CN113515889B - Dynamic wind speed prediction model building method - Google Patents

Dynamic wind speed prediction model building method Download PDF

Info

Publication number
CN113515889B
CN113515889B CN202110557310.3A CN202110557310A CN113515889B CN 113515889 B CN113515889 B CN 113515889B CN 202110557310 A CN202110557310 A CN 202110557310A CN 113515889 B CN113515889 B CN 113515889B
Authority
CN
China
Prior art keywords
wind speed
model
prediction
speed prediction
error
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.)
Active
Application number
CN202110557310.3A
Other languages
Chinese (zh)
Other versions
CN113515889A (en
Inventor
李永刚
王月
吴滨源
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
North China Electric Power University
Original Assignee
North China Electric Power University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by North China Electric Power University filed Critical North China Electric Power University
Priority to CN202110557310.3A priority Critical patent/CN113515889B/en
Publication of CN113515889A publication Critical patent/CN113515889A/en
Application granted granted Critical
Publication of CN113515889B publication Critical patent/CN113515889B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

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
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • 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
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • 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
    • 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
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Human Resources & Organizations (AREA)
  • Economics (AREA)
  • General Physics & Mathematics (AREA)
  • Strategic Management (AREA)
  • Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Development Economics (AREA)
  • General Business, Economics & Management (AREA)
  • General Engineering & Computer Science (AREA)
  • Tourism & Hospitality (AREA)
  • Marketing (AREA)
  • Software Systems (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Artificial Intelligence (AREA)
  • Biophysics (AREA)
  • Educational Administration (AREA)
  • Game Theory and Decision Science (AREA)
  • Molecular Biology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computational Linguistics (AREA)
  • Mathematical Physics (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Data Mining & Analysis (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Public Health (AREA)
  • Water Supply & Treatment (AREA)
  • Primary Health Care (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Feedback Control In General (AREA)

Abstract

The invention discloses a method for establishing a dynamic wind speed prediction model, which comprises the following steps: obtaining measured wind speed data of a target area and preprocessing the measured wind speed data; training and predicting the preprocessed actually measured wind speed data by utilizing a plurality of prediction algorithms to obtain a plurality of wind speed prediction models, and forming a Q learning model set; adding a wind speed fluctuation condition and attribute factors in the Q learning model set, selecting an optimal wind speed prediction model of each period through a Q reinforcement learning algorithm to obtain preliminary wind speed prediction data, and calculating a wind speed prediction error; and constructing an error Q learning model library based on the wind speed prediction error, and selecting an optimal wind speed prediction error model from the error Q learning model library through a Q reinforcement learning algorithm to correct the preliminary wind speed prediction value so as to obtain final wind speed prediction data. The method provided by the invention adopts the twice Q reinforcement learning algorithm to construct the dynamic wind speed prediction model, and has the characteristics of strong generalization capability, good robustness and high prediction precision.

