps: it comprises the following steps:
1) Wind power prediction error decomposition:
setting a wind power sequence as x (t), and performing denoising processing on the sequence to obtain d (t), wherein the power sequence is expressed as:
x(t)＝d(t)+φ (1)
in the formula, d (t) represents an intrinsic component of a wind power sequence, phi represents a noise error caused by observation, and t represents the time;
the h-step prediction relationship considering the power sequence is as follows: d (t) → d (t + h), x (t) → x (t + h), the predictor f () is constructed such that:
d(t+h)＝f(d(t)) (2)
x(t+h)＝f(t) (3)
in the formula, f () represents the truest eigen mapping relation between an ideal power sequence reflecting historical time intervals and actual power of a prediction time interval, and a predictor f obtained by fitting a neural network is adopted w () (ii) de-approximating f () such that:
d w (t+h)＝f w (w,d(t)) (4)
x w (t+h)＝f w (w,x(t)) (5)
in the formula (f) w A neural network predictor; w is the adjustable connection weight of the neural network; d is a radical of w Predicting a predicted value of the denoised power sequence through a neural network; x is the number of w Predicting a predicted value of the power sequence which is not denoised after prediction by a neural network;
when the power sequence contains noise, the power sequence shows a rule completely inconsistent with the actual characteristic in certain time intervals, so that the construction process and the actual prediction process of the predictor are influenced by the noise, and the error of the actual prediction process is difficult to analyze;
in order to accurately analyze the source and the constituent mechanism of the prediction error, the prediction error is described from the following three directions:
(1) defining the difference value of intrinsic mapping of the noise-containing power sequence and the de-noising power sequence as an observation error E l ：
E l ＝x(t+h)-d(t+h)＝f(x(t))-f(d(t)) (6)
(2) Defining the difference value of the intrinsic mapping and the neural network fitting mapping of the denoising power sequence as a predictor error E b ：
E b ＝d(t+h)-d w (t+h)＝f(d(t))-f w (w,d(t)) (7)
(3) Defining the difference value of the noisy power sequence and the de-noised power sequence through the fitting mapping of the neural network as an input disturbanceDynamic error E n ：
E n ＝d w (t+h)-x w (t+h)＝f w (w,d(t))-f w (w,x(t)) (8)
Therefore, the actual prediction error E of the noisy power sequence can be obtained by the predictor error, the input disturbance error and the observation error p ：
E p ＝E b +E n +E l ＝f(x(t))-f w (w,x(t)) (9)
When the time series contains noise, the actual prediction error E p Upper and lower bounds of (E) and b 、E n and E l Are all related;
the prediction target is to make the mapping fitted by the predictor approximate to the eigen mapping as much as possible, and the neural network usually adopts a gradient descent training method, so that stable data can obtain larger weight in model training, and for the stable data, the mapping relation f obtained by fitting the neural network is used w Can more approximate the actual physical mapping f, and shows that the input transformation can not only be used for the final actual error E p Has an influence on the predictor error E b Producing an influence; input disturbance error E n The size of (2) essentially represents how the system initial state sensitivity acts on the actual prediction error; and an observation error E l Depending on the sampling accuracy of the data sampling equipment, the data sampling equipment cannot be controlled through a prediction link;
2) Processing data based on Kalman filtering:
the estimation of the Kalman filtering algorithm is based on a least square method, firstly a priori estimation value is obtained, then the state of the current moment is optimally estimated under the action of Kalman gain by combining the measurement value of the current moment, the minimum mean square error is met,
the state equation is to estimate the current system state by using the linear random difference equation and the last system state, as shown in formula (10):
x k ＝Ax k-1 +Bu k-1 +ω (10)
in the formula, x k Representing the state to be estimated, x k-1 Representing the state at the previous moment, A represents the transition matrix from the previous state to the current state; u. of k-1 A control input representing a state at a previous time; b represents a transition matrix of control inputs to the current state; ω represents process noise;
ignoring the control input u at the previous moment in use k-1 And then:
x k ＝Ax k-1 +ω (11)
for the measurement equation of the current state again, then:
z k ＝Hx k +v (12)
