CN110991743A - Wind power short-term combination prediction method based on cluster analysis and neural network optimization - Google Patents

Wind power short-term combination prediction method based on cluster analysis and neural network optimization Download PDF

Info

Publication number
CN110991743A
CN110991743A CN201911231533.XA CN201911231533A CN110991743A CN 110991743 A CN110991743 A CN 110991743A CN 201911231533 A CN201911231533 A CN 201911231533A CN 110991743 A CN110991743 A CN 110991743A
Authority
CN
China
Prior art keywords
neural network
wind
wind power
population
prediction model
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201911231533.XA
Other languages
Chinese (zh)
Other versions
CN110991743B (en
Inventor
王冰
陈桂儒
曹智杰
王绍平
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.)
Hohai University HHU
Original Assignee
Hohai University HHU
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 Hohai University HHU filed Critical Hohai University HHU
Priority to CN201911231533.XA priority Critical patent/CN110991743B/en
Publication of CN110991743A publication Critical patent/CN110991743A/en
Application granted granted Critical
Publication of CN110991743B publication Critical patent/CN110991743B/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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Electricity, gas 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

Abstract

The invention discloses a wind power short-term combination prediction method based on cluster analysis and an optimized neural network, which comprises the following steps of: s1: determining the influence factors of the wind power output power according to the wind power generated by the fan, the fan efficiency under wind directions with different wind speeds and the wake effect; s2: clustering the input samples according to a K-means clustering algorithm, and classifying the input samples; s3: establishing a BP neural network prediction model corresponding to each type of input sample, and simultaneously optimizing each BP neural network prediction model through a thought evolution algorithm; s4: and inputting the input sample into the corresponding optimized BP neural network prediction model, predicting the wind power, and obtaining a future fan output curve. The method utilizes a thought evolution algorithm to optimize the initial weight and the threshold, thereby not only identifying the wind conditions and respectively establishing a prediction model for each type, but also improving the wind power prediction speed and the prediction precision.