Description

Dynamic wind speed prediction model building method
Technical Field
The invention relates to the technical field of wind speed prediction, in particular to a method for establishing a dynamic wind speed prediction model.
Background
In recent years, energy structures are continuously developed in a low-carbon direction, and the permeability of renewable energy power grids represented by wind power is increased year by year. With the large-scale grid connection of renewable energy sources represented by wind power generation, the dispatching economy of a power grid is gradually improved, but the running reliability of a power system is seriously reduced due to the characteristics of fluctuation, indirection, low energy density and the like of wind speed. Therefore, in order to better utilize wind power generation and simultaneously consider the stability of the power system, short-term accurate prediction needs to be performed on the wind speed, but no clear definition is given yet on how to build a prediction model and how to optimize the prediction model to improve the prediction precision and generalization capability.
Disclosure of Invention
The invention aims to provide a method for establishing a dynamic wind speed prediction model, which adopts a twice Q reinforcement learning algorithm to construct the dynamic wind speed prediction model and has the characteristics of strong generalization capability, good robustness and high prediction precision.
In order to achieve the above object, the present invention provides the following solutions:
a method for establishing a dynamic wind speed prediction model comprises the following steps:
s1) obtaining measured wind speed data of a target area, and preprocessing the measured wind speed data;
s2) dividing the preprocessed actually measured wind speed data into a wind speed training set, a wind speed testing set and a wind speed checking set, training the wind speed training set by utilizing a plurality of prediction algorithms, predicting the wind speed testing set to obtain a plurality of wind speed prediction models, and forming a Q learning model set;
s3) adding wind speed fluctuation conditions and attribute factors in the Q learning model set, selecting an optimal wind speed prediction model of each period through a Q reinforcement learning algorithm to obtain preliminary wind speed prediction data, and calculating a wind speed prediction error according to the preliminary wind speed prediction data and corresponding actually measured wind speed data;
s4) constructing an error Q learning model library based on the wind speed prediction error, and selecting an optimal wind speed prediction error model from the error Q learning model library through a Q reinforcement learning algorithm to correct the preliminary wind speed prediction value so as to obtain final wind speed prediction data.
Optionally, the preprocessing of the measured wind speed data in step S1) refers to replacing missing and abnormal values in the measured wind speed data by adopting an adjacent data complementation method.
Optionally, the method further comprises, after step S4):
s5) verifying the validity of the optimal wind speed prediction error model by using the wind speed test set.
Optionally, verifying the validity of the optimal wind speed prediction error model selects a mean square error ε 1 Relative error epsilon 2 Determining a coefficient R 2 The three evaluation indexes evaluate the final wind speed prediction data, and the calculation formulas are respectively as follows:
Figure BDA0003077769640000021
Figure BDA0003077769640000022
/>
Figure BDA0003077769640000023
wherein: x is x t 、y t 、
Figure BDA0003077769640000024
The measured wind speed value, the final predicted wind speed value, the measured wind speed average value and the final predicted wind speed average value at the time t are respectively.
Optionally, the plurality of prediction algorithms adopted in the Q learning model set in step S2) is 5, including a LSTM, XGBoost, SVR, BP neural network and a KRR learning algorithm.
Optionally, the plurality of prediction algorithms adopted in the error Q learning model library in step S4) is 5, including learning algorithms of SVR, BP neural network, GKRR, PKRR and MHKRR.
Optionally, the calculation formula of the wind speed prediction error in step S3) and in step S4) is as follows:
Figure BDA0003077769640000025
wherein:
Figure BDA0003077769640000026
for the wind speed prediction error, x is the measured wind speed value, < >>
Figure BDA0003077769640000027
Is a preliminary wind speed predictor.
Optionally, the calculation formula of the final wind speed prediction data in step S4) is as follows:
Figure BDA0003077769640000028
wherein: y is the final wind speed predictor value,
Figure BDA0003077769640000029
for the preliminary wind speed forecast->
Figure BDA00030777696400000210
Is the corrected wind speed prediction error.
Optionally, the Q reinforcement learning algorithm in step S3) and in step S4) uses a reward function with a mixture of error and model rank, and the calculation formula of the reward function is as follows:
Figure BDA0003077769640000031
