WO2023134478A1 - 超短期风电功率预测方法和装置 - Google Patents

超短期风电功率预测方法和装置 Download PDF

Info

Publication number
WO2023134478A1
WO2023134478A1 PCT/CN2022/144103 CN2022144103W WO2023134478A1 WO 2023134478 A1 WO2023134478 A1 WO 2023134478A1 CN 2022144103 W CN2022144103 W CN 2022144103W WO 2023134478 A1 WO2023134478 A1 WO 2023134478A1
Authority
WO
WIPO (PCT)
Prior art keywords
wind
power
correlation coefficient
wind turbine
generator
Prior art date
Application number
PCT/CN2022/144103
Other languages
English (en)
French (fr)
Inventor
肖建华
刘冬明
龚贤夫
陈鸿琳
罗苑萍
傅惠芹
刘满
张莉
林晓波
李暖群
Original Assignee
广东电网有限责任公司揭阳供电局
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 广东电网有限责任公司揭阳供电局 filed Critical 广东电网有限责任公司揭阳供电局
Publication of WO2023134478A1 publication Critical patent/WO2023134478A1/zh

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
    • G06N3/04—Architecture, e.g. interconnection topology
    • 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
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Definitions

  • the embodiments of the present application relate to the technical field of wind power forecasting, for example, to a method and device for ultra-short-term wind power forecasting.
  • the present application provides an ultra-short-term wind power prediction method and device.
  • the embodiment of the present application provides a method for ultra-short-term wind power prediction.
  • the ultra-short-term wind power prediction method includes:
  • the optimal power prediction result is obtained by using a clustering algorithm.
  • the embodiment of the present application also provides an ultra-short-term wind power prediction device, the ultra-short-term wind power prediction device includes:
  • the historical power data acquisition module is configured to obtain the historical power data of each wind turbine
  • the feature data acquisition module is configured to acquire various feature data affecting wind power of each wind turbine
  • a characteristic matrix building module is configured to establish a characteristic matrix of each wind generator according to historical power data of each wind generator and various characteristic data affecting wind power;
  • a multiple correlation coefficient calculation module is configured to calculate multiple correlation coefficients between each wind power generator according to the characteristic matrix of each wind power generator
  • a comprehensive correlation coefficient matrix construction module is configured to construct a comprehensive correlation coefficient matrix between each wind power generator according to multiple correlation coefficients between each wind power generator;
  • the input feature parameter determination module is configured to determine the required input feature parameters for each wind generator power prediction according to the comprehensive correlation coefficient matrix between each wind generator;
  • a variety of forecasting model building modules are set to establish multiple forecasting models according to the input characteristic parameters required for power forecasting of each wind turbine;
  • the power prediction result obtaining module is configured to obtain corresponding power prediction results according to each prediction model training
  • the optimal power prediction result solving module is configured to use a clustering algorithm to obtain the optimal power prediction result according to each power prediction result.
  • Fig. 1 is a flow chart of a method for ultra-short-term wind power prediction in Embodiment 1 of the present application;
  • Fig. 2 is a flow chart of a method for ultra-short-term wind power prediction in Embodiment 2 of the present application;
  • Fig. 3 is a structural block diagram of an ultra-short-term wind power prediction device in Embodiment 3 of the present application.
  • Figure 1 is a flow chart of a method for ultra-short-term wind power prediction provided in Embodiment 1 of the present application.
  • This embodiment can be applied to an offshore wind power management platform to implement a method for improving the accuracy of ultra-short-term wind power prediction.
  • This method can Executed by the ultra-short-term wind power forecasting device, the device can be implemented by software and/or hardware, and the device can be configured in the server of the management platform, referring to Figure 1, including the following steps:
  • Step 110 obtaining historical power data of each wind motor and various characteristic data affecting wind power
  • offshore wind power prediction is affected by many factors, such as atmospheric temperature, wind speed, sea water temperature, wind speed at the wind turbine blades, angle between wind turbine blades and wind direction, etc. Although many factors will affect the power output of offshore wind turbines, the degree of influence of each factor is different, and some factors can even be ignored directly. Therefore, various characteristic data affecting wind power refer to factors other than factors that can be ignored directly.
  • the historical power data of each wind turbine can be obtained from the data acquisition and monitoring control system (Supervisory Control And Data Acquisition, SCADA) of the offshore wind farm to obtain the power time series of each wind turbine.
  • SCADA Supervisory Control And Data Acquisition
  • Various characteristic data affecting wind power of each wind motor can be collected by data collection equipment or sensors, for example, temperature can be collected by a temperature sensor.
  • the collected historical power data of each wind motor and various characteristic data affecting wind power are sent to the data storage device for storage. Therefore, the server can directly obtain the historical power data of each wind turbine and various characteristic data affecting wind power from the data storage device.
  • Step 120 establishing a characteristic matrix of each wind generator according to the historical power data of each wind generator and various characteristic data affecting wind power;
  • each wind turbine After obtaining the various characteristic data affecting the wind power of each wind turbine, in order to unify the statistical distribution of the sample data and facilitate subsequent processing, it is necessary to normalize the various characteristic data, that is, each wind turbine
  • the wind speed, wind direction and temperature of the surrounding environment, as well as the flow speed, flow direction and temperature of the surrounding sea water of each wind turbine are normalized.
  • the normalized characteristic data of each wind turbine and the corresponding historical power data are spliced into a characteristic matrix X of each wind turbine.
  • the normalization process uses the min-max normalization method to normalize the data to [0, 1], the formula is:
  • x is various feature data vectors
  • x min is the minimum value of the corresponding feature data vector
  • x max is the maximum value of the corresponding feature data vector
  • Step 130 according to the characteristic matrix of each wind turbine, calculate various correlation coefficients between each wind turbine;
  • the calculated correlation coefficient between each wind turbine can reflect the correlation between each wind turbine in terms of different characteristics or dimensions.
  • the correlation coefficient can be characterized in many ways.
  • the Pearson correlation coefficient can determine the correlation between each wind turbine according to the waveform;
  • the Spearman correlation coefficient can determine the correlation between each wind turbine according to the monotonicity.
  • Correlation; the R2 correlation coefficient can explain the degree of mutual influence of each wind turbine;
  • the Euclidean distance is used to determine the degree of correlation between the characteristics of two offshore wind turbines.
  • the correlation coefficient may also include other types, which can be set according to actual conditions, and are not limited here.
  • Step 140 constructing a comprehensive correlation coefficient matrix between each wind turbine according to various correlation coefficients among each wind generator
  • the various correlation coefficients among the wind power generators can reflect the correlation of characteristics between the wind power generators from different angles. Different correlation coefficient calculation methods show different characteristics. Considering that different correlation coefficients have different emphases, in order to improve the prediction accuracy, when evaluating the correlation between offshore wind turbines, multiple correlation coefficients are combined, and the Various correlation coefficients are weighted to construct a comprehensive correlation coefficient matrix between each wind turbine to more accurately reflect the interaction of characteristics between offshore wind turbines, for example, the interaction of wakes between wind turbines .
  • Step 150 according to the comprehensive correlation coefficient matrix between each wind turbine, determine the input characteristic parameters required for power prediction of each wind turbine;
  • the input characteristic parameters required for power prediction of each wind turbine are the basic parameters for training various power prediction models.
  • the input characteristic parameters are the characteristics of important factors affecting the power change of the wind turbine. Therefore, before training according to the prediction model, it is necessary to perform feature screening on the input feature parameters required for prediction, that is, according to the comprehensive correlation coefficient matrix between each wind turbine generator, the final input characteristics of each wind turbine power prediction can be determined parameter.
  • Step 160 according to the input characteristic parameters required for power prediction of each wind turbine, establish multiple prediction models; and obtain corresponding power prediction results according to the training of each prediction model;
  • multiple prediction models are established for each wind turbine, and the input characteristic parameters required for power prediction of each wind turbine are input to each wind generator
  • Each prediction model of the wind turbine is trained, and the power prediction results of each prediction model of the corresponding wind turbine are obtained.
  • a variety of prediction models are established for each wind turbine, and training is performed according to each prediction model, in order to obtain the training situation of each wind turbine under multiple prediction models, so that the prediction of multiple prediction models can be subsequently
  • the results are coupled to realize the coupling of the advantages of various prediction models to obtain the optimal prediction results, thereby improving the accuracy and reliability of power prediction.
