CN106875033B - Wind power cluster power prediction method based on dynamic self-adaption - Google Patents

Wind power cluster power prediction method based on dynamic self-adaption Download PDF

Info

Publication number
CN106875033B
CN106875033B CN201611215714.XA CN201611215714A CN106875033B CN 106875033 B CN106875033 B CN 106875033B CN 201611215714 A CN201611215714 A CN 201611215714A CN 106875033 B CN106875033 B CN 106875033B
Authority
CN
China
Prior art keywords
prediction
power
wind power
time
cluster
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
CN201611215714.XA
Other languages
Chinese (zh)
Other versions
CN106875033A (en
Inventor
彭小圣
樊闻翰
文劲宇
邓迪元
熊磊
宴青
张勇
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.)
Huazhong University of Science and Technology
State Grid Corp of China SGCC
State Grid Xinjiang Electric Power Co Ltd
Original Assignee
Huazhong University of Science and Technology
State Grid Corp of China SGCC
State Grid Xinjiang Electric Power Co Ltd
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 Huazhong University of Science and Technology, State Grid Corp of China SGCC, State Grid Xinjiang Electric Power Co Ltd filed Critical Huazhong University of Science and Technology
Priority to CN201611215714.XA priority Critical patent/CN106875033B/en
Publication of CN106875033A publication Critical patent/CN106875033A/en
Application granted granted Critical
Publication of CN106875033B publication Critical patent/CN106875033B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

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
    • 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
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation

Landscapes

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

Abstract

The invention provides a wind power cluster power prediction method based on dynamic self-adaptation, which is carried out according to the following steps: step 1: collecting historical data, and dividing the wind power clusters; step 2: according to the divided wind power clusters, three prediction models, namely a time sequence prediction model, a numerical weather forecast prediction model and a space resource matching prediction model, are established, and power prediction of the three prediction models of the wind power clusters is trained; and step 3: selecting a prediction model with the best training error evaluation result according to the training error evaluation results of the three models; and 4, step 4: collecting real-time numerical weather forecast NWP data and real-time power measurement data; and 5: and substituting the real-time NWP data and the real-time power measurement data according to the prediction model selected in the training process to obtain a sub-cluster prediction result, and adding the power prediction results of the sub-clusters to obtain a cluster overall prediction result. According to the method, the optimal prediction model is selected for the wind power clusters under different working conditions, and the prediction precision is improved.