wherein: r (s, a) is a reward function; s is a state space, S= { S 1 ,…,s I ,…,s N }，s I N is the number of wind speed prediction models for the current wind speed prediction model; a is an action space, a= { a 1 ,…,a J ,…,a N }，a J An act of switching from the current wind speed predictive model to the next wind speed predictive model for a next predicted time step; RANK (M) I,t ) And RANK (M) I,t+1 ) Wind speed prediction model M at the t time and the t+1 time respectively I Is a ranking of (2); TIME (M) I,t ) And TIME (M) I,t+1 ) Wind speed prediction model M for t time and t+1 time I Is calculated according to the calculation time of (2); α, β are weight coefficients, and α+β=1 is satisfied.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects: according to the method for establishing the dynamic wind speed prediction model, the dynamic wind speed prediction model is established by adopting a twice Q reinforcement learning algorithm, wherein one Q learning agent is responsible for selecting an optimal wind speed prediction model to conduct preliminary wind speed prediction, and the other Q learning agent is responsible for inputting the optimal wind speed prediction model into an error correction part through calculating errors, and selecting an optimal wind speed prediction error model from the optimal wind speed prediction model to obtain an optimal prediction strategy; and Q learning effectively selects the optimal prediction model in both the wind speed prediction part and the error correction part; the error correction reduces the average relative error of prediction by 50%, and the error correction link has effectiveness on the mature prediction model; according to the method, the wind speed prediction model is constructed to predict typical months in different seasons, and the result shows that the method has strong generalization capability, good robustness and high prediction precision, solves the problem of reduced operation reliability of the power system caused by the characteristics of wind speed fluctuation, indirection, low energy density and the like, and can remarkably improve the scheduling economy of a power grid containing renewable energy source grid connection and the operation safety of a wind power plant.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions of the prior art, the drawings that are needed in the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a method for establishing a dynamic wind speed prediction model according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a dynamic wind speed prediction model building method according to an embodiment of the present invention;
FIG. 3 is a convergence diagram of a bonus function according to an embodiment of the invention;
FIG. 4 is a diagram showing a 2019-year wind speed fluctuation of an actual wind field in northeast according to an embodiment of the present invention;
FIG. 5a is a graph of predicted wind speed data for a typical month (3 months) QWSP, LSTM, BP neural network in spring according to an embodiment of the present invention;
FIG. 5b is a graph of predicted wind speed data for a typical month in summer (6 months) QWSP, LSTM, BP neural network according to an embodiment of the present invention;
FIG. 5c is a graph of wind speed prediction data for a autumn representative month (9 months) QWSP, LSTM, BP neural network according to an embodiment of the invention;
FIG. 5d is a graph of predicted wind speed data for a typical month in winter (12 months) QWSP, LSTM, BP neural network according to an embodiment of the present invention;
FIG. 6a is a graph of DPDQ wind speed prediction data for a typical month (3 months) of spring according to an embodiment of the present invention;
FIG. 6b is a graph of DPDQ wind speed forecast data for a typical month in summer (6 months) according to an embodiment of the present invention;
FIG. 6c is a graph of DPDQ wind speed prediction data for a typical month in autumn (9 months) according to an embodiment of the present invention;
FIG. 6d is a graph of DPDQ wind speed forecast data for a typical month in winter (12 months) according to an embodiment of the invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The invention aims to provide a method for establishing a dynamic wind speed prediction model, which adopts a twice Q reinforcement learning algorithm to construct the dynamic wind speed prediction model and has the characteristics of strong generalization capability, good robustness and high prediction precision.
In order that the above-recited objects, features and advantages of the present invention will become more readily apparent, a more particular description of the invention will be rendered by reference to the appended drawings and appended detailed description.
As shown in fig. 1 to 2, the method for establishing a dynamic wind speed prediction model provided by the embodiment of the invention includes the following steps:
s1) obtaining measured wind speed data of a target area, and preprocessing the measured wind speed data;
s2) dividing the preprocessed actually measured wind speed data into a wind speed training set, a wind speed testing set and a wind speed checking set, training the wind speed training set by utilizing a plurality of prediction algorithms, predicting the wind speed testing set to obtain a plurality of wind speed prediction models, and forming a Q learning model set;