  • Step 170 according to each power prediction result, a clustering algorithm is used to obtain an optimal power prediction result.
  • the clustering algorithm is used to couple the prediction results of various prediction models to solve the optimal prediction results, which fully retains the advantages of different prediction models, and the use of clustering algorithms and its organic combination further improves the prediction accuracy of the prediction model. It has important practical engineering significance.
  • the working principle of the ultra-short-term wind power prediction method is as follows: firstly, obtain the historical power data of each wind power motor and various characteristic data affecting wind power; A characteristic matrix of each wind generator is established based on various characteristic data affecting wind power; according to the characteristic matrix of each wind generator, various correlation coefficients between each wind generator are calculated; According to the various correlation coefficients of each wind turbine, the comprehensive correlation coefficient matrix between each wind turbine is constructed; according to the comprehensive correlation coefficient matrix between each wind turbine, the input characteristic parameters required for each wind turbine power prediction are determined; according to The input characteristic parameters required for the power prediction of each wind turbine, establish a variety of prediction models; and obtain the corresponding power prediction results according to the training of each prediction model; according to the power prediction results, use the clustering algorithm to solve the optimal power prediction results .
  • the ultra-short-term wind power prediction method includes: obtaining historical power data of each wind power motor and various characteristic data affecting wind power; Power data and various characteristic data affecting wind power to establish the characteristic matrix of each wind turbine; according to the characteristic matrix of each wind turbine, calculate the various correlation coefficients between each wind turbine; According to the multiple correlation coefficients between wind turbines, the comprehensive correlation coefficient matrix between each wind turbine is constructed; according to the comprehensive correlation coefficient matrix between each wind turbine, the input characteristics required for power prediction of each wind turbine are determined Parameters; according to the input characteristic parameters required for each wind turbine power prediction, a variety of prediction models are established; and the corresponding power prediction results are obtained according to the training of each prediction model; according to each power prediction result, the clustering algorithm is used to solve the optimal Power prediction results.
  • the power of offshore wind turbines can be predicted, and various characteristic data that affect offshore wind power prediction can be fully considered, and the characteristic matrix of each wind turbine can be generated by combining the historical power data of each wind turbine.
  • the characteristic matrix constructs the comprehensive correlation coefficient matrix between each wind turbine and determines the final input characteristic parameters to the prediction model according to the comprehensive correlation coefficient matrix, so as to improve the prediction accuracy.
  • the feature data at least includes the wind speed, wind direction and temperature of the surrounding environment of each wind power generator, the blade angle of each wind power generator, and the surrounding sea water of each wind power generator flow velocity, flow direction and temperature.
  • the wind speed and direction of the surrounding environment of each wind turbine will affect the wake of the offshore wind turbine.
  • the wake factors of offshore wind turbines will affect the power prediction of wind turbines to a certain extent. Therefore, in order to improve the accuracy of wind turbine power prediction, it is necessary to consider the two factors of wind speed and wind direction in the surrounding environment of each wind turbine. characteristic data.
  • some data need to be preprocessed to improve the accuracy of the data.
  • the wind speed and wind direction data of the surrounding environment of each wind turbine assuming that the wind speed is v speed and the wind direction is v windD , it is decomposed into the meridional wind speed v windV and the latitudinal wind speed v windU , and the calculation formula is as follows:
  • Fig. 2 is a flowchart of an ultra-short-term wind power forecasting method provided in Embodiment 2 of the present application.
  • the various correlation coefficients at least include: Pearson correlation coefficient, Spearman correlation coefficient, R2 coefficient and Euclidean distance.
  • the ultra-short-term wind power prediction method includes the following steps:
  • Step 210 obtaining historical power data of each wind turbine and various characteristic data affecting wind power
  • Step 220 establishing a characteristic matrix of each wind generator according to the historical power data of each wind generator and various characteristic data affecting wind power;
  • Step 230 according to the characteristic matrix of each wind turbine, calculate various correlation coefficients between each wind turbine;
  • Step 240 constructing a comprehensive correlation coefficient formula between each wind turbine according to various correlation coefficients between each wind turbine;
  • the comprehensive correlation coefficient formula between each wind power generator is:
  • the calculation formula of the coefficient ⁇ i corresponding to various correlation coefficients is:
  • i 0, 1, 2, 3.
  • ⁇ 3 is the determinant of matrix A
  • ⁇ 0 , ⁇ 1 , and ⁇ 2 are sub-determinants of the determinant of matrix A.
  • the method of calculating the coefficient ⁇ i corresponding to various correlation coefficients through the calculation formula of the coefficient ⁇ i and the calculation formulas of ⁇ 0 , ⁇ 1 , ⁇ 2 and ⁇ 3 can improve the simplicity of calculation.
  • Step 250 calculate the comprehensive correlation coefficient between each wind power generator according to the comprehensive correlation coefficient formula between each wind power generator; The comprehensive correlation coefficient matrix of .
  • ⁇ ij is the comprehensive correlation coefficient between offshore wind turbines i and j.
  • Step 260 sequentially judge whether each element in the comprehensive correlation coefficient matrix meets the preset threshold; if yes, set the corresponding element value to 1; if not, set the corresponding element value to 0; and obtain a standardized comprehensive correlation coefficient matrix .
  • each element ⁇ ij in the comprehensive correlation coefficient matrix Z exceeds the preset threshold k, if it exceeds, the value of the corresponding element is set to 1; if it is not exceeded, the value of the corresponding element is set to 0, by
  • each column in the standardized comprehensive correlation coefficient matrix Z* represents a wind turbine
  • an element with an element value of 1 in each column corresponds to a characteristic data of the wind turbine in the column, that is, the element value in each column is 1
  • the number of represents the total number of wind turbine features used in the prediction of the row of wind turbines.
  • the preset threshold k can be set according to actual conditions, and is not limited here.
  • Step 270 according to the standardized comprehensive correlation coefficient matrix, according to the preset screening rules to determine the input characteristic parameters required for power prediction of each wind turbine.
  • the comprehensive correlation coefficient matrix is standardized according to the preset threshold, which is only the initial screening of the input features required by the prediction model of each wind turbine.
  • the preset screening rules perform secondary screening on the standardized comprehensive correlation coefficient matrix to determine the final input characteristic parameters required for power prediction of each wind turbine.
  • the elements of each column in the standardized comprehensive correlation coefficient matrix are screened out according to the preset screening rules to select the characteristic elements that meet the preset screening conditions, and are used as the power prediction requirements of the wind power generators in the corresponding column.
  • each column vector in the standardized comprehensive correlation coefficient matrix represents a characteristic parameter vector of a wind turbine.
  • the i-th column in the standardized comprehensive correlation coefficient matrix corresponds to the n-th wind turbine
  • set the i-th column as Y [x 1 , x 2 , . . . , x n ], where, in Y
  • the value of each element in Y is 0 or 1
  • the value of each element in Y is 1, which represents a feature of the nth wind turbine.
  • filter each element in Y according to the preset filtering rules to select the feature elements that meet the preset filtering conditions. Assuming that the value of the element x 2 in Y is 1 and meets the preset filtering conditions, then take the element x 2 as the nth An input characteristic parameter required for power prediction of a wind turbine.
  • the preset screening rules all the elements satisfying the preset conditions in the column matrix Y are screened out, that is, all the input characteristic parameters required for the power prediction of the nth wind turbine are screened out.
  • the default filtering rules are:
  • the strong correlation of the Pearson correlation coefficient means that the absolute value of the Pearson correlation coefficient reaches the preset coefficient threshold and above, and there is correlation, and the larger the absolute value coefficient of the Pearson correlation coefficient, the stronger the correlation.
  • the preset coefficient threshold may be 0.6, which may be set according to actual conditions, and is not limited here.
  • Step 280 according to the input characteristic parameters required for power prediction of each wind turbine, establish multiple prediction models; and obtain corresponding power prediction results according to the training of each prediction model;
  • Step 290 according to each power prediction result, use a clustering algorithm to obtain an optimal power prediction result.
  • the plurality of prediction models at least include: BP neural network model, convolutional neural network model (Convolutional Neural Networks, CNN), gated recurrent neural network model (Recurrent Neural Network, RNN) and CNN-GRU model.
  • BP neural network model convolutional neural network model