Description

Wind power cluster power prediction method based on dynamic self-adaption
Technical Field
The invention relates to the technical field of wind power generation, in particular to a wind power cluster power prediction method based on dynamic self-adaptation, which is suitable for power prediction of large-scale wind power clusters.
Background
In recent years, with the global energy problem becoming more severe, the development of renewable energy power generation, particularly wind power generation, has become more important. However, wind energy has inherent volatility, instability and intermittency, so that the output of wind power fluctuates with the change of wind speed. If the output of the wind power at the future moment can be correctly predicted, the safe and stable operation of the power grid can be positively influenced. By predicting the wind power generation amount at the future moment, the power grid side can adjust the dispatching plan in advance, so that the problems of unstable electric energy, shortage of supply and the like are avoided. The output value of the wind power plant in a certain day can be obtained in advance at the wind power plant side, so that equipment maintenance and fault maintenance are scientifically arranged.
Most of wind power prediction systems at home and abroad aim at a single wind power plant, and the adopted methods include a physical method, a time series method, an artificial intelligence method and the like. But the power prediction of a single wind farm cannot meet the requirements of grid dispatching. For power grid dispatching, the fluctuation significance of the total power of the wind power cluster formed by a plurality of wind power plants is more important. The wind power cluster power prediction system at home and abroad mainly adopts an overlay method and a statistical upscaling method. And accumulating the power prediction results of the single wind power plant by an superposition method to form the total power of the wind power cluster. The statistical scale-up method includes the steps of firstly selecting a reference wind power plant, predicting the power of the reference wind power plant, and obtaining the power of the wind power cluster through the scale-up of the power prediction result of the reference wind power plant. The methods have certain effect on the power prediction of the cluster, but have the problems of long model training time and low precision.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and improve the power prediction precision of a wind power cluster, provides a dynamic self-adaptive wind power cluster power prediction method, and aims at selecting an optimal prediction model for the wind power cluster under different working conditions to improve the prediction precision.
The technical scheme adopted by the invention is as follows:
a wind power cluster power prediction method based on dynamic self-adaptation is characterized by comprising the following steps:
step 1: collecting historical data of a wind power plant, and dividing wind power clusters according to local geographical positions and a power grid topological structure;
step 2: according to the divided wind power clusters, three prediction models, namely a time sequence prediction model, a numerical weather forecast prediction model and a space resource matching prediction model, are established, and power prediction of the three prediction models of the wind power clusters is trained;
and step 3: selecting a prediction model with the best training error evaluation result according to the training error evaluation results of the three models;
and 4, step 4: collecting real-time numerical weather forecast NWP data and real-time power measurement data;
and 5: and substituting the real-time NWP data and the real-time power measurement data according to the prediction model selected in the training process to obtain a sub-cluster prediction result, and adding the power prediction results of the sub-clusters to obtain a cluster overall prediction result.
The step 1 specifically comprises the following steps:
step 1-1: collecting historical weather forecast data of the wind power plant, wherein the historical weather forecast data contains parameters of wind speed, wind direction, temperature, humidity and air pressure;
step 1-2: collecting geographical position data of a wind power plant, and dividing wind power clusters according to a geographical position approach principle;
step 1-3: and collecting historical power data of each wind power plant.
The step 2 specifically comprises the following steps:
step 2-1: establishing a time series prediction model: the method comprises the steps that an autoregressive moving average model ARMA is used as a time sequence prediction model, parameter identification is conducted on the ARMA model through power data of a historical wind power cluster, and an upstream and downstream effect prediction model is formed; namely, it is
Figure BDA0001191727430000031
Wherein xtRepresenting the power at the time t to be predicted, xt-jRepresenting the measured power at the t-j moment; epsilont-kM and n are the order of ARMA model respectively,
Figure BDA0001191727430000032
θkand the ARMA model order m and n are obtained by a long self-regressive method;
Figure BDA0001191727430000033
is the coefficient of an autoregressive model, thetakIs the moving average model coefficient;