s3) adding wind speed fluctuation conditions and attribute factors (wind speed, temperature, humidity, wind direction and turbulence speed) in the Q learning model set, selecting an optimal wind speed prediction model of each period through a Q reinforcement learning algorithm to obtain preliminary wind speed prediction data, and calculating a wind speed prediction error according to the preliminary wind speed prediction data and corresponding measured wind speed data;
s4) constructing an error Q learning model library based on the wind speed prediction error, selecting an optimal wind speed prediction error model from the error Q learning model library through a Q reinforcement learning algorithm to correct the preliminary wind speed prediction value, and correcting the error to obtain final wind speed prediction data.
In the step S2), 5 kinds of prediction algorithms are adopted in the Q learning model set, considering the high variability of the wind speed sequence, deep learning can perform deep feature mining on measured wind speed data, a wind speed transformation trend can be predicted better when wind speed fluctuation is severe, but in detail, a phenomenon of over fitting may exist, SVR and BP neural networks often have higher prediction precision in a period of relatively mild wind speed fluctuation, and larger prediction errors exist when wind speed fluctuation is severe, so that five algorithms, namely a deep learning algorithm LSTM, an integrated learning algorithm XGBoost, a shallow learning algorithm SVR, a BP neural network and a KRR, are selected as basic models in the wind speed prediction model set, so that for different fluctuation situations, Q learning can select a more suitable prediction model in the deep learning algorithm, wherein KRR selects PKRR based on a polynomial kernel function.
The principle of the Q reinforcement learning algorithm is briefly described as follows:
to train the Q-learning representation, a mathematical framework based on reinforcement-learning dynamic model selection is first defined in a markov decision process, typically the Q-learning agent takes sequential operations in a series of states according to a state-effect value matrix (Q-matrix) until a final goal is reached, and a reward update Q-matrix is obtained by evaluating the predictive effect of the current state space, which consists of the current predictive model:
S＝{s 1 ,…,s I ,…,s N }
wherein: s is(s) I Representing a current wind speed prediction model; n is the number of wind speed prediction models. Likewise, the action space a consists of the wind speed prediction model of the next step:
A＝{a 1 ,…,a J ,…,a N }
wherein a is J Representing an action of switching from the current wind speed predictive model to the next wind speed predictive model at the next predicted time step. In order to successfully solve the Markov decision process using Q learning, the most central part is to derive the reward matrix R through the appropriate reward function R (s, a), and embodiments of the present invention define the error and model rank mix reward function as follows:
R t (s I ,a J )＝α[RANK(M I,t )-RANK(M J,t+1 )]+β[TIME(M I,t )-TIME(M J,t+1 )]
wherein: RANK (M) I,t ) And RANK (M) I,t+1 ) Wind speed prediction model M at the t time and the t+1 time respectively I Is a ranking of (2); TIME (M) I,t ) And TIME (M) I,t+1 ) Wind speed prediction model M at the t time and the t+1 time respectively I Is calculated according to the calculation time of (2); α, β are weight coefficients, and α+β=1 is satisfied. Dynamic model selection as Q learning model-freeThe frame is selected, and the model is selected preferentially for prediction, so that when the two models are ranked as 1, the model is 0, and the rewarding and punishment effects are lost. Therefore, the two weighted Q learning frameworks are selected to make the reward function more universal. After defining the state space, action space and reward function, training data set T is trained by using Q learning t Training Q to learn the dynamic prediction model.
The Q learning agent adopting decay t greedy method takes completely random action from the beginning, reduces randomness through decay in the learning process, and is characterized by N e After the secondary training, the Q learning algorithm will eventually converge to an optimal strategy Q, which is used to find the optimal action a during Q learning. The method comprises the following specific steps:
(1) Defining a model step length k, a prediction scale N and a model library size N M Q learning data set T t Dynamic predictive model dataset T c The learning rate kappa of the aggressiveness of learning is controlled, the discount factor gamma of future return is weighed, and the training times N are controlled e Ensure at T c Selecting the best model from the N models;
(2) Initializing Q (s, a), ω=1, starting training; selecting random action a with probability of ω e Otherwise select
Figure BDA0003077769640000061
(3) Calculating and updating a reward matrix R according to a reward function calculation formula;
(4) Update Q (s, a) by:
Figure BDA0003077769640000062
(5) Repeating (2) - (4) k times to find the optimal action each time
Figure BDA0003077769640000063