  • CNN convolutional Neural Networks
  • RNN gated recurrent neural network model
  • CNN-GRU model CNN-GRU model
  • the BP neural network model, convolutional neural network model, gated recurrent neural network model and CNN-GRU model of each wind turbine are respectively established.
  • the establishment and training methods of the BP neural network model, convolutional neural network model, gated recurrent neural network model and CNN-GRU model corresponding to each wind turbine are as follows:
  • BP neural network model For the BP neural network model: first, establish a three-layer BP neural network, for example, set the number of neurons to 64, 128, 256 in order to construct the BP neural network model; Input the characteristic parameters, train the BP neural network model, and finally output the predicted power y predict1 of the offshore wind turbine.
  • the convolutional neural network model first, establish a two-layer CNN model, for example, set the number of neurons to 4 and 8 in order to construct the convolutional neural network model; then, according to the input of the wind turbine finally determined above Feature parameters, train the convolutional neural network model, and finally output the predicted power y predictet2 of the offshore wind turbine.
  • the gated cyclic neural network model first, establish a two-layer GRU model, for example, set the number of neurons to 4 and 8 in turn to construct the gated cyclic neural network model; then, according to the above-mentioned finalized wind turbine The input feature parameters of , train the gated recurrent neural network model, and finally output the predicted power y predict3 of the offshore wind turbine.
  • For the CNN-GRU model First, build a combined model of one layer of CNN and two layers of GRU. For example, set the number of neurons to 4, 8, and 4 in order to build a CNN-GRU model; then, according to the above final determination The input characteristic parameters of the wind turbine, train the CNN-GRU model, and finally output the predicted power y predict4 of the offshore wind turbine.
  • PSO particle swarm optimization
  • y predict ⁇ 1 ⁇ y predict1 + ⁇ 2 ⁇ y predictet2 + ⁇ 3 ⁇ y predict3 + ⁇ 4 ⁇ y predict4
  • y predict is the final prediction result of the comprehensive model
  • ⁇ i is the weight corresponding to the prediction results of various prediction models, satisfying:
  • the particle swarm optimization is used to solve the optimal value of ⁇ i to obtain the optimal power prediction result.
  • f obj is the optimization target; n is the number of samples, that is, the number of all elements in the matrix; i is the first element in the matrix; y predict is the predicted value, which is a matrix; y truth is the real value.
  • x i and v i are the velocity and position of the particle respectively; c 1 and c 2 are random numbers between 0 and 1; c 1 and c 2 are the learning factors; ⁇ is the inertia factor.
  • the particle position and velocity update formula of the PSO algorithm is executed to obtain the new particle position and velocity.
  • the update formula for updating the population P is as follows:
  • L is the particle population corresponding to the minimum target value calculated by particles P(i) and X 1(t+1) (i) according to the above formula.
  • Fig. 3 is a structural block diagram of an ultra-short-term wind power prediction device provided in Embodiment 3 of the present application.
  • Embodiment 3 of the present application provides an ultra-short-term wind power prediction device.
  • the device 100 includes:
  • Historical power data acquisition module 10 used to obtain the historical power data of each wind-driven motor
  • Feature data acquisition module 20 used to acquire various feature data affecting wind power of each wind-driven motor
  • the feature matrix building module 30 is used to set up the feature matrix of each wind generator according to the historical power data of each wind generator and various characteristic data affecting wind power;
  • Multiple correlation coefficient calculation module 40 used for calculating multiple correlation coefficients between each wind power generator according to the characteristic matrix of each wind power generator;
  • the comprehensive correlation coefficient matrix construction module 50 is used to construct the comprehensive correlation coefficient matrix between each wind power generator according to various correlation coefficients between each wind power generator;
  • the input characteristic parameter determination module 60 is used to determine the input characteristic parameters required for power prediction of each wind-driven generator according to the comprehensive correlation coefficient matrix between each wind-driven generator;
  • a variety of prediction model building modules 70 used to establish a variety of prediction models according to the input characteristic parameters required for power prediction of each wind turbine;
  • the power prediction result obtaining module 80 is used to obtain the corresponding power prediction result according to each prediction model training
  • the optimal power prediction result solving module 90 is configured to use a clustering algorithm to obtain the optimal power prediction result according to each power prediction result.
  • the ultra-short-term wind power prediction device includes: a historical power data acquisition module, used to acquire the historical power data of each wind power motor; a feature data acquisition module, used to acquire Various characteristic data affecting wind power of each wind power generator; characteristic matrix building module, used to establish the characteristic matrix of each wind power generator according to the historical power data of each wind power motor and various characteristic data affecting wind power; A correlation coefficient calculation module is used to calculate multiple correlation coefficients between each wind generator according to the characteristic matrix of each wind generator; a comprehensive correlation coefficient matrix construction module is used to calculate various correlation coefficients between each wind generator A variety of correlation coefficients are used to construct the comprehensive correlation coefficient matrix between each wind turbine; the input feature parameter determination module is used to determine the power prediction value of each wind turbine according to the comprehensive correlation coefficient matrix between each wind turbine.
  • the required input characteristic parameters; a variety of prediction model building modules are used to establish a variety of prediction models according to the input characteristic parameters required for power prediction of each wind turbine; the power prediction result acquisition module is used to obtain according to each prediction model training.
  • the corresponding power prediction result; the module for solving the optimal power prediction result is used to obtain the optimal power prediction result by using a clustering algorithm according to each power prediction result.
  • multiple correlation coefficients include at least: Pearson correlation coefficient, Spearman correlation coefficient, R2 coefficient and Euclidean distance;
  • a comprehensive correlation coefficient matrix building block 50 comprising:
  • the comprehensive correlation coefficient formula construction unit is used for constructing the comprehensive correlation coefficient formula between each wind power generator according to various correlation coefficients between each wind power generator;
  • the comprehensive correlation coefficient calculation unit is used to calculate the comprehensive correlation coefficient between each wind power generator according to the comprehensive correlation coefficient formula between each wind power generator;
  • the comprehensive correlation coefficient matrix forming unit is used to form the comprehensive correlation coefficient matrix among the wind power generators based on the comprehensive correlation coefficients among the wind power generators.
  • the comprehensive correlation coefficient formula between each wind power generator is:
  • Corr R2 is the R2 coefficient between each wind turbine;
  • ⁇ 2 is the coefficient corresponding to Corr R2 ;
  • Corr d is the Euclidean distance between each wind generator;
  • ⁇ 3 is the coefficient corresponding to Corr d ;
  • ⁇ ij is the comprehensive correlation coefficient between offshore wind turbines i and j.
  • the calculation formula of the coefficient ⁇ i corresponding to various correlation coefficients is:
  • i 0, 1, 2, 3.
  • the input feature parameter determination module 60 includes:
  • the standardized comprehensive correlation coefficient matrix obtaining unit is used to sequentially determine whether each element in the comprehensive correlation coefficient matrix meets the preset threshold according to the comprehensive correlation coefficient matrix; if so, set the corresponding element value to 1; if not, set the corresponding The element value is set to 0; and a standardized comprehensive correlation coefficient matrix is obtained;
  • the input characteristic parameter determination unit is used to determine the input characteristic parameters required for power prediction of each wind power generator according to the preset screening rules according to the standardized comprehensive correlation coefficient matrix.
  • the input feature parameter determination unit is also used to: select each element of each column in the standardized comprehensive correlation coefficient matrix according to the preset screening rules to filter out the feature elements that meet the preset screening conditions, and use them as the corresponding columns
  • each column vector in the standardized comprehensive correlation coefficient matrix represents a characteristic parameter vector of a wind turbine.
  • the default filtering rules are:
  • the characteristic data at least include the wind speed, wind direction and temperature of the surrounding environment of each wind generator, the blade angle of each wind generator, and the flow velocity, flow direction and temperature of sea water around each wind generator.
  • the multiple prediction models at least include: BP neural network model, convolutional neural network model, gated recurrent neural network model and CNN-GRU model.
  • the ultra-short-term wind power forecasting device provided in the embodiment of the present application can execute the ultra-short-term wind power forecasting method provided in any embodiment of the present application, and has corresponding functional modules and beneficial effects for executing the method.
  • Embodiments of the present application provide a method and device for ultra-short-term wind power prediction, so as to realize power prediction of offshore wind power and improve prediction accuracy.