step 2-2: establishing a numerical weather forecast prediction model: the prediction model is based on a BP neural network, and is trained by taking the wind speed and the wind direction of all NWP forecast points in the cluster and the power of the cluster in the first 12 hours of prediction as input parameters and the actual power of the cluster as output parameters; in the training process, the number of hidden layer nodes of the BP neural network is obtained through traversal optimization;
step 2-3: establishing a space resource matching prediction model: the calculation method of the prediction model is shown in formula (2);
Figure BDA0001191727430000034
wherein the content of the first and second substances,
Figure BDA0001191727430000035
the predicted value of the wind power cluster power after h hours; l represents that the weight coefficient of the total L matching sets and the t + h moment to be predicted is the highest by calculating the weight coefficient; p is a radical ofiThe measured value of the wind power cluster power in the matching set is obtained; omegai,t+hThe weight coefficient is a weight coefficient, and the larger the weight coefficient value is, the larger the weight value occupied by the set is; determination of L in equation (2), and weighting factor ωi,t+hThe method of calculation of (1); for prediction of a wind power cluster, the essence of a weight coefficient is to calculate the distance of a space resource parameter between two clusters; the distance di,t+hIs shown in the formula (3);
Figure BDA0001191727430000041
m in the formula (3) represents the number of wind power plants in the cluster, ηkThe weight coefficient of the importance degree of a certain space resource parameter to the whole measurement, for example, the wind speed is the most important parameter for wind power prediction, the weight coefficient can be set to be the highest, and the weight coefficient corresponding to the wind power plant with large capacity is higher than the weight coefficient corresponding to the wind power plant with small capacity; v. ofk,t+hA certain spatial resource parameter, v, for the moment to be predictedk,ia certain space resource parameter of history matching object, wherein β is the weight coefficient occupied by the power distance, Pi,Pt+h-1Representing the power measurements at time i and time t + h-1; drawing an example of a historical power and space resource distance scatter diagram according to the distance calculated by the formula (3); for the historical power and space resource distance scatter diagram, a threshold value delta is sets(ii) a Less than deltasThe historical power corresponding to the matching set of (d) is used for the prediction of real-time power, and is larger than deltasThe set of (b) is considered to be independent of the power to be predicted and can therefore be excluded. Threshold value deltasIs calculated as shown in equation (4), where dminIs the minimum distance value; dmedIs the median of the distance scatter distribution diagram; p is a radical ofrIs from dminAnd dmedWithin interval intercept close to dminThe percentage of data of (c);
δs＝dmin+pr·(dmed-dmin) (4)
for the model calculation formula (2), after the matching sets are determined, the weight coefficient ω of each set needs to be further determinedi,t+hIt is calculated as shown in formula (5), wherein
Figure BDA0001191727430000042
In order to be the distance-weighting factor,
Figure BDA0001191727430000043
is a time weight coefficient;
Figure BDA0001191727430000044
distance weight coefficient
Figure BDA0001191727430000051
The calculation is shown in formula (6), wherein di,t+hand (4) for the distance obtained by calculation of the formula (3), wherein mu is a median in a distance distribution scatter diagram, α is a undetermined coefficient, and optimization selection is carried out in training.
Figure BDA0001191727430000052
Time weight coefficient
Figure BDA0001191727430000053
The obvious effect of the time factor in wind power prediction is reflected, the more important the effect is in the historical data which is closer to the current prediction time point, and the time weight coefficient
Figure BDA0001191727430000054
τiIs a distance in time, τiT + h-i, λ is a time factor, 0<λ<1, optimization selection is needed in the training process. For different prediction time scales, the model corresponds to different optimal parameters.
The step 4 specifically comprises the following steps:
step 4-1: collecting real-time power output data in the SCADA system;
step 4-2: and collecting real-time NWP data of the numerical weather forecast center.
The step 5 specifically includes the following contents:
and (4) selecting a prediction model with the minimum training error according to the step (3), substituting the data in the step (4) into the selected prediction model to obtain a sub-cluster prediction result, and adding the power prediction results of the sub-clusters to obtain a cluster overall prediction result.
The prediction process of the three prediction models is different, and the following discussion is divided into three cases:
and if the upstream and downstream effect prediction model is selected, substituting the power data in the step 4-1 into the formula (1) to obtain a wind power cluster prediction result of 12 hours.
If the weather forecast prediction model is selected, NWP data is first corrected using equation (7).
yt＝x0,t+x1,tvt+x2,tvt 2+x3,tvt 3+qt(7)