in the formula, z k Representing the measured value corresponding to the current state; h represents a transition matrix from the current state to the measurement; v represents measurement noise;
to analyze input disturbance error E n For the predicted total error E p The influence of the noise-containing time sequence is analyzed by filtering the original noise-containing time sequence by adopting a Kalman filtering algorithm, removing noise errors in the original noise-containing time sequence and then performing prediction comparative analysis through a prediction model, so that the influence of the noise-containing condition of the time sequence input into the prediction model on the prediction errors is analyzed;
3) Model parameter optimization based on BAS optimization algorithm:
solving the multi-dimensional model optimization problem by adopting a longicorn stigma search algorithm, wherein a longicorn individual is represented by 3 points of a mass center, a left stigma and a right stigma, and the specific optimization process comprises the following steps:
(1) creating a random vector of the orientation of the longicorn stigma and carrying out normalization treatment:
in the formula, rands is a random function; c represents a spatial dimension;
(2) creating space coordinates of the longicorn left and right whiskers:
in the formula, x rt Representing the position coordinates of the right tassel of the longicorn at the t-th iteration; x is the number of lt Representing the position coordinates of the longicorn left hair at the t-th iteration;
representing the centroid coordinates of the longicorn at the tth iteration; d 0 The distance between the left and the right whiskers is represented;
(3) determining the odor intensity of the anoplophora chinensis, namely f (x), according to a fitness function rt ) And f (x) lt ) The f function is a fitness function;
(4) iteratively updating the position of the longicorn:
in the formula, delta t Represents the step size at the t-th iteration; f (x) rt ) And f (x) lt ) Respectively representing the odor intensity of the long horns and the small horns of the longicorn in the s iteration; sign function; if the fitness of the right whisker is larger than that of the left whisker, sign is 1, and the longicorn turns to the right whisker direction by a step length delta t Moving, otherwise, moving towards the left beard direction;
to analyze predictor error E b For the predicted total error E p Optimizing the internal parameters of the prediction model by using a BAS (basic-energy analysis) optimization algorithm, and comparing the optimized internal parameters with the unoptimized model, thereby analyzing the influence of the optimization of the internal parameters of the predictor on the prediction error;
4) Simulation calculation:
in order to research the sensitivity of actual prediction errors of different prediction models to input disturbance errors, predictor errors and observation errors, a certain wind power plant data with a sampling interval of 15min is adopted for example analysis, an Extreme Learning Machine (ELM) is used for analyzing an error composition mechanism,
(1) evaluation index
Prediction standardRate of determination r 1 The calculation method is as formula (16):
in the formula, N represents a sampling point in the test set, i represents a current sampling point in the test set, cap represents wind power installed capacity, and P m (i) Representing the actual power, P, of the sample point i p (i) Representing the predicted power of sample point i;
predicted yield r 2 Comprises the following steps:
wherein B (i) represents a flag indicating whether the sample i prediction is acceptable or not, if so
The sample prediction is passed, B (i) =1, if
The sample prediction is not qualified, B (i) =0;
mean absolute error r 3 ：
Inputting simulation input quantity, and carrying out error calculation on the predicted power calculated by the ELM model and the measured power through error evaluation standard formulas (16), (17) and (18).
According to the method for decomposing the wind power prediction error, the day-ahead prediction error of the wind power is decomposed into a predictor error, and a disturbance error and an observation error are input. And (3) respectively researching input disturbance errors and predictor errors of different models through a Kalman filtering algorithm and a longicorn parameter optimization algorithm, and analyzing the influence of different decisions on the prediction effect of the ELM model. The prediction method is scientific and reasonable, the prediction process is simple, the prediction precision is high, the physical significance is clear, the prediction result is effective, and the practicability is high.
Drawings
FIG. 1 is a prediction error composition mechanism diagram;
FIG. 2 is a block diagram of a Kalman filter;
fig. 3BAS optimization algorithm flow chart.