Landscapes

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

Abstract

本申请实施例公开了一种超短期风电功率预测方法和装置。该方法包括：获取每台风力电机的历史功率数据和影响风电功率的各种特征数据；根据每台风力电机的历史功率数据和影响风电功率的各种特征数据建立每台风力发电机的特征矩阵；根据每台风力发电机的特征矩阵计算各台风力发电机之间的多种相关系数；根据各台风力发电机之间的多种相关系数构建各台风力发电机之间的综合相关系数矩阵；根据各台风力发电机之间的综合相关系数矩阵确定每台风力发电机功率预测所需的输入特征参数；根据每台风力发电机功率预测所需的输入特征参数建立多种预测模型；根据各预测模型训练得到对应的功率预测结果，并采用聚类算法求解得到最优功率预测结果。

Description

超短期风电功率预测方法和装置
本申请要求在2022年01月13日提交中国专利局、申请号为202210034461.5的中国专利申请的优先权，以上申请的全部内容通过引用结合在本申请中。
技术领域
本申请实施例涉及风电功率预测技术领域，例如涉及一种超短期风电功率预测方法和装置。
背景技术
传统的风电功率预测方法分为物理模型和统计模型，随着大数据技术的发展，人工智能技术被应用于风电预测中。但针对海上风电功率预测的研究却较少，由于海水比热容大、风浪、风机尾流的影响明显，陆地风电场的预测方法难以适应海上风电预测。近年来深度学习模型也逐渐被应用于海上风电功率预测中，但大多是将风电场作为一个整体进行预测，并未考虑风电场内风机尾流的影响。因此，急需一种应用于海上风电功率预测的预测方法。
发明内容
本申请提供一种超短期风电功率预测方法和装置。
第一方面，本申请实施例提供了一种超短期风电功率预测方法，超短期风电功率预测方法包括：
获取每台风力电机的历史功率数据和影响风电功率的各种特征数据；
根据每台风力电机的历史功率数据和影响风电功率的各种特征数据建立每台风力发电机的特征矩阵；
根据每台风力发电机的特征矩阵，计算各台风力发电机之间的多种相关系数；
根据各台风力发电机之间的多种相关系数，构建各台风力发电机之间的综合相关系数矩阵；
根据各台风力发电机之间的综合相关系数矩阵，确定每台风力发电机功率预测所需的输入特征参数；
根据每台风力发电机功率预测所需的输入特征参数，建立多种预测模型；并根据各预测模型训练得到对应的功率预测结果；
根据各功率预测结果，采用聚类算法求解得到最优功率预测结果。
第二方面，本申请实施例还提供了一种超短期风电功率预测装置，该超短期风电功率预测装置包括：
历史功率数据获取模块，设置为获取每台风力电机的历史功率数据；
特征数据获取模块，设置为获取每台风力电机的影响风电功率的各种特征数据；
特征矩阵建立模块，设置为根据每台风力电机的历史功率数据和影响风电功率的各种特征数据建立每台风力发电机的特征矩阵；
多种相关系数计算模块，设置为根据每台风力发电机的特征矩阵，计算各台风力发电机之间的多种相关系数；
综合相关系数矩阵构建模块，设置为根据各台风力发电机之间的多种相关系数，构建各台风力发电机之间的综合相关系数矩阵；
输入特征参数确定模块，设置为根据各台风力发电机之间的综合相关系数矩阵，确定每台风力发电机功率预测所需的输入特征参数；
多种预测模型建立模块，设置为根据每台风力发电机功率预测所需的输入特征参数，建立多种预测模型；
功率预测结果获得模块，设置为根据各预测模型训练得到对应的功率预测结果；
最优功率预测结果求解模块，设置为根据各功率预测结果，采用聚类算法求解得到最优功率预测结果。
附图说明
图1是本申请实施例一中的一种超短期风电功率预测方法的流程图；
图2是本申请实施例二中的一种超短期风电功率预测方法的流程图；
图3是本申请实施例三中的一种超短期风电功率预测装置的结构框图。
具体实施方式
下面结合附图和实施例对本申请进行说明。
实施例一
图1为本申请实施例一中提供的一种超短期风电功率预测方法的流程图，本实施例可适用于在海上风电管理平台中，实现提高超短期风电功率预测精度的方法，该方法可以由超短期风电功率预测装置来执行，该装置可以由软件和/或硬件的方式来实现，该装置可配置于管理平台的服务器中，参考图1，包括如 下步骤：
步骤110、获取每台风力电机的历史功率数据和影响风电功率的各种特征数据；
通常，海上风电功率预测受到众多因素的影响，例如大气温度、风速、海水温度、风机轮翼处风速、风机叶片与风向的夹角等。虽然由众多的因数都会影响海上风机功率的输出，但每个因素的影响程度不一，有的因素甚至可以直接忽略。因此，影响风电功率的各种特征数据是指除去可以直接忽略的因素以外的因素。
其中，每台风力电机的历史功率数据可以从海上风电场的数据采集与监视控制***(Supervisory Control And Data Acquisition，SCADA)获取各台风力发电机的功率时间序列。每台风力电机的影响风电功率的各种特征数据可以通过数据采集设备或传感器等进行采集，例如，温度可以采用温度传感器采集。将采集到的每台风力电机的历史功率数据和影响风电功率的各种特征数据都发送到数据存储设备进行存储。因而，服务器可以直接从数据存储设备获取每台风力电机的历史功率数据和影响风电功率的各种特征数据。
步骤120、根据每台风力电机的历史功率数据和影响风电功率的各种特征数据建立每台风力发电机的特征矩阵；
其中，在获取每台风力电机的影响风电功率的各种特征数据后，为了统一样本数据的统计分布性，方便后续处理，需要对各种特征数据做归一化处理，即将每台风力发电机的周围环境风速、风向和温度，以及每台风力发电机的周围海水的流速、流向和温度都进行归一化处理。并将归一化处理后的每台风力发电机的各种特征数据与对应的历史功率数据拼接为每台风力发电机的特征矩阵X。
其中，归一化处理采用min-max归一化方法，将数据归一化至[0，1]，公式为：
Figure PCTCN2022144103-appb-000001
其中，x为各种特征数据向量，x min为对应的特征数据向量的最小值，x max为对应的特征数据向量的最大值。
步骤130、根据每台风力发电机的特征矩阵，计算各台风力发电机之间的多种相关系数；
其中，各台风力发电机之间存在一定的相关性。因此，基于各台风力发电 机的特征矩阵，计算得到的各台风力发电机之间的相关系数可以反应各台风力发电机之间在不同的特征或维度方面的相关性。
其中，相关系数可以采用多种方式去表征，例如，Pearson相关系数可以根据波形来确定各台风力发电机之间的相关性；Spearman相关系数可以根据单调性来确定各台风力发电机之间的相关性；R2相关系数可以说明各台风力发电机相互间的影响程度；欧氏距离用来确定两个海上风力发电机的特征间的关联程度。此外，相关系数还可以包括其他种类，可根据实际情况进行设置，在此不做限定。
步骤140、根据各台风力发电机之间的多种相关系数，构建各台风力发电机之间的综合相关系数矩阵；
其中，各台风力发电机之间的多种相关系数可以从多种不同的角度反应各台风力发电机之间特征的相关性。不同相关系数计算方法表现出不同的特征，考虑到不同的相关系数的侧重点不同，因此，为了提高预测的精度，在评估海上风力发电机间的相关性时，结合多种相关系数，并对各种相关系数进行加权，构建各台风力发电机之间的综合相关系数矩阵，以更为准确的反映海上风力发电机组之间特征的相互影响，例如，风力发电机组之间尾流的相互影响。
步骤150、根据各台风力发电机之间的综合相关系数矩阵，确定每台风力发电机功率预测所需的输入特征参数；
其中，每台风力发电机功率预测所需的输入特征参数是各种功率预测模型进行训练的基础参数，为了提高每台风力发电机功率预测的准确性，要确保输入到功率预测模型进行训练的输入特征参数为影响风力发电机功率变化的重要因素特征。因此，在根据预测模型训练之前，需要对预测所需的输入特征参数进行特征筛选，即根据各台风力发电机之间的综合相关系数矩阵可以确定每台风力发电机功率预测最终需要输入的特征参数。
步骤160、根据每台风力发电机功率预测所需的输入特征参数，建立多种预测模型；并根据各预测模型训练得到对应的功率预测结果；
例如，基于每台风力发电机功率预测所需的输入特征参数，对每台风力发电机建立多种预测模型，并将每台风力发电机功率预测所需的输入特征参数输入到每台风力发电机的各个预测模型，进行训练，得到对应的风力发电机的各个预测模型的功率预测结果。其中，针对每台风力发电机都建立多种预测模型，并根据各个预测模型进行训练，是为了得到每台风力发电机在多种预测模型下的训练情况，以便后续将多种预测模型的预测结果进行耦合，实现将多种预测 模型的优点进行耦合，以得到最优的预测结果，从而提高功率预测的精度和可靠性。
步骤170、根据各功率预测结果，采用聚类算法求解得到最优功率预测结果。
其中，将多种预测模型的预测结果采用聚类算法进行耦合求解最优预测结果，充分保留了不同预测模型的优点，采用聚类算法与其有机结合使得预测模型的预测精度得到了进一步的提升，具有重要的实际工程意义。
在本实施例中，该超短期风电功率预测方法的工作原理为：首选先，获取每台风力电机的历史功率数据和影响风电功率的各种特征数据；根据每台风力电机的历史功率数据和影响风电功率的各种特征数据建立每台风力发电机的特征矩阵；根据每台风力发电机的特征矩阵，计算各台风力发电机之间的多种相关系数；根据各台风力发电机之间的多种相关系数，构建各台风力发电机之间的综合相关系数矩阵；根据各台风力发电机之间的综合相关系数矩阵，确定每台风力发电机功率预测所需的输入特征参数；根据每台风力发电机功率预测所需的输入特征参数，建立多种预测模型；并根据各预测模型训练得到对应的功率预测结果；根据各功率预测结果，采用聚类算法求解得到最优功率预测结果。由此可知，通过全面地考虑影响海上风电预测的各种特征数据，并结合每台风力电机的历史功率数据生成每台风力发电机的特征矩阵，基于特征矩阵构建各台风力发电机之间的综合相关系数矩阵并根据综合相关系数矩阵确定最终输入到预测模型的输入特征参数，从而提高预测的精度。且通过聚类算法将多种预测模型的预测结果进行耦合，在充分保留不同预测模型的优点的同时，将聚类算法与各种预测模型有机结合使得预测精度得到了进一步的提升。