Wherein v istIs the wind speed output of the NWP model at time t, ytIs the wind speed prediction error at time t. x is the number ofi,t(i ═ 0,1,2,3) are coefficients estimated using a kalman filter. And substituting the power data obtained in the step 4 and the corrected weather forecast data into a BP neural network model to obtain a 1 st hour prediction result. The predicted power of the 1 st hour is required to be substituted into the input parameter of the 2 nd hour, and so on.
And if the space resource matching prediction model is selected, substituting the NWP data and the predicted previous hour power data into the formulas (2) - (6) for prediction. It is noted that the input parameters contain the power one hour prior to the predicted point in the first 4 hours of prediction, and the input parameters do not contain the power one hour prior to the predicted point in the last several hours of prediction. Iterations of the parameters are entered in the first 4 hours of prediction.
Compared with the prior art, the invention has the following beneficial effects:
the method can realize the wind power cluster power prediction based on the dynamic self-adaptive technology, and further improve the power prediction precision. The method comprises the following specific steps:
(1) the method selects the optimal prediction model type according to the prediction error in the training stage, and avoids low precision caused by random selection of the prediction model.
(2) The invention provides an effective space resource matching model which is simple in modeling, low in calculation complexity, high in precision and strong in practicability.
(3) According to the space resource matching prediction model provided by the invention, the predicted input parameters in the first four hours contain the measured power at the previous moment of the prediction point, and the predicted input parameters in the last 8 hours do not contain the measured power at the previous moment, so that the prediction precision in the first 4 hours is improved, and the prediction precision in the last 4 hours is not influenced.
Drawings
FIG. 1 is sample distance data provided by the present invention;
FIG. 2 is a diagram of an iterative process of BP neural network input parameters;
FIG. 3 is an iterative process diagram of input parameters of a wind power cluster power space resource matching method.
FIG. 4 is a flow chart of overall prediction provided by the present invention.
Detailed Description
The prediction flow of the present invention is further illustrated with reference to the accompanying drawings, and the following examples are illustrative of the present invention but are not intended to limit the scope of the present invention.
As shown in fig. 4, a wind power cluster power prediction method based on dynamic self-adaptation is characterized by comprising the following steps:
step 1: collecting historical weather forecast data of wind speed, wind direction, temperature, humidity and air pressure of each wind power plant, collecting geographical position data of each wind power plant, dividing wind power clusters according to power grid topology, and collecting historical power data of each wind power plant;
step 2: according to the divided wind power clusters, three prediction models, namely a time sequence prediction model, a numerical weather forecast prediction model and a space resource matching prediction model, are established, and power prediction of the three prediction models of the wind power clusters is trained;
the method comprises the following specific steps:
step 2-1: establishing a time series prediction model: the method comprises the steps that an autoregressive moving average model ARMA is used as a time sequence prediction model, parameter identification is conducted on the ARMA model through power data of a historical wind power cluster, and an upstream and downstream effect prediction model is formed; namely, it is
Figure BDA0001191727430000071
Wherein xtRepresenting the power at the time t to be predicted, xt-jRepresenting the measured power at the t-j moment; epsilont-kM and n are the order of ARMA model respectively,
Figure BDA0001191727430000072
θkand the ARMA model order m and n are obtained by a long self-regressive method;
Figure BDA0001191727430000073
is the coefficient of an autoregressive model, thetakIs the moving average model coefficient;
step 2-2: establishing a numerical weather forecast prediction model: the prediction model is based on a BP neural network, and is trained by taking the wind speed and the wind direction of all NWP forecast points in the cluster and the power of the cluster in the first 12 hours of prediction as input parameters and the actual power of the cluster as output parameters; in the training process, the number of hidden layer nodes of the BP neural network is obtained through traversal optimization;
step 2-3: establishing a space resource matching prediction model: the calculation method of the prediction model is shown in formula (2);
Figure BDA0001191727430000081
wherein the content of the first and second substances,
Figure BDA0001191727430000082
the predicted value of the wind power cluster power after h hours; l represents that the weight coefficient of the total L matching sets and the t + h moment to be predicted is the highest by calculating the weight coefficient; p is a radical ofiThe measured value of the wind power cluster power in the matching set is obtained; omegai,t+hThe weight coefficient is a weight coefficient, and the larger the weight coefficient value is, the larger the weight value occupied by the set is; determination of L in equation (2), and weighting factor ωi,t+hThe method of calculation of (1); for prediction of a wind power cluster, the essence of a weight coefficient is to calculate the distance of a space resource parameter between two clusters; the distance di,t+hIs calculated as shown in formula (3);