Detailed Description
The wind power prediction error decomposition method of the present invention is further described below with reference to the accompanying drawings and specific embodiments. With reference to fig. 1 to 3, the method for decomposing the wind power prediction error of the present invention includes the following steps:
1) Decomposing wind power prediction errors:
a wind power prediction error mechanism diagram is shown in fig. 1, a wind power time sequence is set as x (t), denoising processing is performed on the sequence to obtain d (t), and then the time sequence is represented as:
x(t)＝d(t)+φ (1)
in the formula, d (t) represents an intrinsic component of the wind power sequence, phi represents a noise error caused by observation, and t represents time.
The h-step prediction relationship considering the time series is as follows: d (t) → d (t + h), x (t) → x (t + h), it is desirable to construct the predictor f () so that:
d(t+h)＝f(d(t)) (2)
x(t+h)＝f(t) (3)
in the formula, f () represents the most true eigen mapping relationship between the ideal power sequence reflecting the historical period and the actual power of the prediction period, which cannot be obtained exactly in any way.
But a derived predictor f that can be fitted by a neural network w () To approximate f () as closely as possible so that:
d w (t+h)＝f w (w,d(t)) (4)
x w (t+h)＝f w (w,x(t)) (5)
in the formula (f) w A neural network predictor; w is the adjustable connection weight of the neural network; d w Predicting a predicted value of the denoised power sequence through a neural network; x is the number of w And predicting the predicted value of the power sequence which is not denoised through a neural network.
When the time series itself contains noise, the series shows a rule completely different from the actual characteristics in some time periods, and therefore both the predictor construction process and the actual prediction process are affected by the noise, so that the error of the actual prediction process becomes difficult to analyze.
In order to accurately analyze the source and the mechanism of the prediction error, the prediction error is described from the following three directions:
(1) defining the difference value of intrinsic mapping of the noise-containing power sequence and the de-noising power sequence as an observation error E l ：
E l ＝x(t+h)-d(t+h)＝f(x(t))-f(d(t)) (6)
(2) Defining the difference value of the intrinsic mapping and the neural network fitting mapping of the denoising power sequence as a predictor error E b ：
E b ＝d(t+h)-d w (t+h)＝f(d(t))-f w (w,d(t)) (7)
(3) Defining the difference value of the noisy power sequence and the de-noising power sequence through neural network fitting mapping as an input disturbance error E n ：
E n ＝d w (t+h)-x w (t+h)＝f w (w,d(t))-f w (w,x(t)) (8)
Therefore, the actual prediction error E of the noise-containing power sequence can be obtained through the predictor error, the input disturbance error and the observation error p ：
E p ＝E b +E n +E l ＝f(x(t))-f w (w,x(t)) (9)
At this time the actual prediction error E p From E b 、E n And E l When the time series contains noise, the actual prediction error E p Of (2)Lower bound and E b 、E n And E l Are all relevant.
The prediction target is to make the mapping fitted by the predictor approximate to the eigen mapping as much as possible, and the neural network usually adopts a gradient descent training method, so that stable data can obtain larger weight in model training, and for the stable data, the mapping relation f obtained by fitting the neural network is used w Can more approximate the actual physical mapping f, and shows that the input transformation can not only be used for the final actual error E p Will have an impact on the predictor error E b An influence is generated; input disturbance error E n The size of (A) essentially represents how the sensitivity of the initial state of the system acts on the actual prediction error; and an observation error E l Depending on the sampling accuracy of the data sampling device, it cannot be controlled by the prediction stage.
2) Processing data based on Kalman filtering:
the estimation of the Kalman filtering algorithm is based on the least square method, firstly, a priori estimation value is obtained, then, the state of the current moment is optimally estimated under the action of Kalman gain by combining the measurement value of the current moment, the minimum mean square error is met, and the structure diagram of a filter is shown in FIG. 2. The state equation is to estimate the current system state by using the linear random difference equation and the last system state, as shown in formula (10):
x k ＝Ax k-1 +Bu k-1 +ω (10)
in the formula, x k Representing the state to be estimated, x k-1 Representing the state at the last moment, wherein A represents a transition matrix from the last state to the current state; u. of k-1 A control input representing a state at a previous time; b represents a transition matrix of control inputs to the current state; ω represents process noise.