本实施例，通过提供一种超短期风电功率预测方法，该超短期风电功率预测方法包括：获取每台风力电机的历史功率数据和影响风电功率的各种特征数据；根据每台风力电机的历史功率数据和影响风电功率的各种特征数据建立每台风力发电机的特征矩阵；根据每台风力发电机的特征矩阵，计算各台风力发电机之间的多种相关系数；根据各台风力发电机之间的多种相关系数，构建各台风力发电机之间的综合相关系数矩阵；根据各台风力发电机之间的综合相关系数矩阵，确定每台风力发电机功率预测所需的输入特征参数；根据每台风力发电机功率预测所需的输入特征参数，建立多种预测模型；并根据各预测模型训练得到对应的功率预测结果；根据各功率预测结果，采用聚类算法求解得到最优功率预测结果。通过该方法可以实现对海上风力发电机的功率进行预测，且全面地考虑影响海上风电预测的各种特征数据，并结合每台风力电机的历史 功率数据生成每台风力发电机的特征矩阵，基于特征矩阵构建各台风力发电机之间的综合相关系数矩阵并根据综合相关系数矩阵确定最终输入到预测模型的输入特征参数，从而提高预测的精度。
在上述实施例的基础上，在一实施例中，特征数据至少包括每台风力发电机的周围环境风速、风向和温度，每台风力发电机的叶片角度，以及每台风力发电机的周围海水的流速、流向和温度。
其中，每台风力发电机的周围环境风速和风向会影响海上风力发电机的尾流。而海上风力发电机的尾流因素会在一定程度上影响风力发电机的功率预测，因此，为了提高风力发电机的功率预测的精度，需要考虑每台风力发电机的周围环境风速和风向这两个特征数据。
在一实施例中，在获取到每台风力电机的影响风电功率的各种特征数据后，需要对部分数据进行预处理，以提高数据的准确性。例如，对于每台风力发电机的周围环境的风速和风向数据，假设风速为v speed，风向为v windD，将其分解为经向风速v windV和纬向风速v windU，其计算公式如下：
Figure PCTCN2022144103-appb-000002
对于每台风力发电机的周围海水的流向数据，假设海水流向数据为v seaD，将其分解为流向余弦v seaD cos和流向正弦v seaD sin，其计算公式如下：
Figure PCTCN2022144103-appb-000003
对于每台风力发电机的叶片角度数据，假设风力发电机的叶片角度为v seaD，将其分解为叶片角度正弦D Ysin和余弦D Ycos，其计算公式如下：
Figure PCTCN2022144103-appb-000004
实施例二
图2是本申请实施例二中提供的一种超短期风电功率预测方法的流程图。在上述实施例一的基础上，在一实施例中，多种相关系数至少包括：Pearson相关系数、Spearman相关系数、R2系数和欧式距离。
其中，对各台风力发电机之间的Pearson相关系数的计算公式为：
Figure PCTCN2022144103-appb-000005
其中，对各台风力发电机之间的Spearman相关系数的计算公式为：
Figure PCTCN2022144103-appb-000006
其中，对各台风力发电机之间的R2系数的计算公式为：
Figure PCTCN2022144103-appb-000007
其中，对各台风力发电机之间的欧式距离的计算公式为：
Figure PCTCN2022144103-appb-000008
在一实施例中，参考图2，该超短期风电功率预测方法包括如下步骤：
步骤210、获取每台风力电机的历史功率数据和影响风电功率的各种特征数据；
步骤220、根据每台风力电机的历史功率数据和影响风电功率的各种特征数据建立每台风力发电机的特征矩阵；
步骤230、根据每台风力发电机的特征矩阵，计算各台风力发电机之间的多种相关系数；
步骤240、根据各台风力发电机之间的多种相关系数，构建各台风力发电机之间的综合相关系数式；
其中，在一实施例中，各台风力发电机之间的综合相关系数式为：
δ＝α 0Corr p+α 1Corr s+α 2Corr R2+α 3Corr d
其中，Corr p为各台风力发电机之间的Pearson相关系数；α 0为Corr p所对应的系数；Corr s为各台风力发电机之间的Spearman相关系数；α 1为Corr s所对应的系数；Corr R2为各台风力发电机之间的R2系数；α 2为Corr R2所对应的系数；Corr d为各台风力发电机之间的欧式距离；α 3为Corr d所对应的系数；
在一实施例中，各种相关系数所对应的系数α i的计算公式为：
Figure PCTCN2022144103-appb-000009
其中，β Q、β 1、β 2、β 3的计算公式分别为：
Figure PCTCN2022144103-appb-000010
Figure PCTCN2022144103-appb-000011
Figure PCTCN2022144103-appb-000012
Figure PCTCN2022144103-appb-000013
其中，i的取值为0，1，2，3。
示例性的，设矩阵A为：
Figure PCTCN2022144103-appb-000014
其中，β 3为矩阵A的行列式，β 0、β 1、β 2为矩阵A的行列式的子行列式。通过系数α i的计算公式，以及β 0、β 1、β 2和β 3的计算公式计算各种相关系数所对应的系数α i的方法可以提高计算的简便性。
步骤250、根据各台风力发电机之间的综合相关系数式计算各台风力发电机之间的综合相关系数；并基于各台风力发电机之间的综合相关系数形成各台风力发电机之间的综合相关系数矩阵。
其中，各台风力发电机之间的综合相关系数矩阵为：
Figure PCTCN2022144103-appb-000015
其中，δ ij为海上风力发电机i和j之间的综合相关系数。
步骤260、依次判断综合相关系数矩阵中的各个元素是否满足预设阈值；若是，则将对应的元素值置1；若否，则将对应的元素值置0；并得到标准化的综合相关系数矩阵。
例如，依次判断综合相关系数矩阵Z中的各个元素δ ij是否超过预设阈值k，若超过，则将对应的元素值置为1；若未超过，则将对应的元素值置为0，由此将综合相关系数矩阵Z中的所有元素的值标准化为只有1和/或0的标准化的综合相关系数矩阵Z*。其中，标准化的综合相关系数矩阵Z*中的每一列表示一台风力发电机，且每一列中元素值为1的元素对应该列风力发电机的一个特征数据，即每一列中元素值为1的个数表示该列风力发电机在预测时所用到的风机特征 的总数。换句话说，通过将综合相关系数矩阵按照预设阈值进行标准化处理后，相当于对每台风力发电机的预测模型所需的输入特征进行了一次筛选，即根据每台风力发电机的实际情况选择其对应所需的特征输入到预测模型进行功率预测(因为每台风力发电机由于周围风速、风向、海水流速、流向等影响因素或影响程度可能不同，对其功率预测的影响也就不同，因此，每台风力发电机进行功率预测所需考虑的特征数据也不一定相同)，从而可以提高预测的精度。
其中，预设阈值k可根据实际情况进行设置，在此不做限定。
步骤270、根据标准化的综合相关系数矩阵，按照预设筛选规则确定每台风力发电机功率预测所需的输入特征参数。
其中，将综合相关系数矩阵按照预设阈值进行标准化处理，只是对每台风力发电机的预测模型所需的输入特征进行了初次筛选，为了提高输入特征筛选的可靠性和提高预测的精度，按照预设筛选规则对标准化的综合相关系数矩阵进行二次筛选，以确定每台风力发电机功率预测最终所需的输入特征参数。
在一实施例中，将标准化的综合相关系数矩阵中的每一列的各个元素，按照预设筛选规则筛选出满足预设筛选条件的特征元素，并作为对应列的风力发电机的功率预测所需的输入特征参数；
其中，标准化的综合相关系数矩阵中的每一列向量表示一台风力发电机的特征参数向量。
示例性的，假设标准化的综合相关系数矩阵中的第i列对应第n台风力发电机，设第i列为Y＝[x 1，x 2，...，x n]，其中，Y中的各个元素值为0或1，Y中的各个元素值为1的元素表示第n台风力发电机的一个特征。则将Y中的各个元素按照预设筛选规则筛选出满足预设筛选条件的特征元素，假设Y中的元素x 2的值为1，且满足预设筛选条件，则将元素x 2作为第n台风力发电机的功率预测所需的一个输入特征参数。按照预设筛选规则筛选出列矩阵Y中所有满足预设条件的元素，即筛选出第n台风力发电机的功率预测所需的所有输入特征参数。
在一实施例中，预设筛选规则为：
计算每台风力发电机在标准化的综合相关系数矩阵中对应列的每个特征元素与对应的风力电机的历史功率数据之间的Pearson相关系数；
根据Pearson相关系数的计算结果，判断是否保留对应的特征元素。
示例性的，仍以上述第n台风力发电机为例，且设第n台风力发电机的历史功率数据为P，分别计算列矩阵Y中的各个元素与第n台风力发电机的历史功率数据P之间的Pearson相关系数；根据Pearson相关系数的计算结果，判断 是否保留对应的特征元素。若该元素与历史功率数据的Pearson相关系数表现为强相关性则保留该元素，即保留该元素对应的风力发电机的特征元素；若否，则移除该元素，即移除该元素对应的风力发电机的特征元素。
其中，Pearson相关系数表现为强相关性是指Pearson相关系数的绝对值达到预设系数阈值及以上则为具有相关性，且Pearson相关系数的绝对值系数越大，相关性越强。其中，预设系数阈值可以为0.6，可根据实际情况进行设置，在此不做限定。
步骤280、根据每台风力发电机功率预测所需的输入特征参数，建立多种预测模型；并根据各预测模型训练得到对应的功率预测结果；
步骤290、根据各功率预测结果，采用聚类算法求解得到最优功率预测结果。
在一实施例中，多种预测模型至少包括：BP神经网络模型、卷积神经网络模型(Convolutional Neural Networks，CNN)、门控循环神经网络模型(Recurrent Neural Network，RNN)和CNN-GRU模型。
例如，根据每台风力发电机功率预测所需的输入特征参数，分别建立每台风力发电机的BP神经网络模型、卷积神经网络模型、门控循环神经网络模型和CNN-GRU模型。