The calculation formula of the wind speed prediction error in step S3) and in step S4) is as follows:
Figure BDA0003077769640000064
wherein:
Figure BDA0003077769640000065
for the wind speed prediction error, x is the measured wind speed value, < >>
Figure BDA0003077769640000066
Is a preliminary wind speed predictor.
The number of the multiple prediction algorithms adopted in the error Q learning model library in the step S4) is 5; for the selection of an error correction model set (namely an error Q learning model library formed by wind speed prediction error models), because the fluctuation and variability of the prediction errors are not as severe as the actually measured wind speed sequence, more accurate prediction is needed for the error sequence, five models with higher efficiency, namely SVR, BP neural network and GKRR, PKRR, MHKRR, are selected to form the error correction model set, wherein GKRR, PKRR and MHKRR models are KRR models adopting different kernel functions.
The method further comprises, after step S4): s5) verifying the validity of the optimal wind speed prediction error model by using the wind speed test set; verifying the validity of the optimal wind speed prediction error model, and selecting a mean square error epsilon 1 Relative error epsilon 2 Determining a coefficient R 2 Three evaluation indexes evaluate the final wind speed prediction data, wherein epsilon 1 、ε 2 Most desirably 0, R 2 The optimal expectation is 1, and the calculation formulas are respectively as follows:
Figure BDA0003077769640000071
Figure BDA0003077769640000072
Figure BDA0003077769640000073
wherein: x is x t 、y t 、
Figure BDA0003077769640000074
The measured wind speed value, the final predicted wind speed value, the measured wind speed average value and the final predicted wind speed average value at the time t are respectively.
Selecting 2019 wind speed and related attribute data of a certain practical wind field in northeast China to develop research, carrying out short-term wind speed prediction on typical months (3 months, 6 months, 9 months and 12 months) of each quarter of the wind field, preprocessing measured wind speed data, replacing missing and abnormal data values by adopting an adjacent data complementation method, sampling the measured wind speed data at a sampling interval of 10min, taking 20 days of data before each month as a wind speed training set, 21-25 days of data as a wind speed test set and 26-30 days as a wind speed test set (used for checking whether model parameter setting is proper or not)
The model super parameters were set as follows:
the specific parameters of the learning framework Q of the embodiment of the invention are set to be kappa=0.1 and gamma=0.8 so as to ensure the learning speed of dynamic model selection, N e =100, and fully considers future rewards and punishments of the reward function, choosing α=0.9, β=0.1; according to the actual running condition of the wind field, selecting to conduct the day-ahead prediction (k=6, n=144) with the step length of 6, namely adopting the optimal strategy to make model selection decision for the next k steps according to the training result of wind speed training set data, wherein the super-parameter setting of the basic model is shown in table 1.
Table 1 super parameter settings for different algorithms
Figure BDA0003077769640000081
Regarding the setting of the Q learning reward function, the adaptive error function commonly used in artificial intelligence algorithms is currently used, but it is found in the training process that Q learning using the adaptive error function as the reward function fails to converge because the size of the predictive evaluation index not only depends on the predictive model, but also changes over time, and negative returns (due to the decrease in the predictive evaluation index) may still be received by taking the operation of switching from the poor model to the optimal model. Meanwhile, whether one prediction model is mature or not is related to not only prediction precision but also time cost paid by the prediction model, so that another rewarding function is provided for evaluating the model effect, namely model ranking improvement and model prediction time are comprehensively considered. The training results of the two methods are shown in fig. 3, and it can be seen that the reward function is successfully converged, so that the time sequence effect is effectively avoided.
As shown in FIG. 4, the wind speed fluctuation condition of the northeast wind field 2019 in the whole year can be seen that the wind energy density of the wind field is larger, wherein the wind speed fluctuation of the spring and winter seasons is more severe, the wind speed difference at different moments is larger, and the highest wind speed exceeds 25m/s. The wind speed in summer and autumn is mostly lower than 10m/s, the fluctuation is gentle, and the wind energy density is obviously lower than that in spring and winter.
To test and explain the effectiveness of dynamic model selection based on Q learning, an artificial intelligent algorithm of two different prediction principles of a single prediction model LSTM and BP neural network is selected to carry out simulation comparison analysis on a wind speed prediction part (Q learning wind speed prediction, QWSP) in a dynamic prediction (Dynamic prediction based on double Q learning, DPDQ) model based on double Q learning, and day-ahead prediction with a sliding step length of 6 is carried out on the wind speed of a typical month in each quarter, and specific prediction results are shown in fig. 5a to 5 d.