In use, the control input u at the previous moment is generally ignored k-1 And then:
x k ＝Ax k-1 +ω (11)
plus the measurement equation for the current state, then:
z k ＝Hx k +v (12)
in the formula, z k Representing the measured value corresponding to the current state; h represents a transition matrix from the current state to the measurement; v represents measurement noise.
To analyze input disturbance error E n For the predicted total error E p The influence of the noise-containing time sequence input into the prediction model on the prediction error is analyzed by adopting a Kalman filtering algorithm to filter the original noise-containing time sequence, removing the noise error in the original noise-containing time sequence and then carrying out prediction contrast analysis through the prediction model.
3) Model parameter optimization based on BAS optimization algorithm:
a longicorn stigma search algorithm (BAS) algorithm is a new technology which is proposed in 2017 and is based on the foraging principle of longicorn and suitable for multi-objective function optimization, and the biological principle of the technology is as follows: when foraging, longhorn beetles do not know where food is, but seek food according to the intensity of the food odor. The longicorn has two long antennae, if the left antenna receives stronger smell than the right antenna, the longicorn will fly to the left, otherwise fly to the right. According to the simple principle, the longicorn can effectively find the food. When solving the multi-dimensional model optimization problem, the longicorn individual can be represented by 3 points of the mass center, the left whisker and the right whisker. The algorithm execution flow chart is shown in fig. 3, and the specific optimization flow is as follows:
(1) creating a random vector of the orientation of the longicorn stigma and carrying out normalization treatment:
wherein rands is a random function; c represents the spatial dimension.
(2) Creating space coordinates of the longicorn left and right whiskers:
in the formula, x rt Representing the position coordinates of the right tassel of the longicorn at the t-th iteration; x is a radical of a fluorine atom lt Representing the position coordinates of the longicorn left hair at the t-th iteration;
representing the centroid coordinates of the longicorn at the tth iteration; d 0 Indicating the distance between the two whiskers.
(3) Determining the odor intensity of the anoplophora chinensis, namely f (x), according to a fitness function rt ) And f (x) lt ) The f-function is a fitness function.
(3) Iteratively updating the position of the longicorn:
in the formula, delta t Represents the step size at the t-th iteration; f (x) rt ) And f (x) lt ) Respectively representing the odor intensity of the longicorn beard and the odor intensity of the longicorn beard in the s-th iteration; sign function; if the fitness of the right palpus is larger than that of the left palpus, sign is 1, and the longicorn turns to the right palpus direction by the step length delta t And otherwise, moving towards the left direction.
To analyze predictor error E b For the predicted total error E p The method adopts the BAS optimization algorithm to optimize the internal parameters of the prediction model, and compares the internal parameters with the model which is not optimized, thereby analyzing the influence of the optimization of the internal parameters of the predictor on the prediction error.
4) Simulation calculation:
in order to explore the sensitivity of actual prediction errors of different prediction models to input disturbance errors, predictor errors and observation errors, a certain wind power plant data with a sampling interval of 15min is adopted for example analysis, and an error composition mechanism is analyzed through an Extreme Learning Machine (ELM).
(1) Evaluation index
Prediction accuracy rate r 1 The calculation method is formula (16):
in the formula, N represents a sampling point in the test set, i represents a current sampling point in the test set, cap represents wind power installed capacity, and P m (i) Representing the actual power, P, of the sample point i p (i) Representing the predicted power at sample point i.
Predicted yield r 2 ：
Wherein B (i) represents a flag indicating whether the sample i is qualified or not, if so
The sample prediction is passed, B (i) =1, if
The sample prediction is not good, B (i) =0.