其中，每台风力发电机所对应的BP神经网络模型、卷积神经网络模型、门控循环神经网络模型和CNN-GRU模型的建立和训练方法分别如下：
对于BP神经网络模型：首先，建立三层BP神经网络，示例性的，将神经元数量依次设置为64，128，256，以构建BP神经网络模型；然后，根据上述最终确定的风力发电机的输入特征参数，训练BP神经网络模型，最后输出海上风力发电机的预测功率y predict1。
对于卷积神经网络模型：首先，建立两层CNN模型，示例性的，将神经元数量依次设置为4和8，以构建卷积神经网络模型；然后，根据上述最终确定的风力发电机的输入特征参数，训练卷积神经网络模型，最后输出海上风力发电机的预测功率y predicet2。
对于门控循环神经网络模型：首先，建立两层GRU模型，示例性的，将神经元数量依次设置为4和8，以构建门控循环神经网络模型；然后，根据上述最终确定的风力发电机的输入特征参数，训练门控循环神经网络模型，最后输出海上风力发电机的预测功率y predict3。
对于CNN-GRU模型：首先，建立一层CNN和两层GRU的组合模型，示例性的，将神经元数量依次设置为4，8，4，以构建CNN-GRU模型；然后，根 据上述最终确定的风力发电机的输入特征参数，训练CNN-GRU模型，最后输出海上风力发电机的预测功率y predict4。
在一实施例中，根据各种预测模型得到的功率预测结果，采用粒子群算法(Particle swarm optimization，PSO)求解得到最优功率预测结果。求解过程如下：
首先，将4种预测模型的预测结果进行加权，并求和得到综合模型的最终预测结果，其关系式如下：
y predict＝θ 1·y predict1+θ 2·y predicet2+θ 3·y predict3+θ 4·y predict4
其中，y predict为综合模型的最终预测结果；θ i为对应各种预测模型预测结果的权重，满足：
Figure PCTCN2022144103-appb-000016
然后，以4种预测模型的预测结果的均方根误差为最小目标函数，采用粒子群算法(Particle swarm optimization，PSO)求解θ i的最优值，以求解得到最优功率预测结果。
其次，执行PSO算法的粒子位置与速度更新公式，得到新的粒子的位置与速度；
接着，计算新粒子的适应度值，若适应度值优于其个体最优位置P b，则将其值赋予P b；若适应度值优于其全局最优位置g b，则将其值赋予g b；
最后，根据预设迭代次数重复执行更新粒子的位置与速度，以及计算更新后的粒子的适应度值及判断的操作，直到满足预设迭代次数为止；并将满足预设迭代次数时的结果作为θ i的最优值，以求解得到最优功率预测结果。
其中，目标函数的表达式为：
Figure PCTCN2022144103-appb-000017
其中，f obj为优化目标；n为样本数量，即矩阵中的所有元素的个数；i为矩阵中的第个元素；y predict为预测值，是个矩阵；y truth为真实值。
其中，粒子群算法的更新公式如下：
Figure PCTCN2022144103-appb-000018
其中，
Figure PCTCN2022144103-appb-000019
其中，x i和v i分别为粒子的速度和位置；c 1和c 2是0到1之间的随机数；c 1和 c 2为学习因子；ω为惯性因子。
其中，执行PSO算法的粒子位置与速度更新公式，得到新的粒子的位置与速度，根据种群粒X 1(t+1)，更新种群P的更新公式如下：
P(i)＝L best←min(f obj(P(i))，f obj(X 1(t+1)(i)))，i∈[1，N]
其中，L best是粒子P(i)和X 1(t+1)(i)根据上式计算出的最小目标值对应的粒子种群。
实施例三
图3是本申请实施例三中提供的一种超短期风电功率预测装置的结构框图。本申请实施例三提供了一种超短期风电功率预测装置，参考图3，该装置100包括：
历史功率数据获取模块10，用于获取每台风力电机的历史功率数据；
特征数据获取模块20，用于获取每台风力电机的影响风电功率的各种特征数据；
特征矩阵建立模块30，用于根据每台风力电机的历史功率数据和影响风电功率的各种特征数据建立每台风力发电机的特征矩阵；
多种相关系数计算模块40，用于根据每台风力发电机的特征矩阵，计算各台风力发电机之间的多种相关系数；
综合相关系数矩阵构建模块50，用于根据各台风力发电机之间的多种相关系数，构建各台风力发电机之间的综合相关系数矩阵；
输入特征参数确定模块60，用于根据各台风力发电机之间的综合相关系数矩阵，确定每台风力发电机功率预测所需的输入特征参数；
多种预测模型建立模块70，用于根据每台风力发电机功率预测所需的输入特征参数，建立多种预测模型；
功率预测结果获得模块80，用于根据各预测模型训练得到对应的功率预测结果；
最优功率预测结果求解模块90，用于根据各功率预测结果，采用聚类算法求解得到最优功率预测结果。
本实施例，通过提供一种超短期风电功率预测装置，该超短期风电功率预测装置包括：历史功率数据获取模块，用于获取每台风力电机的历史功率数据；特征数据获取模块，用于获取每台风力电机的影响风电功率的各种特征数据；特征矩阵建立模块，用于根据每台风力电机的历史功率数据和影响风电功率的 各种特征数据建立每台风力发电机的特征矩阵；多种相关系数计算模块，用于根据每台风力发电机的特征矩阵，计算各台风力发电机之间的多种相关系数；综合相关系数矩阵构建模块，用于根据各台风力发电机之间的多种相关系数，构建各台风力发电机之间的综合相关系数矩阵；输入特征参数确定模块，用于根据各台风力发电机之间的综合相关系数矩阵，确定每台风力发电机功率预测所需的输入特征参数；多种预测模型建立模块，用于根据每台风力发电机功率预测所需的输入特征参数，建立多种预测模型；功率预测结果获得模块，用于根据各预测模型训练得到对应的功率预测结果；最优功率预测结果求解模块，用于根据各功率预测结果，采用聚类算法求解得到最优功率预测结果。通过该装置可以实现对海上风力发电机的功率进行预测，且全面地考虑影响海上风电预测的各种特征数据，并结合每台风力电机的历史功率数据生成每台风力发电机的特征矩阵，基于特征矩阵构建各台风力发电机之间的综合相关系数矩阵并根据综合相关系数矩阵确定最终输入到预测模型的输入特征参数，从而提高预测的精度。
在一实施例中，多种相关系数至少包括：Pearson相关系数、Spearman相关系数、R2系数和欧式距离；
综合相关系数矩阵构建模块50，包括：
综合相关系数式构建单元，用于根据各台风力发电机之间的多种相关系数，构建各台风力发电机之间的综合相关系数式；
综合相关系数计算单元，用于根据各台风力发电机之间的综合相关系数式计算各台风力发电机之间的综合相关系数；
综合相关系数矩阵形成单元，用于基于各台风力发电机之间的综合相关系数形成各台风力发电机之间的综合相关系数矩阵。
在一实施例中，各台风力发电机之间的综合相关系数式为：
δ＝α 0Corr p+α 1Corr s+α 2Corr R2+α 3Corr d
其中，Corr p为各台风力发电机之间的Pearson相关系数；α 0为Corr p所对应的系数；Corr S为各台风力发电机之间的Spearman相关系数；α 1为Corr S所对应的系数；Corr R2为各台风力发电机之间的R2系数；α 2为Corr R2所对应的系数；Corr d为各台风力发电机之间的欧式距离；α 3为Corr d所对应的系数；
各台风力发电机之间的综合相关系数矩阵为：
Figure PCTCN2022144103-appb-000020
其中，δ ij为海上风力发电机i和j之间的综合相关系数。
在一实施例中，各种相关系数所对应的系数α i的计算公式为：
Figure PCTCN2022144103-appb-000021
其中，β 0、β 1、β 2、β 3的计算公式分别为：
Figure PCTCN2022144103-appb-000022
Figure PCTCN2022144103-appb-000023
Figure PCTCN2022144103-appb-000024
Figure PCTCN2022144103-appb-000025
其中，i的取值为0，1，2，3。
在一实施例中，输入特征参数确定模块60，包括：
标准化的综合相关系数矩阵获得单元，用于根据综合相关系数矩阵依次判断综合相关系数矩阵中的各个元素是否满足预设阈值；若是，则将对应的元素值置1；若否，则将对应的元素值置0；并得到标准化的综合相关系数矩阵；
输入特征参数确定单元，用于根据标准化的综合相关系数矩阵，按照预设筛选规则确定每台风力发电机功率预测所需的输入特征参数。
在一实施例中，输入特征参数确定单元还用于：将标准化的综合相关系数矩阵中的每一列的各个元素，按照预设筛选规则筛选出满足预设筛选条件的特征元素，并作为对应列的风力发电机的功率预测所需的输入特征参数；
其中，标准化的综合相关系数矩阵中的每一列向量表示一台风力发电机的特征参数向量。
在一实施例中，预设筛选规则为：
计算每台风力发电机在标准化的综合相关系数矩阵中对应列的每个特征元素与对应的风力电机的历史功率数据之间的Pearson相关系数；
根据Pearson相关系数的计算结果，判断是否保留对应的特征元素。
在一实施例中，特征数据至少包括每台风力发电机的周围环境风速、风向和温度，每台风力发电机的叶片角度，以及每台风力发电机的周围海水的流速、流向和温度。
在一实施例中，多种预测模型至少包括：BP神经网络模型、卷积神经网络模型、门控循环神经网络模型和CNN-GRU模型。
本申请实施例所提供的超短期风电功率预测装置可执行本申请任意实施例所提供的超短期风电功率预测方法，具备执行方法相应的功能模块和有益效果。
本申请实施例提供一种超短期风电功率预测方法和装置，以实现对海上风电的功率预测，并提高预测的精度。
本领域技术人员会理解，本申请不限于这里所述的特定实施例，对本领域技术人员来说能够进行各种变化、重新调整和替代而不会脱离本申请的保护范围。因此，虽然通过以上实施例对本申请进行了说明，但是本申请不仅仅限于以上实施例，在不脱离本申请构思的情况下，还可以包括更多其他等效实施例，而本申请的范围由所附的权利要求范围决定。