Figure BDA0001191727430000083
m in the formula (3) represents the number of wind power plants in the cluster, ηkThe weight coefficient of the importance degree of a certain space resource parameter to the whole measurement, for example, the wind speed is the most important parameter for wind power prediction, the weight coefficient can be set to be the highest, and the weight coefficient corresponding to the wind power plant with large capacity is higher than the weight coefficient corresponding to the wind power plant with small capacity; v. ofk,t+hA certain spatial resource parameter, v, for the moment to be predictedk,ia certain space resource parameter of history matching object, wherein β is the weight coefficient occupied by the power distance, Pi,Pt+h-1Representing the power measurements at time i and time t + h-1; an example of a plot of historical power and spatial resource distance scatter is shown in FIG. 2, based on the distances calculated by equation (3). For this figure, a threshold value δ is sets. Less than deltasThe historical power corresponding to the matching set of (d) is used for the prediction of real-time power, and is larger than deltasThe set of (b) is considered to be independent of the power to be predicted and can therefore be excluded. The dashed line in fig. 1 is the threshold value and the box of the solid line is the selected set. Threshold value deltasIs calculated as shown in equation (4), where dminIs the minimum distance value; dmedIs the median of the distance scatter distribution diagram; p is a radical ofrIs from dminAnd dmedWithin interval intercept close to dminThe percentage of data of (c).
δs＝dmin+pr·(dmed-dmin) (4)
For the model calculation formula (2), after the matching sets are determined, the weight coefficient ω of each set needs to be further determinedi,t+hWhich is represented by the calculation formula (5), wherein
Figure BDA0001191727430000091
In order to be the distance-weighting factor,
Figure BDA0001191727430000092
is a time weight coefficient;
Figure BDA0001191727430000093
distance weight coefficient
Figure BDA0001191727430000094
The calculation is shown in formula (6), wherein di,t+hfor the distance calculated by the formula (3), mu is the median in the distance distribution scatter diagram, α is the undetermined coefficient, and optimization selection is carried out in training;
Figure BDA0001191727430000095
time weight coefficient
Figure BDA0001191727430000096
The significant effect of the time factor in wind power prediction is reflected, and the effect of the historical data which is closer to the current prediction time point is more important; time weight coefficient
Figure BDA0001191727430000097
τiIs a distance in time, τiT + h-i, λ is a time factor, 0<λ<1, carrying out optimization selection in the training process. For different prediction time scales, the model corresponds to different optimal parameters.
And step 3: selecting a prediction model with the minimum training error according to the training error evaluation results of the three models;
and 4, step 4: and collecting real-time power output data in the SCADA system and real-time NWP data of a numerical weather forecast center.
And 5: substituting the real-time NWP data and the real-time power measurement data according to the prediction model selected in the training process to obtain a sub-cluster prediction result, and adding the power prediction results of the sub-clusters to obtain a cluster total prediction result; the method specifically comprises the following steps: selecting a prediction model with the minimum training error according to the step 3, substituting the data in the step 4 into the selected prediction model, wherein the prediction processes of the three prediction models are different, and the following discussion is carried out according to three cases:
if the upstream and downstream effect prediction model is selected, substituting the power data of the real-time power output data in the SCADA system collected in the step 4 into the formula (1) to obtain a 12-hour wind power cluster prediction result;
if a weather forecast prediction model is selected, NWP data correction is carried out by using a formula (7);
yt＝x0,t+x1,tvt+x2,tvt 2+x3,tvt 3+qt(7)
wherein v istIs the wind speed output of the NWP model at time t, ytIs the wind speed prediction error at time t. x is the number ofi,t(i ═ 0,1,2,3) are coefficients estimated using a kalman filter. And substituting the power data obtained in the step 4 and the corrected weather forecast data into a BP neural network model to obtain a 1 st hour prediction result. The predicted power of the 1 st hour is required to be substituted into the input parameter of the 2 nd hour, and so on. The detailed iterative process of the input parameters is shown in fig. 2.
And if the space resource matching prediction model is selected, substituting the NWP data and the predicted previous hour power data into the formulas (2) - (6) for prediction. It is noted that the input parameters contain the power one hour prior to the predicted point in the first 4 hours of prediction, and the input parameters do not contain the power one hour prior to the predicted point in the last several hours of prediction. The iterative process of inputting parameters during the first 4 hours of prediction is shown in fig. 3.