As can be seen from fig. 5a to fig. 5d, the QWSP can perform well under the condition of coping with different wind speed fluctuation in each season, and the overall prediction effect is better than that of a single prediction model; from the detail, the wind speed fluctuation in summer and autumn is more gentle, the wind speed is lower, and the wind speed in spring and winter is relatively higher. As can be seen by expanding the parts 121-126 in FIG. 5a and the parts 115-120 in FIG. 5b, each model can obtain a good prediction effect; in the parts 61-66 in FIG. 5c, the BP neural network and the LSTM cannot cope with all wind speed change conditions due to the fact that dynamic selection is not carried out, and the deviation between the prediction result and the actual value is large; the detail prediction result in the graph (d) is not the same, and as can be seen from the right graph, the prediction result deviation of the QWSP model is larger than that of the BP neural network, which is mainly the prediction deviation caused by the selection error of the Q learning model, and the ranking of the selected model should be relatively later due to the arrangement of the reward function mechanism, so that the reward is negative, and the correction is carried out in the next model selection so as to obtain a better prediction result. When the dynamic selection of Q learning is correct, the prediction precision is generally higher, and the model has better robustness; when the single prediction model is fitted, the situation of larger prediction error occurs, and the model can be corrected in the next period in time due to the mechanism of the reward function. Therefore, the model selection strategy based on Q learning can improve the performance of the wind speed prediction model as a whole. The prediction errors of the wind speed of the typical month in each quarter are shown in table 2.
Table 2 short-term prediction error for different methods
Figure BDA0003077769640000091
As can be seen from the data in Table 2, since the fluctuation of the wind speed in summer is minimal, each model R 2 All reach about 0.9, and the fluctuation of wind speed in winter is large, so that a certain prediction difficulty is brought to the model, and a prediction error epsilon is caused 1 And also increases. The model provided by the embodiment of the invention selects the R of the dynamic model based on reinforcement learning 2 The result is closest to 1, ε in each quarter 1 Error is also the smallest of the three models.
In order to verify the effectiveness of the error correction link in the embodiment of the present invention, the result of predicting the wind speed before each season by using the DPDQ is shown in fig. 6a to 6d, it can be seen that the predicted value of the DPDQ model in each season can reach a better effect, but for certain extreme wind conditions with too large wind speed difference, such as the highest wind speed point in fig. 6c, a certain predicted error still exists, which is unavoidable.
According to the method for establishing the dynamic wind speed prediction model, the dynamic wind speed prediction model is established by adopting a twice Q reinforcement learning algorithm, wherein one Q learning agent is responsible for selecting an optimal wind speed prediction model to conduct preliminary wind speed prediction, and the other Q learning agent is responsible for inputting the optimal wind speed prediction model into an error correction part through calculating errors, and selecting an optimal wind speed prediction error model from the optimal wind speed prediction model to obtain an optimal prediction strategy; and Q learning effectively selects the optimal prediction model in both the wind speed prediction part and the error correction part; the error correction reduces the average relative error of prediction by 50%, and the error correction link has effectiveness on the mature prediction model; according to the method, the wind speed prediction model is constructed to predict typical months in different seasons, and the result shows that the method has strong generalization capability, good robustness and high prediction precision, solves the problem of reduced operation reliability of the power system caused by the characteristics of wind speed fluctuation, indirection, low energy density and the like, and can remarkably improve the scheduling economy of a power grid containing renewable energy source grid connection and the operation safety of a wind power plant.
The principles and embodiments of the present invention have been described herein with reference to specific examples, the description of which is intended only to assist in understanding the methods of the present invention and the core ideas thereof; also, it is within the scope of the present invention to be modified by those of ordinary skill in the art in light of the present teachings. In view of the foregoing, this description should not be construed as limiting the invention.