Mean absolute error r 3 ：
Inputting simulation input quantity, and carrying out error calculation on the predicted power calculated by the ELM model and the measured power through error evaluation standard expressions (16), (17) and (18).
Detailed description of the invention
The method takes the measured data and the NWP data of a certain wind power plant as an example for analysis, and the sampling interval is 15min. The installed capacity of the power station is 250MW; the evaluation indices of the predicted results are shown in tables 1 and 2.
TABLE 1 ELM model prediction result evaluation table under different decisions
TABLE 2ELM correlation degree between various errors and total error
The description of the present invention is not intended to be exhaustive or to limit the scope of the claims, and those skilled in the art, on the basis of the teachings contained in this example, will be able to conceive of other substantially equivalent alternatives, all of which are within the scope of this invention.
Claims (1)
1. A wind power prediction error decomposition method is characterized by comprising the following steps: it comprises the following steps:
1) Wind power prediction error decomposition:
setting a wind power sequence as x (t), and performing denoising processing on the sequence to obtain d (t), wherein the power sequence is expressed as:
x(t)＝d(t)+φ (1)
in the formula, d (t) represents an intrinsic component of a wind power sequence, phi represents a noise error caused by observation, and t represents the time;
the h-step prediction relationship considering the power sequence is as follows: d (t) → d (t + h), x (t) → x (t + h), and the predictor f () is constructed such that:
d(t+h)＝f(d(t)) (2)
x(t+h)＝f(t) (3)
in the formula, f () represents the truest eigen mapping relation between an ideal power sequence reflecting historical time intervals and actual power of a prediction time interval, and a predictor f obtained by fitting a neural network is adopted w () (iv) de-approximating f () such that:
d w (t+h)＝f w (w,d(t)) (4)
x w (t+h)＝f w (w,x(t)) (5)
in the formula (f) w Predicting for a neural networkA device; w is the adjustable connection weight of the neural network; d w Predicting a predicted value of the denoised power sequence through a neural network; x is a radical of a fluorine atom w Predicting a predicted value of the power sequence which is not denoised after prediction by a neural network;
when the power sequence contains noise, the power sequence shows a rule completely inconsistent with the actual characteristic in certain time intervals, so that the construction process and the actual prediction process of the predictor are influenced by the noise, and the error of the actual prediction process is difficult to analyze;
in order to accurately analyze the source and the constituent mechanism of the prediction error, the prediction error is described from the following three directions:
(1) defining the difference value of the intrinsic mapping of the noise-containing power sequence and the de-noising power sequence as an observation error E l ：
E l ＝x(t+h)-d(t+h)＝f(x(t))-f(d(t)) (6)
(2) Defining the difference value of the intrinsic mapping and the neural network fitting mapping of the denoising power sequence as a predictor error E b ：
E b ＝d(t+h)-d w (t+h)＝f(d(t))-f w (w,d(t)) (7)
(3) Defining the difference value of the noisy power sequence and the de-noising power sequence through neural network fitting mapping as an input disturbance error E n ：
E n ＝d w (t+h)-x w (t+h)＝f w (w,d(t))-f w (w,x(t)) (8)
Therefore, the actual prediction error E of the noisy power sequence can be obtained by the predictor error, the input disturbance error and the observation error p ：
E p ＝E b +E n +E l ＝f(x(t))-f w (w,x(t)) (9)
When the time series contains noise, the actual prediction error E p Upper and lower bounds of (1) and E b 、E n And E l Are all related;
wherein the goal of the prediction is to make the mapping fitted by the predictor as much as possibleThe eigen mapping is approximated, and the neural network usually adopts a gradient descent training method, so that the stationary data can obtain larger weight in the model training, and therefore, for the stationary data, the mapping relation f obtained by fitting the neural network w Can more approximate the actual physical mapping f, and shows that the input transformation can not only be used for the final actual error E p Has an influence on the predictor error E b Producing an influence; input disturbance error E n The size of (A) essentially represents how the sensitivity of the initial state of the system acts on the actual prediction error; and an observation error E l Depending on the sampling accuracy of the data sampling equipment, the data sampling equipment cannot be controlled through a prediction link;
2) Data processing based on Kalman filtering:
the estimation of the Kalman filtering algorithm is based on a least square method, firstly a priori estimation value is obtained, then the state of the current moment is optimally estimated under the action of Kalman gain by combining the measurement value of the current moment, the minimum mean square error is met,
the state equation is to estimate the current system state by using the linear random difference equation and the last system state, as shown in formula (10):
x k ＝Ax k-1 +Bu k-1 +ω (10)
in the formula, x k Representing the state to be estimated, x k-1 Representing the state at the previous moment, A represents the transition matrix from the previous state to the current state; u. of k-1 A control input representing a state at a previous time; b represents a transition matrix of control inputs to the current state; ω represents process noise;
ignoring the control input u at the previous moment in use k-1 Then:
x k ＝Ax k-1 +ω (11)
for the measurement equation of the current state again, then:
z k ＝Hx k +v (12)
in the formula, z k A corresponding measured value representing the current state(ii) a H represents a transition matrix from the current state to the measurement; v represents measurement noise;
to analyze input disturbance error E n For the predicted total error E p The influence of the noise-containing time sequence is analyzed by filtering the original noise-containing time sequence by adopting a Kalman filtering algorithm, removing the noise error in the original noise-containing time sequence and then performing prediction contrast analysis through a prediction model, so that the influence of the noise-containing condition of the time sequence input into the prediction model on the prediction error is analyzed;
3) Model parameter optimization based on BAS optimization algorithm:
solving the multi-dimensional model optimization problem by adopting a longicorn stigma search algorithm, wherein a longicorn individual is represented by 3 points of a mass center, a left stigma and a right stigma, and the specific optimization flow is as follows:
(1) creating a random vector of the orientation of the longicorn stigma and carrying out normalization treatment:
in the formula, rands is a random function; c represents a spatial dimension;
(2) creating space coordinates of the longicorn left and right whiskers:
in the formula, x rt Representing the position coordinates of the right tassel of the longicorn at the t-th iteration; x is the number of lt Representing the position coordinates of the left tassel of the longicorn at the t-th iteration;
representing the centroid coordinates of the longicorn at the tth iteration; d 0 Representing the distance between the left and right whiskers;
(3) determining the odor intensity of the anoplophora chinensis, namely f (x), according to a fitness function rt ) And f (x) lt ) The f function is a fitness function;
(4) iteratively updating the position of the longicorn:
in the formula, delta t Represents the step size at the t-th iteration; f (x) rt ) And f (x) lt ) Respectively representing the odor intensity of the longicorn beard and the odor intensity of the longicorn beard in the s-th iteration; sign function; if the fitness of the right palpus is larger than that of the left palpus, sign is 1, and the longicorn turns to the right palpus direction by the step length delta t Moving, otherwise, moving towards the left beard direction;
to analyze predictor error E b For the predicted total error E p Optimizing the internal parameters of the prediction model by using a BAS (basic-energy analysis) optimization algorithm, and comparing the optimized internal parameters with the unoptimized model, thereby analyzing the influence of the optimization of the internal parameters of the predictor on the prediction error;
4) Simulation calculation:
in order to research the sensitivity of actual prediction errors of different prediction models to input disturbance errors, predictor errors and observation errors, a certain wind power plant data with a sampling interval of 15min is adopted for example analysis, an Extreme Learning Machine (ELM) is used for analyzing an error composition mechanism,
(1) evaluation index
Prediction accuracy rate r 1 The calculation method is as formula (16):
in the formula, N represents a sampling point in the test set, i represents a current sampling point in the test set, cap represents wind power installed capacity, and P m (i) Representing the actual power, P, of sample point i p (i) Representing the predicted power of sample point i;
predicted yield r 2 Comprises the following steps:
wherein B (i) represents a flag indicating whether the sample i is qualified or not, if so
The sample prediction is passed, B (i) =1, if
The sample prediction is not qualified, B (i) =0;
mean absolute error r 3 ：
Inputting simulation input quantity, and carrying out error calculation on the predicted power calculated by the ELM model and the measured power through error evaluation standard expressions (16), (17) and (18).