Claims (7)

1. A wind power cluster power prediction method based on dynamic self-adaptation is characterized by comprising the following steps:
step 1: collecting historical data of a wind power plant, and dividing wind power clusters according to local geographical positions and a power grid topological structure;
step 2: according to the divided wind power clusters, three prediction models, namely a time sequence prediction model, a numerical weather forecast prediction model and a space resource matching prediction model, are established, and power prediction of the three prediction models of the wind power clusters is trained; the method specifically comprises the following steps:
step 2-1: establishing a time series prediction model: an autoregressive moving average model ARMA is used as a time sequence prediction model, and parameter identification is carried out on the ARMA model by using power data of a historical wind power cluster to form an upstream and downstream effect prediction model; namely, it is
Figure FDA0002290648820000011
Wherein xtRepresenting the power at the time t to be predicted, xt-jRepresenting the measured power at the t-j moment; epsilont-kM and n are the order of ARMA model respectively,
Figure FDA0002290648820000012
θkand the ARMA model order m and n are obtained by a long self-regressive method;
Figure FDA0002290648820000013
is the coefficient of an autoregressive model, thetakIs the moving average model coefficient;
step 2-2: establishing a numerical weather forecast prediction model: the prediction model is based on a BP neural network, and is trained by taking the wind speed and the wind direction of all NWP forecast points in the cluster and the power of the cluster in 12 hours before prediction as input parameters and the actual power of the cluster as output parameters; in the training process, the number of hidden layer nodes of the BP neural network is obtained through traversal optimization;
step 2-3: establishing a space resource matching prediction model: the calculation method of the prediction model is shown in formula (2);
Figure FDA0002290648820000021
wherein the content of the first and second substances,
Figure FDA0002290648820000022
the predicted value of the wind power cluster power after h hours; l represents that the weight coefficient of the total L matching sets and the t + h moment to be predicted is the highest by calculating the weight coefficient; p is a radical ofiThe measured value of the wind power cluster power in the matching set is obtained; omegai,t+hThe weight coefficient is a weight coefficient, and the larger the weight coefficient value is, the larger the weight value occupied by the set is; determination of L in equation (2), and weighting factor ωi,t+hThe method of calculation of (1); for prediction of a wind power cluster, the essence of a weight coefficient is to calculate the distance of a space resource parameter between two clusters; the distance di,t+hIs shown in the formula (3);
Figure FDA0002290648820000023
m in the formula (3) represents the number of wind power plants in the cluster, ηkThe weight coefficient of the importance degree of a certain space resource parameter to the whole measurement is the weight coefficient, for example, the wind speed is the most important parameter for wind power prediction, the weight coefficient is set to be the highest, and the weight coefficient corresponding to the wind power plant with large capacity is higher than the weight coefficient corresponding to the wind power plant with small capacity; v isk,t+hA certain spatial resource parameter, v, for the moment to be predictedk,ia certain space resource parameter of history matching object, wherein β is the weight coefficient occupied by the power distance, Pi,Pt+h-1Representing the power measurements at time i and time t + h-1; drawing an example of a historical power and space resource distance scatter diagram according to the distance calculated by the formula (3); for the historical power and space resource distance scatter diagram, a threshold value delta is sets(ii) a Less than deltasThe historical power corresponding to the matching set of (d) is used for the prediction of real-time power, and is larger than deltasThe set of (a) is regarded as being independent of the power to be predicted and excluded; threshold value deltasIs shown in the formula (4), wherein dminIs the minimum distance value; dmedIs the median of the distance scatter distribution diagram; p is a radical ofrIs from dminAnd dmedWithin interval intercept close to dminThe percentage of data of (c);
δs＝dmin+pr·(dmed-dmin) (4)
for the model calculation formula (2), after the matching sets are determined, the weight coefficient ω of each set needs to be further determinedi,t+hWhich is represented by the calculation formula (5), wherein
Figure FDA0002290648820000031
In order to be the distance-weighting factor,
Figure FDA0002290648820000032
is a time weight coefficient;
Figure FDA0002290648820000033
distance weight coefficient
Figure FDA0002290648820000034
The calculation is shown in formula (6), wherein di,t+hfor the distance calculated by the formula (3), mu is the median in the distance distribution scatter diagram, α is the undetermined coefficient, and optimization selection is carried out in training;
Figure FDA0002290648820000035
time weight coefficient
Figure FDA0002290648820000036
The significant effect of the time factor in wind power prediction is reflected, and the effect of the historical data which is closer to the current prediction time point is more important; time weight coefficient
Figure FDA0002290648820000037
τiIs a distance in time, τiT + h-i, λ is a time factor, 0<λ<1, carrying out optimization selection in a training process; for different prediction time scales, the model corresponds to noThe same optimal parameters;
and step 3: selecting a prediction model with the best training error evaluation result according to the training error evaluation results of the three models;
and 4, step 4: collecting real-time numerical weather forecast NWP data and real-time power measurement data;
and 5: and substituting the real-time NWP data and the real-time power measurement data according to the prediction model selected in the training process to obtain a sub-cluster prediction result, and adding the power prediction results of the sub-clusters to obtain a cluster overall prediction result.
2. The dynamic self-adaptive wind power cluster power prediction method according to claim 1, wherein the step 1 specifically comprises the following steps:
step 1-1: collecting historical weather forecast data of the wind power plant, wherein the historical weather forecast data contains parameters of wind speed, wind direction, temperature, humidity and air pressure;
step 1-2: collecting geographical position data of a wind power plant, and dividing wind power clusters according to a geographical position approach principle;
step 1-3: and collecting historical power data of each wind power plant.
3. The dynamic self-adaptive wind power cluster power prediction method according to claim 1, wherein the step 4 specifically comprises the following steps:
step 4-1: collecting real-time power output data in the SCADA system;
step 4-2: and collecting real-time NWP data of the numerical weather forecast center.
4. The wind power cluster power prediction method based on dynamic self-adaptation according to claim 1, wherein the step 5 specifically comprises: and (4) selecting a prediction model with the minimum training error according to the step (3), substituting the data in the step (4) into the selected prediction model to obtain a sub-cluster prediction result, and adding the power prediction results of the sub-clusters to obtain a cluster overall prediction result.
5. The wind power cluster power prediction method based on dynamic self-adaptation as claimed in claim 3, wherein if an upstream and downstream effect prediction model is selected, the power data in step 4-1 is substituted into formula (1) to obtain a 12-hour wind power cluster prediction result.
6. The wind power cluster power prediction method based on dynamic self-adaptation as claimed in claim 4, characterized in that if a weather forecast prediction model is selected, NWP data correction is performed by using formula (7) first;
yt＝x0,t+x1,tvt+x2,tvt 2+x3,tvt 3+qt(7)
wherein v istIs the wind speed output of the NWP model at time t, ytIs the wind speed prediction error at time t; x is the number ofi,t(i is 0,1,2,3) is a coefficient estimated by adopting a Kalman filter, and then the power data obtained in the step 4 and the corrected weather forecast data are substituted into a BP neural network model to obtain a prediction result at the 1 st hour; the predicted power of the 1 st hour is required to be substituted into the input parameter of the 2 nd hour, and so on.
7. The wind power cluster power prediction method based on dynamic self-adaptation as claimed in claim 4, wherein if a space resource matching prediction model is selected, the NWP data and the power data predicted in the previous hour are substituted into equations (2) - (6) for prediction; it is noted that the input parameters contain the power one hour before the predicted point in the first 4 hours of prediction, the input parameters do not contain the power one hour before the predicted point in the last several hours of prediction, and the input parameters iterate in the first 4 hours of prediction.
CN201611215714.XA 2016-12-26 2016-12-26 Wind power cluster power prediction method based on dynamic self-adaption Active CN106875033B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201611215714.XA CN106875033B (en) 2016-12-26 2016-12-26 Wind power cluster power prediction method based on dynamic self-adaption