Claims (9)

1. The method for establishing the dynamic wind speed prediction model is characterized by comprising the following steps of:
s1) obtaining measured wind speed data of a target area, and preprocessing the measured wind speed data;
s2) dividing the preprocessed actually measured wind speed data into a wind speed training set, a wind speed testing set and a wind speed checking set, training the wind speed training set by utilizing a plurality of prediction algorithms, predicting the wind speed testing set to obtain a plurality of wind speed prediction models, and forming a Q learning model set;
s3) adding wind speed fluctuation conditions and attribute factors in the Q learning model set, selecting an optimal wind speed prediction model of each period through a Q reinforcement learning algorithm to obtain preliminary wind speed prediction data, and calculating a wind speed prediction error according to the preliminary wind speed prediction data and corresponding actually measured wind speed data;
s4) constructing an error Q learning model library based on the wind speed prediction error, and selecting an optimal wind speed prediction error model from the error Q learning model library through a Q reinforcement learning algorithm to correct the preliminary wind speed prediction value so as to obtain final wind speed prediction data.
2. The method for building a dynamic wind speed prediction model according to claim 1, wherein the preprocessing of the measured wind speed data in step S1) means to replace missing and abnormal values in the measured wind speed data by an adjacent data complementation method.
3. The method for building a dynamic wind speed prediction model according to claim 1, wherein the method further comprises, after step S4):
s5) verifying the validity of the optimal wind speed prediction error model by using the wind speed test set.
4. A method of building a dynamic wind speed prediction model according to claim 3, wherein verifying the validity of the optimal wind speed prediction error model selects a mean square error epsilon 1 Relative error epsilon 2 Determining a coefficient R 2 The three evaluation indexes evaluate the final wind speed prediction data, and the calculation formulas are respectively as follows:
Figure FDA0003077769630000011
Figure FDA0003077769630000012
Figure FDA0003077769630000013
wherein: x is x t 、y t 、
Figure FDA0003077769630000014
The measured wind speed value, the final predicted wind speed value, the measured wind speed average value and the final predicted wind speed average value at the time t are respectively.
5. The method according to claim 1, wherein the plurality of prediction algorithms adopted in the Q learning model set in step S2) is 5, including learning algorithms of LSTM, XGBoost, SVR, BP neural network and KRR.
6. The method according to claim 1, wherein the plurality of prediction algorithms adopted in the error Q learning model library in step S4) is 5, including learning algorithms of SVR, BP neural network, GKRR, PKRR and MHKRR.
7. The method according to claim 1, wherein the calculation formula of the wind speed prediction error in step S3) is as follows:
Figure FDA0003077769630000021
/>
wherein:
Figure FDA0003077769630000022
for the wind speed prediction error, x is the measured wind speed value, < >>
Figure FDA0003077769630000023
Is a preliminary wind speed predictor.
8. The method according to claim 1, wherein the calculation formula of the final wind speed prediction data in step S4) is as follows:
Figure FDA0003077769630000024
wherein: y is the final wind speed predictor value,
Figure FDA0003077769630000025
for the preliminary wind speed forecast->
Figure FDA0003077769630000026
Is the corrected wind speed prediction error.
9. The method for building a dynamic wind speed prediction model according to claim 1, wherein the Q reinforcement learning algorithm in step S3) and in step S4) adopts a reward function with a mixture of errors and model ranks, and the calculation formula of the reward function is as follows:
Figure FDA0003077769630000027
wherein: r (s, a) is a reward function; s is a state space, S= { S 1 ,…,s I ,·…,s N }，s I N is the number of wind speed prediction models for the current wind speed prediction model; a is an action space, a= { a 1 ,…,a J ,…,a N }，a J An act of switching from the current wind speed predictive model to the next wind speed predictive model for a next predicted time step; RANK (M) I,t ) And RANK (M) I,t+1 ) Wind speed prediction model M at the t time and the t+1 time respectively I Is a ranking of (2); TIME (M) I,t ) And TIME (M) I,t+1 ) Wind speed prediction model M for t time and t+1 time I Is calculated according to the calculation time of (2); α, β are weight coefficients, and α+β=1 is satisfied.
CN202110557310.3A 2021-05-21 2021-05-21 Dynamic wind speed prediction model building method Active CN113515889B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110557310.3A CN113515889B (en) 2021-05-21 2021-05-21 Dynamic wind speed prediction model building method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110557310.3A CN113515889B (en) 2021-05-21 2021-05-21 Dynamic wind speed prediction model building method