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211045066.3A CN115423175A (en) | 2022-08-30 | 2022-08-30 | Wind power prediction error decomposition method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211045066.3A CN115423175A (en) | 2022-08-30 | 2022-08-30 | Wind power prediction error decomposition method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115423175A true CN115423175A (en) | 2022-12-02 |
Family
ID=84200702
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211045066.3A Pending CN115423175A (en) | 2022-08-30 | 2022-08-30 | Wind power prediction error decomposition method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115423175A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116047423A (en) * | 2022-12-13 | 2023-05-02 | 扬州宇安电子科技有限公司 | Interference resource allocation method and allocation system thereof |
| CN118013235A (en) * | 2024-04-08 | 2024-05-10 | 西安热工研究院有限公司 | Energy storage auxiliary black start wind speed prediction method based on cyclic integrated error compensation |
-
2022
- 2022-08-30 CN CN202211045066.3A patent/CN115423175A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116047423A (en) * | 2022-12-13 | 2023-05-02 | 扬州宇安电子科技有限公司 | Interference resource allocation method and allocation system thereof |
| CN116047423B (en) * | 2022-12-13 | 2024-04-26 | 扬州宇安电子科技股份有限公司 | Interference resource allocation method and allocation system thereof |
| CN118013235A (en) * | 2024-04-08 | 2024-05-10 | 西安热工研究院有限公司 | Energy storage auxiliary black start wind speed prediction method based on cyclic integrated error compensation |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Thomas et al. | Likelihood-free inference by ratio estimation | |
| CN109886498B (en) | EMD-GRU short-term power load prediction method based on feature selection | |
| CN115423175A (en) | Wind power prediction error decomposition method | |
| CN110414788B (en) | Electric energy quality prediction method based on similar days and improved LSTM | |
| CN109711609B (en) | Photovoltaic power station output power prediction method based on wavelet transformation and extreme learning machine | |
| CN110163429B (en) | Short-term load prediction method based on similarity day optimization screening | |
| CN111931983B (en) | Precipitation prediction method and system | |
| CN109523155B (en) | Power grid risk assessment method of Monte Carlo and least square support vector machine | |
| CN111127246A (en) | Intelligent prediction method for transmission line engineering cost | |
| CN112328588A (en) | Industrial fault diagnosis unbalanced time sequence data expansion method | |
| WO2023103130A1 (en) | Quantum walk-based time-series multiscale analysis method | |
| CN109212631B (en) | Satellite observation data three-dimensional variation assimilation method considering channel correlation | |
| CN110490366A (en) | Runoff forestry method based on variation mode decomposition and iteration decision tree | |
| CN113094860A (en) | Industrial control network flow modeling method based on attention mechanism | |
| CN114970952A (en) | Photovoltaic output short-term prediction method and system considering environmental factors | |
| CN111222689A (en) | LSTM load prediction method, medium, and electronic device based on multi-scale temporal features | |
| CN116384244A (en) | Electromagnetic field prediction method based on physical enhancement neural network | |
| CN116578551A (en) | GRU-GAN-based power grid data restoration method | |
| CN114330815A (en) | Ultra-short-term wind power prediction method and system based on improved GOA (generic object oriented architecture) optimized LSTM (least Square TM) | |
| CN106528679A (en) | Time series analysis method based on multilinear autoregression model | |
| CN111931990B (en) | Power load prediction method based on optimized least square support vector machine | |
| CN115239029B (en) | Wind power prediction method and system considering power time sequence and meteorological dependent characteristics | |
| CN110909492A (en) | Sewage treatment process soft measurement method based on extreme gradient lifting algorithm | |
| CN116384586A (en) | Wind speed prediction method based on VMD-SVR | |
| CN113466681B (en) | Breaker service life prediction method based on small sample learning |
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 |