Description

Wind power short-term combination prediction method based on cluster analysis and neural network optimization
Technical Field
The invention relates to the technical field of wind power prediction, in particular to a wind power short-term combined prediction method based on cluster analysis and an optimized neural network.
Background
With the modernization of science and technology and industry, the utilization rate of renewable energy sources gradually rises. Wind energy is an important part of renewable energy, but wind power generation has great randomness, and large-scale development of the wind power generation is restricted. In order to more effectively utilize wind energy and reduce the influence of wind power generation on a power system, the power prediction method which is high in research precision and suitable for wind power has important significance on the development of the power industry.
The traditional BP neural network prediction method has the defects of easy falling into local optimization and slow convergence speed, and although some methods for combining the optimization algorithm with the neural network improve the accuracy and the convergence speed, the traditional BP neural network prediction method still has the problems of local convergence and long optimization time.
Disclosure of Invention
The purpose of the invention is as follows: the invention provides a wind power short-term combined prediction method based on cluster analysis and an optimized neural network, aiming at the problems of local convergence and long optimization time of the existing BP neural network prediction method.
The technical scheme is as follows: in order to realize the purpose of the invention, the technical scheme adopted by the invention is as follows:
a wind power short-term combination prediction method based on cluster analysis and an optimized neural network comprises the following steps:
s1: determining the influence factors of the wind power output power according to the wind power generated by the fan, the fan efficiency under wind directions with different wind speeds and the wake effect;
s2: clustering input samples according to a K-means clustering algorithm, and classifying the input samples according to the influence factors of the wind power output power;
s3: establishing a BP neural network prediction model corresponding to each type of input sample according to the classified input samples, and simultaneously optimizing each BP neural network prediction model through a thought evolution algorithm;
s4: and inputting the input sample into a corresponding optimized BP neural network prediction model according to the category of the input sample, calling parameters of the optimized BP neural network prediction model, predicting the wind power, and obtaining a future fan output curve.
Further, in step S1, the wind power generated by the wind turbine, the wind efficiency at different wind speeds and wind directions, and the calculation formula of the wake effect are specifically as follows:
Figure BDA0002303685590000021
wherein: p is the output power of the fan, CpIs the power coefficient of the fan, A is the swept area of the fan, ρ is the air density, v is the wind speed, η is the efficiency coefficient of the wind power, P is the efficiency coefficient of the wind powermFor actually measuring the output power, P, of the wind powerfThe output power of the wind power is under the condition of not being influenced by the wake flow.
Further, in step S2, the input samples are classified as follows:
s2.1: determining the number of the types of clusters according to the influence factors of the wind power output power;
s2.2: inputting the number of the types of the clusters and a sample data object to be clustered into a wind type clustering model of the K-means;
s2.3: selecting initial clustering centers from the sample data objects to be clustered, wherein the number of the initial clustering centers is the same as the number of the types of clustering;
s2.4: calculating the distance between each sample data object to be clustered except the initial clustering center and each initial clustering center, selecting the minimum distance corresponding to each sample data object to be clustered from the distances, and taking the class to which the initial clustering center corresponding to the minimum distance belongs as the class to which the sample data object to be clustered belongs;
s2.5: calculating the average value of each type of data object according to the class of the sample data object to be clustered, and taking the average value as a new clustering center of each type of data;
s2.6: and (5) repeating the step (S2.4) to the step (S2.5) until the clustering center is not changed any more, and obtaining the best clustering result.
Further, the objective function of the K-means clustering algorithm is specifically as follows:
Figure BDA0002303685590000022
wherein: y is the target function of the K-means clustering algorithm, | | xij-cj||2For clustering sample points xijCluster center c to jth groupjJ is the cluster group number, k is the number of classes of the cluster, niThe number of sample points in the ith type of sample data object.
Further, in step S3, each of the BP neural network prediction models is optimized as follows:
s3.1: according to the classified input samples, taking the wind speed and the wind direction in each class as the input of the BP neural network prediction model, taking the power in each class as the output of the BP neural network prediction model, and establishing the BP neural network prediction model corresponding to each class of input samples;
s3.2: calculating the coding length according to the structure of the BP neural network prediction model, wherein the calculation formula of the coding length specifically comprises the following steps:
s＝nl+ml+l+m
wherein: s is the coding length, n is the number of input nodes, m is the number of output nodes, and l is the number of hidden layer nodes;
s3.3: defining iteration times and an initial population size, pre-allocating a dominant sub-population and a temporary sub-population size, and determining the size of the sub-population, wherein a calculation formula of the size of the sub-population specifically comprises the following steps:
Figure BDA0002303685590000031
wherein: SG is the number of individuals in the sub-population, popsize is the initial population size, bestsize is the size of the dominant sub-population, tempsize is the size of the temporary sub-population;
s3.4: calculating score functions of each individual and each population, sequencing according to the score functions, and setting winning individuals and temporary individuals, wherein a calculation formula of the score functions of each individual and each population specifically comprises the following steps:
Figure BDA0002303685590000032
wherein: val is a score function of each individual and population, SE is mean square error, T is expected output, and A2 is an output value of an output layer after each iteration;
s3.5: performing convergence and differentiation operations on the sub-populations;
s3.6: judging whether the iteration times meet the preset maximum iteration times, if so, executing the next step, and if not, returning to the step S3.5;
s3.7: and analyzing the optimal individual obtained under the maximum iteration times to obtain an initial weight and a threshold of each BP neural network prediction model.