Publications (2)

Publication Number Publication Date
CN113515889A CN113515889A (en) 2021-10-19
CN113515889B true CN113515889B (en) 2023-06-13

Family

ID=78064601

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110557310.3A Active CN113515889B (en) 2021-05-21 2021-05-21 Dynamic wind speed prediction model building method

Country Status (1)

Country Link
CN (1) CN113515889B (en)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104899446A (en) * 2015-06-05 2015-09-09 上海大学 Method for simulating fluctuating wind speeds on basis of data drive
CN108229732A (en) * 2017-12-20 2018-06-29 上海电机学院 ExtremeLearningMachine wind speed ultra-short term prediction method based on error correction
CN110474339A (en) * 2019-08-07 2019-11-19 国网福建省电力有限公司 A kind of electric network reactive-load control method based on the prediction of depth generation load
CN111064229A (en) * 2019-12-18 2020-04-24 广东工业大学 Wind-light-gas-storage combined dynamic economic dispatching optimization method based on Q learning
CN111985712A (en) * 2020-08-19 2020-11-24 华北电力大学（保定） Multi-step wind speed combined prediction model building method

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104899446A (en) * 2015-06-05 2015-09-09 上海大学 Method for simulating fluctuating wind speeds on basis of data drive
CN108229732A (en) * 2017-12-20 2018-06-29 上海电机学院 ExtremeLearningMachine wind speed ultra-short term prediction method based on error correction
CN110474339A (en) * 2019-08-07 2019-11-19 国网福建省电力有限公司 A kind of electric network reactive-load control method based on the prediction of depth generation load
CN111064229A (en) * 2019-12-18 2020-04-24 广东工业大学 Wind-light-gas-storage combined dynamic economic dispatching optimization method based on Q learning
CN111985712A (en) * 2020-08-19 2020-11-24 华北电力大学（保定） Multi-step wind speed combined prediction model building method

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
Yonggang Li 等.Short-Term Direct Probability Prediction Model of Wind Power Based on Improved Natural Gradient Boosting.《energies》.2020,第1-15页. *
刘永强 ; 续毅 ; 贺永辉 ; 柳文斌 ; .基于双向长短期记忆神经网络的风电预测方法.天津理工大学学报.2020,(第05期),第51-56+61页. *
李永刚 ; 王月 ; 刘丰瑞 ; 吴滨源 ; .基于Stacking融合的短期风速预测组合模型.电网技术.2020,(第08期),第85-92页. *
李永刚 等.基于双重 Q 学习的动态风速预测模型.《电工技术学报》.2022,第37卷(第7期),第1810-1819页. *
王毅 ; 于明 ; 李永刚 ; .基于模型预测控制方法的风电直流微网集散控制.电工技术学报.2016,(第21期),第 61-70页. *

Also Published As

Publication number Publication date
CN113515889A (en) 2021-10-19

Similar Documents

Publication Publication Date Title
US11581740B2 (en) Method, system and storage medium for load dispatch optimization for residential microgrid
CN112186743B (en) Dynamic power system economic dispatching method based on deep reinforcement learning
CN105846461B (en) Control method and system for large-scale energy storage power station self-adaptive dynamic planning
CN110474339B (en) Power grid reactive power control method based on deep power generation load prediction
CN112149879B (en) New energy medium-and-long-term electric quantity prediction method considering macroscopic volatility classification
CN105868853B (en) Method for predicting short-term wind power combination probability
CN104638672B (en) Determining method of photovoltaic transmission power limit considering variable correlation
CN105512745A (en) Wind power section prediction method based on particle swarm-BP neural network
CN106849097A (en) A kind of active distribution network tidal current computing method
CN110222883A (en) Load Prediction In Power Systems method based on wind Drive Optimization BP neural network
CN105787592A (en) Wind turbine generator set ultra-short period wind power prediction method based on improved RBF network
CN114362175B (en) Wind power prediction method and system based on depth certainty strategy gradient algorithm
CN113361761A (en) Short-term wind power integration prediction method and system based on error correction
CN115526401A (en) Novel power supply optimal planning method for electric power system based on digital twinning
CN115049292B (en) Intelligent single reservoir flood control scheduling method based on DQN deep reinforcement learning algorithm
CN111612244A (en) QRA-LSTM-based method for predicting nonparametric probability of photovoltaic power before day
CN114626573A (en) Load prediction method for optimizing extreme learning machine based on improved multivariate universe algorithm
CN116544934B (en) Power scheduling method and system based on power load prediction
CN113515889B (en) Dynamic wind speed prediction model building method
CN113435595A (en) Two-stage optimization method for extreme learning machine network parameters based on natural evolution strategy
US20230344242A1 (en) Method for automatic adjustment of power grid operation mode base on reinforcement learning
CN115622056B (en) Energy storage optimal configuration method and system based on linear weighting and selection method
CN115764870A (en) Multivariable photovoltaic power generation power prediction method and device based on automatic machine learning
CN113991742B (en) Distributed photovoltaic double-layer collaborative optimization investment decision-making method for power distribution network
CN109615142A (en) A kind of wind farm wind velocity combination forecasting method based on wavelet analysis

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
GR01 Patent grant
GR01 Patent grant