Claims (10)

  1. 一种超短期风电功率预测方法，包括：
    获取每台风力电机的历史功率数据和影响风电功率的各种特征数据；
    根据每台风力电机的历史功率数据和影响风电功率的各种特征数据建立每台风力发电机的特征矩阵；
    根据每台风力发电机的特征矩阵，计算各台风力发电机之间的多种相关系数；
    根据各台风力发电机之间的多种相关系数，构建各台风力发电机之间的综合相关系数矩阵；
    根据各台风力发电机之间的综合相关系数矩阵，确定每台风力发电机功率预测所需的输入特征参数；
    根据每台风力发电机功率预测所需的输入特征参数，建立多种预测模型；并根据各预测模型训练得到对应的功率预测结果；
    根据各功率预测结果，采用聚类算法求解得到最优功率预测结果。
  2. 根据权利要求1所述的超短期风电功率预测方法，其中，所述多种相关系数至少包括：Pearson相关系数、Spearman相关系数、R2系数和欧式距离；
    所述根据各台风力发电机之间的多种相关系数，构建各台风力发电机之间的综合相关系数矩阵，包括：
    根据各台风力发电机之间的多种相关系数，构建各台风力发电机之间的综合相关系数式；
    根据各台风力发电机之间的综合相关系数式计算各台风力发电机之间的综合相关系数；并基于各台风力发电机之间的综合相关系数形成各台风力发电机之间的综合相关系数矩阵。
  3. 根据权利要求2所述的超短期风电功率预测方法，其中，所述各台风力发电机之间的综合相关系数式为：
    δ＝α 0Corr p+α 2Corr s+α 2Corr R2+α 3Corr d
    其中，Corr p为各台风力发电机之间的Pearson相关系数；α 0为Corr p所对应的系数；Corr s为各台风力发电机之间的Spearman相关系数；α 1为Corr s所对应的系数；Corr R2为各台风力发电机之间的R2系数；α 2为Corr R2所对应的系数；Corr d为各台风力发电机之间的欧式距离；α 3为Corr d所对应的系数；
    所述各台风力发电机之间的综合相关系数矩阵为：
    Figure PCTCN2022144103-appb-100001
    其中，δ ij为海上风力发电机i和j之间的综合相关系数。
  4. 根据权利要求3所述的超短期风电功率预测方法，其中，各种相关系数所对应的系数α i的计算公式为：
    Figure PCTCN2022144103-appb-100002
    其中，β 0、β 1、β 2、β 3的计算公式分别为：
    Figure PCTCN2022144103-appb-100003
    Figure PCTCN2022144103-appb-100004
    Figure PCTCN2022144103-appb-100005
    Figure PCTCN2022144103-appb-100006
    其中，i的取值为0，1，2，3。
  5. 根据权利要求1所述的超短期风电功率预测方法，其中，所述根据各台风力发电机之间的综合相关系数矩阵，确定每台风力发电机功率预测所需的输入特征参数，包括：
    依次判断所述综合相关系数矩阵中的各个元素是否满足预设阈值；若是，则将对应的元素值置1；若否，则将对应的元素值置0；并得到标准化的综合相关系数矩阵；
    根据所述标准化的综合相关系数矩阵，按照预设筛选规则确定每台风力发电机功率预测所需的输入特征参数。
  6. 根据权利要求5所述的超短期风电功率预测方法，其中，所述根据所述标准化的综合相关系数矩阵，按照预设筛选规则确定每台风力发电机功率预测所需的输入特征参数，包括：
    将所述标准化的综合相关系数矩阵中的每一列的各个元素，按照预设筛选规则筛选出满足预设筛选条件的特征元素，并作为对应列的风力发电机的功率 预测所需的输入特征参数；
    其中，所述标准化的综合相关系数矩阵中的每一列向量表示一台风力发电机的特征参数向量。
  7. 根据权利要求6所述的超短期风电功率预测方法，其中，所述预设筛选规则为：
    计算每台风力发电机在所述标准化的综合相关系数矩阵中对应列的每个特征元素与对应的风力电机的历史功率数据之间的Pearson相关系数；
    根据Pearson相关系数的计算结果，判断是否保留对应的特征元素。
  8. 根据权利要求1所述的超短期风电功率预测方法，其中，所述特征数据至少包括每台风力发电机的周围环境风速、风向和温度，每台风力发电机的叶片角度，以及每台风力发电机的周围海水的流速、流向和温度。
  9. 根据权利要求1所述的超短期风电功率预测方法，其中，所述多种预测模型至少包括：BP神经网络模型、卷积神经网络模型、门控循环神经网络模型和CNN-GRU模型。
  10. 一种超短期风电功率预测装置，包括：
    历史功率数据获取模块，设置为获取每台风力电机的历史功率数据；
    特征数据获取模块，设置为获取每台风力电机的影响风电功率的各种特征数据；
    特征矩阵建立模块，设置为根据每台风力电机的历史功率数据和影响风电功率的各种特征数据建立每台风力发电机的特征矩阵；
    多种相关系数计算模块，设置为根据每台风力发电机的特征矩阵，计算各台风力发电机之间的多种相关系数；
    综合相关系数矩阵构建模块，设置为根据各台风力发电机之间的多种相关系数，构建各台风力发电机之间的综合相关系数矩阵；
    输入特征参数确定模块，设置为根据各台风力发电机之间的综合相关系数矩阵，确定每台风力发电机功率预测所需的输入特征参数；
    多种预测模型建立模块，设置为根据每台风力发电机功率预测所需的输入特征参数，建立多种预测模型；
    功率预测结果获得模块，设置为根据各预测模型训练得到对应的功率预测结果；
    最优功率预测结果求解模块，设置为根据各功率预测结果，采用聚类算法求解得到最优功率预测结果。
PCT/CN2022/144103 2022-01-13 2022-12-30 超短期风电功率预测方法和装置 WO2023134478A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202210034461.5 2022-01-13
CN202210034461.5A CN114048930B (zh) 2022-01-13 2022-01-13 一种超短期风电功率预测方法和装置