Further, in step S4, the normalized absolute average error and the normalized root mean square error are used as evaluation indicators of the prediction error, and the calculation formulas of the normalized absolute average error and the normalized root mean square error are specifically:
Figure BDA0002303685590000041
wherein: e.g. of the typeNMAETo normalize the absolute mean error, eNRMSETo normalize root mean square error, PcapRated capacity of wind turbineAnd m is the number of samples,
Figure BDA0002303685590000042
to predict the output, yiIs the actual output.
Has the advantages that: compared with the prior art, the technical scheme of the invention has the following beneficial technical effects:
aiming at some inherent defects of the BP neural network prediction method, the wind power short-term combination prediction method optimizes the initial weight and the threshold value by using a thought evolution algorithm, so that the wind condition can be identified, prediction models are respectively established for each class, the training time of the prediction models is reduced, the wind power prediction speed is increased, and the prediction precision is improved.
Drawings
FIG. 1 is a flow diagram of a wind power short-term combination prediction method;
FIG. 2 is a graph of a wind turbine power curve;
FIG. 3 is a graph of fan efficiency for different wind speeds and winds;
FIG. 4 is a schematic flow chart of a K-means clustering algorithm;
FIG. 5 is a schematic flow chart of a thought evolution algorithm for optimizing a BP neural network;
FIG. 6 is a graph of the wind type clustering results of the K-means algorithm;
FIG. 7 is a comparison graph of the predicted results of the present invention and a conventional wind power model;
FIG. 8 is a wind power prediction error frequency distribution histogram of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention. The described embodiments are a subset of the embodiments of the invention and are not all embodiments of the invention. Thus, the following detailed description of the embodiments of the present invention, presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention.
Example 1
Referring to fig. 1, the present embodiment provides a wind power short-term combination prediction method based on cluster analysis and an optimized neural network, which specifically includes the following steps:
step S1: analyzing the influence factors of the wind power output power, and determining the influence factors of the wind power output power as the wind speed and the wind direction according to the wind power generated by the fan, the wind efficiency and the wake effect under the wind directions with different wind speeds and wind directions by the power curve of the wind turbine and the relation between the wind directions and the wind power output power.
In this embodiment, the calculation formulas of the wind power generated by the wind turbine, the wind efficiency at different wind speeds and the wake effect specifically include:
Figure BDA0002303685590000051
wherein: p is the output power of the fan, CpIs the power coefficient of the fan, A is the swept area of the fan, ρ is the air density, v is the wind speed, η is the efficiency coefficient of the wind power, P is the efficiency coefficient of the wind powermFor actually measuring the output power, P, of the wind powerfThe output power of the wind power is under the condition of not being influenced by the wake flow.
Referring to fig. 2, it can be seen that the power curve of the fan exhibits a distinct piecewise characteristic, and the fan will output power when the wind speed is higher than 3.5 m/s. When the wind speed is 3.5 m/s-13 m/s, the power and the wind speed are approximately in a linear relation, and at the moment, a small wind speed change can cause a large power change. As shown in FIG. 2, the wind speed varied by 2.5m/s and the power varied by about 200 kW. When the wind speed is higher than 13m/s, the saturation state is reached and the power can not be increased any more.
Referring to fig. 3, it can be seen that the effect of wake effect is large at low wind speeds, and the efficiency is only 65% at the lowest at 4 m/s. Meanwhile, the larger the wind speed is, the smaller the influence of the wake effect is, and when the wind speed exceeds the rated wind speed to a certain degree, the wind energy captured by the rear exhaust fan also reaches the rated value, at the moment, the wake effect does not influence the output of power any more, and the efficiency coefficient in all wind directions is 100%.
Step S2: clustering input samples by using a K-means clustering algorithm, inputting the number K of the types of clustering and sample data objects to be clustered, randomly selecting K sample objects as initial clustering centers, continuously updating the clustering centers until the clustering centers are not changed any more, and obtaining the best clustering result. In the present embodiment, a data set X containing n samples is given (X ═ X1,x2,…,xn) And each sample is a d-dimensional real vector, and the target function of the K-means clustering algorithm is specifically as follows:
Figure BDA0002303685590000052
wherein: y is the target function of the K-means clustering algorithm, | | xij-cj||2For clustering sample points xijCluster center c to jth groupjJ is the cluster group number, k is the number of classes of the cluster, niThe number of sample points in the ith type of sample data object.
Referring to fig. 4, the process of obtaining the best clustering result is as follows:
step S2.1: according to the wind speed and the wind direction of the wind electricity output power influencing factors in the step S1, the sample objects are divided into breeze, wind electricity and gale according to the wind speed, and are divided into forward and reverse according to the wind direction, and 6 types are obtained in total, so that input samples under 6 types of different wind conditions are obtained. That is, in the present embodiment, the number k of categories of the cluster is 6.
Step S2.2: inputting the number K of the types of clustering and the sample data object to be clustered into a wind type clustering model of K-means.
Step S2.3: k initial clustering centers are randomly selected from input sample data objects to be clustered, wherein the number of the initial clustering centers is the same as the number of the clustering categories, that is, in this embodiment, 6 initial clustering centers need to be selected.
Step S2.4: and calculating the distance between each sample data object to be clustered except the selected 6 initial clustering centers and each initial clustering center, simultaneously selecting the minimum distance corresponding to each sample data object to be clustered according to the 6 distances corresponding to each sample data object to be clustered, and taking the class to which the initial clustering center corresponding to the minimum distance belongs as the class to which the sample data object to be clustered belongs.
Step S2.5: and calculating the average value of each type of data object according to the class of the sample data object to be clustered, and taking the average value of each type of data object as a new clustering center of the data, thereby finishing the updating of the clustering center.
Step S2.6: and (5) repeating the step (S2.4) to the step (S2.5) until the clustering center is not changed any more, so that the finally obtained clustering result is the optimal clustering result.