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201611215714.XA CN106875033B (en) 2016-12-26 2016-12-26 Wind power cluster power prediction method based on dynamic self-adaption

Publications (2)

Publication Number Publication Date
CN106875033A CN106875033A (en) 2017-06-20
CN106875033B true CN106875033B (en) 2020-06-02

Family

ID=59164185

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201611215714.XA Active CN106875033B (en) 2016-12-26 2016-12-26 Wind power cluster power prediction method based on dynamic self-adaption

Country Status (1)

Country Link
CN (1) CN106875033B (en)

Families Citing this family (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR102089772B1 (en) * 2017-12-18 2020-03-17 두산중공업 주식회사 System and method for predicting power usage
CN108256693B (en) * 2018-02-11 2024-02-13 阳光电源股份有限公司 Photovoltaic power generation power prediction method, device and system
CN108428019B (en) * 2018-05-15 2021-09-03 阳光电源股份有限公司 Method for establishing component battery temperature calculation model and photovoltaic power prediction method
CN111323847B (en) * 2018-12-13 2023-05-23 北京金风慧能技术有限公司 Method and apparatus for determining weight ratios for analog integration algorithms
CN110570030A (en) * 2019-08-22 2019-12-13 国网山东省电力公司经济技术研究院 Wind power cluster power interval prediction method and system based on deep learning
CN111222738B (en) * 2019-10-18 2022-04-15 华中科技大学 Method for predicting power and optimizing parameters of wind power cluster
CN111191815B (en) * 2019-11-25 2022-08-16 清华大学 Ultra-short-term output prediction method and system for wind power cluster
CN112906928B (en) * 2019-12-03 2022-09-16 国网山西省电力公司电力科学研究院 Wind power plant cluster active power prediction method and system
CN114188935A (en) * 2020-09-15 2022-03-15 华为技术有限公司 Power prediction method, device and equipment
CN112633632A (en) * 2020-11-26 2021-04-09 华中科技大学 Integrated short-term wind power cluster power prediction method based on signal decomposition technology
CN112990533B (en) * 2021-01-19 2024-01-05 中国农业大学 Wind power cluster power prediction method adopting sparse constraint and dynamic weight distribution
CN114330097A (en) * 2021-11-26 2022-04-12 中国华能集团清洁能源技术研究院有限公司 Heterogeneous data and deep learning wind power cluster power prediction method and device