Publications (1)

Publication Number Publication Date
WO2023134478A1 true WO2023134478A1 (zh) 2023-07-20

Family

ID=80196387

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/144103 WO2023134478A1 (zh) 2022-01-13 2022-12-30 超短期风电功率预测方法和装置

Country Status (2)

Country Link
CN (1) CN114048930B (zh)
WO (1) WO2023134478A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117763314A (zh) * 2024-02-22 2024-03-26 安徽大学 一种基于时空相关性挖掘的风电功率预测方法

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114048930B (zh) * 2022-01-13 2022-05-17 广东电网有限责任公司揭阳供电局 一种超短期风电功率预测方法和装置
CN115204522B (zh) * 2022-09-13 2023-01-24 华北电力大学 区域发电功率预测方法、装置、设备及存储介质
CN116128130B (zh) * 2023-01-31 2023-10-24 广东电网有限责任公司 一种基于图神经网络的短期风能数据预测方法及装置

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180223814A1 (en) * 2017-02-07 2018-08-09 International Business Machines Corporation Reducing curtailment of wind power generation
CN110188939A (zh) * 2019-05-24 2019-08-30 上海电气风电集团有限公司 风电场的风功率的预测方法、***、设备和存储介质
CN113159452A (zh) * 2021-05-17 2021-07-23 河北工业大学 基于时空相关性的风电集群功率预测方法
CN113507118A (zh) * 2021-07-11 2021-10-15 湘潭大学 一种风电功率预测方法及***
CN114048930A (zh) * 2022-01-13 2022-02-15 广东电网有限责任公司揭阳供电局 一种超短期风电功率预测方法和装置

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2004285858A (ja) * 2003-03-19 2004-10-14 Mitsubishi Electric Corp 風力発電システムおよび風力発電機の制御方法
CN103268366B (zh) * 2013-03-06 2017-09-29 辽宁省电力有限公司电力科学研究院 一种适用于分散式风电场的组合风电功率预测方法
CN109102101B (zh) * 2017-06-21 2021-01-26 北京金风科创风电设备有限公司 风电场风速的预测方法和***
CN109146192B (zh) * 2018-09-03 2021-09-03 贵州电网有限责任公司 一种考虑风电机组运行工况的风电功率预测方法
CN110991743B (zh) * 2019-12-05 2022-08-19 河海大学 一种基于聚类分析和优化神经网络的风电功率短期组合预测方法
CN111310972B (zh) * 2020-01-17 2022-06-03 上海电力大学 一种考虑尾流效应的海上风电机组维护路径随机规划方法
CN112329979A (zh) * 2020-09-23 2021-02-05 燕山大学 一种基于自适应深度残差网络的超短期风电功率预测方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180223814A1 (en) * 2017-02-07 2018-08-09 International Business Machines Corporation Reducing curtailment of wind power generation
CN110188939A (zh) * 2019-05-24 2019-08-30 上海电气风电集团有限公司 风电场的风功率的预测方法、***、设备和存储介质
CN113159452A (zh) * 2021-05-17 2021-07-23 河北工业大学 基于时空相关性的风电集群功率预测方法
CN113507118A (zh) * 2021-07-11 2021-10-15 湘潭大学 一种风电功率预测方法及***
CN114048930A (zh) * 2022-01-13 2022-02-15 广东电网有限责任公司揭阳供电局 一种超短期风电功率预测方法和装置

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117763314A (zh) * 2024-02-22 2024-03-26 安徽大学 一种基于时空相关性挖掘的风电功率预测方法
CN117763314B (zh) * 2024-02-22 2024-05-17 安徽大学 一种基于时空相关性挖掘的风电功率预测方法

Also Published As

Publication number Publication date
CN114048930B (zh) 2022-05-17
CN114048930A (zh) 2022-02-15

Similar Documents

Publication Publication Date Title
WO2023134478A1 (zh) 超短期风电功率预测方法和装置
Yan et al. Forecasting the high penetration of wind power on multiple scales using multi-to-multi mapping
Wang et al. Wind power interval prediction based on improved PSO and BP neural network
CN111310889B (zh) 一种基于深度神经网络的蒸发波导剖面估计方法
CN106529719B (zh) 基于粒子群优化算法风速融合的风电功率预测的方法
WO2021073090A1 (zh) 基于强化学习的风电机组实时变桨距鲁棒控制***及方法
CN114322199B (zh) 基于数字孪生的通风***自主优化运行调控平台及方法
CN106650977A (zh) 应用于新建风电场的短期功率预测方法
Dimitrov et al. From SCADA to lifetime assessment and performance optimization: how to use models and machine learning to extract useful insights from limited data
CN107818340A (zh) 基于k值小波神经网络的二阶段空调负荷预测方法
Chen et al. Research on wind power prediction method based on convolutional neural network and genetic algorithm
CN112819189A (zh) 一种基于历史预测值的风力出力预测方法
CN113705922A (zh) 一种改进的超短期风电功率预测算法及模型建立方法
He et al. Similar day selecting based neural network model and its application in short-term load forecasting
CN111488974B (zh) 基于深度学习神经网络的海洋风能降尺度方法
CN113095547B (zh) 一种基于gra-lstm-ice模型的短期风功率预测方法
Bouhelal et al. Blade element momentum theory coupled with machine learning to predict wind turbine aerodynamic performances
CN107527093B (zh) 一种风电机组运行状态诊断方法及装置
CN116821610B (zh) 一种利用大数据优化风力发电效率的方法
CN112085272A (zh) 一种风功率预测方法
CN117096871A (zh) 一种基于时空分布的风电功率概率密度的预测方法
CN116316629A (zh) 一种考虑拓扑特征学习的数据驱动最优潮流计算方法
Wang et al. Rainfall-runoff simulation using simulated annealing wavelet bp neural networks
CN113673786A (zh) 基于rbf和lstm的风力发电机有效风速估计***及方法
CN114139783A (zh) 基于非线性加权组合的风电短期功率预测方法及装置

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22920108

Country of ref document: EP

Kind code of ref document: A1