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103699944A (en) * 2013-12-31 2014-04-02 国电南京自动化股份有限公司 Wind and photovoltaic generation power prediction system with multiple prediction modes
WO2015159356A1 (en) * 2014-04-15 2015-10-22 株式会社日立製作所 Power control system and power control method
CN106251242A (en) * 2016-08-08 2016-12-21 东南大学 A kind of wind power output interval combinations Forecasting Methodology

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103699944A (en) * 2013-12-31 2014-04-02 国电南京自动化股份有限公司 Wind and photovoltaic generation power prediction system with multiple prediction modes
WO2015159356A1 (en) * 2014-04-15 2015-10-22 株式会社日立製作所 Power control system and power control method
CN106251242A (en) * 2016-08-08 2016-12-21 东南大学 A kind of wind power output interval combinations Forecasting Methodology

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
"风电集群短期及超短期功率预测精度改进方法综述";彭小圣 等;《中国电机工程学报》;20161205;第36卷(第23期);第6315-6323页 *

Also Published As

Publication number Publication date
CN106875033A (en) 2017-06-20

Similar Documents

Publication Publication Date Title
CN106875033B (en) Wind power cluster power prediction method based on dynamic self-adaption
Wang et al. Approaches to wind power curve modeling: A review and discussion
Lange et al. Wind power prediction in Germany–Recent advances and future challenges
CN103268366B (en) A kind of combination wind power forecasting method suitable for distributing wind power plant
CN102880810B (en) Wind power prediction method based on time sequence and neural network method
CN106779226B (en) Fan batch power prediction method based on mixed kernel machine learning
CN111353652B (en) Wind power output short-term interval prediction method
CN105574615B (en) wavelet-BP neural network wind power prediction method based on spatial correlation and GA
Tesfaye et al. Short-term wind power forecasting using artificial neural networks for resource scheduling in microgrids
CN105046374A (en) Power interval predication method based on nucleus limit learning machine model
CN109086928A (en) Photovoltaic plant realtime power prediction technique based on SAGA-FCM-LSSVM model
CN102663251A (en) Physical prediction method for wind power station power based on computational fluid mechanics model
CN110766200A (en) Method for predicting generating power of wind turbine generator based on K-means mean clustering
CN105069236B (en) Consider the broad sense load joint probability modeling method of wind power plant node space correlation
CN104933483A (en) Wind power forecasting method dividing based on weather process
CN111612244B (en) QRA-LSTM-based method for predicting nonparametric probability of photovoltaic power before day
CN104732300A (en) Neural network wind power short-term forecasting method based on fuzzy partition theory
CN103020743A (en) Ultra-short-term wind speed forecasting method for wind power plant
Chen et al. Research on wind power prediction method based on convolutional neural network and genetic algorithm
Japar et al. Estimating the wake losses in large wind farms: A machine learning approach
CN111553536A (en) Power grid load prediction method based on multiple energy structures
CN105844350B (en) Short-term wind power prediction system based on covariance optimal combination model
Kirchner-Bossi et al. Wind farm power density optimization according to the area size using a novel self-adaptive genetic algorithm
CN107453366B (en) UPFC-containing multi-target optimal power flow calculation method considering wind power decision risk
CN106251238B (en) Wind power plant modeling sequence discretization step length selection and model error analysis method

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