Step S3: referring to fig. 5, according to the input samples classified in step S2.6, a BP neural network prediction model corresponding to each type of input sample is established, and optimization is performed in the selectable range of the weight and the threshold of the BP neural network prediction model through a thought evolution algorithm to obtain an optimal initial weight and threshold, so as to complete optimization of each BP neural network prediction model, which is specifically as follows:
step S3.1: and according to the classified input samples, taking the wind speed and the wind direction in each class as the input of a BP neural network prediction model, taking the power in each class as the output of the BP neural network prediction model, and further establishing the BP neural network prediction model corresponding to each class of input samples.
Step S3.2: calculating the coding length according to the structure of the BP neural network prediction model obtained in the step S3.1, wherein the coding length is the number of the initial weight and the threshold obtained according to the structure of the BP neural network prediction model, and the calculation formula of the coding length specifically is as follows:
s＝nl+ml+l+m
wherein: s is the code length, n is the number of input nodes, m is the number of output nodes, and l is the number of hidden layer nodes.
In this embodiment, the coding length is obtained as 41 according to the structure of the BP neural network prediction model.
Step S3.3: defining iteration number iter, defining initial population size popsize, pre-allocating a winner sub-population size bestsize and a temporary sub-population size tempsize, and determining the size of a sub-population SG, wherein a calculation formula of the size of the sub-population SG specifically comprises the following steps:
Figure BDA0002303685590000071
wherein: SG is the number of individuals in the sub-population, popsize is the population size, bestsize is the size of the winner sub-population, tempsize is the size of the tentative sub-population.
In this example, an initial population and individuals were generated, resulting in 50 initial populations. Each population comprises a plurality of individuals with initial weights and threshold values, and the sub-population SG is the number of the individuals in each population.
Step S3.4: calculating the score functions of each individual and each group, arranging the score functions from high to low, selecting the first 25 as winners, wherein 5 groups are used as winning individuals, 20 groups are used as temporary individuals, and the 25 groups are used as centers to generate winning and temporary sub-groups. In particular, the scoring function is a parameter that evaluates the quality of individuals and populations.
In this embodiment, the calculation formula of the score function of each individual and population is specifically:
Figure BDA0002303685590000072
wherein: val is a score function of each individual and population, SE is mean square error, T is expected output, and A2 is the output value of the output layer after each iteration.
Since the population needs to be screened for the highest value of the score, the score function val of each individual and population takes the inverse of the mean square error of the expected value and the actual output value of each iteration.
Step S3.5: the homologation and dissimilarity operations are performed for the respective sub-populations.
Step S3.6: and (4) judging a convergence condition, namely judging whether the iteration times meet a preset maximum iteration time, if so, executing the step (S3.7), and if not, returning to the step (S3.5).
Step S3.7: and analyzing the optimal individual obtained under the maximum iteration times to obtain an initial weight and a threshold of each BP neural network prediction model. Meanwhile, according to the initial weight and the threshold of each BP neural network prediction model, training samples are input into the BP neural network prediction models with the initial weight and the threshold set for training, and relevant parameters of each type of BP neural network prediction models are stored.
Step S4: according to the category of the input sample, the input sample is input into the corresponding optimized BP neural network prediction model according to the category, and the wind power can be predicted by calling the parameters of the optimized BP neural network prediction model, so that the future fan output curve is obtained.
In this embodiment, the normalized absolute average error and the normalized root mean square error are used as evaluation indicators of the prediction error, and the calculation formulas of the normalized absolute average error and the normalized root mean square error are specifically as follows:
Figure BDA0002303685590000081
wherein: e.g. of the typeNMAETo normalize the absolute mean error, eNRMSETo normalize root mean square error, PcapIs the rated capacity of the wind turbine, m is the number of samples,
Figure BDA0002303685590000082
to predict the output, yiIs the actual output.
In the embodiment, an embodiment is also provided to further illustrate a wind power short-term combination prediction method based on cluster analysis and an optimization neural network.
Step 1: referring to FIG. 2, it can be seen from FIG. 2 that the power curve of the fan shows a distinct piecewise characteristic, and when the wind speed is higher than 3.5m/s, the fan will output power; when the wind speed is 3.5 m/s-13 m/s, the power and the wind speed are approximately in a linear relation, and at the moment, a small wind speed change can cause a large power change. As shown in the figure, the wind speed changes by 2.5m/s, and the power changes by about 200 kW; when the wind speed is higher than 13m/s, the saturation state is reached and the power can not be increased any more.
Referring to fig. 3, it can be seen that the effect of wake effect is large at low wind speeds, and the efficiency is only 65% at the lowest at 4 m/s. Meanwhile, the larger the wind speed is, the smaller the influence of the wake effect is, and when the wind speed exceeds the rated wind speed to a certain degree, the wind energy captured by the rear exhaust fan also reaches the rated value, at the moment, the wake effect does not influence the output of power any more, and the efficiency coefficient in all wind directions is 100%.
Step 2: the simulation is carried out by using the measured data of a certain wind power plant in northwest China, the measured data comprises the wind speed and the wind direction of the wind power plant within 12 months and the corresponding output power, and the time resolution is 1 h. In addition, the numerical weather forecast data in the corresponding time is as follows: temperature, humidity, and atmospheric pressure. And dividing the data into two parts, randomly selecting the data of one day as a test sample of the prediction model, and using the data of the rest days as a sample of model training. And the prediction results are compared with the prediction results of the particle swarm optimization artificial neural network model, and the various models adopt the same training and testing samples. Firstly, K-means clustering is carried out on training samples, the number of clustering groups is set to be 6, the wind type identification result based on the K-means clustering algorithm is shown in figure 6, and different wind types in the figure are represented by different colors.
It can be seen from the figure that, in the power axial direction, the wind type classification is relatively average, which indicates that the influence mode of the dispersion of the actual wind power on the prediction error is relatively fixed. In the aspect of the wind direction axial direction, the wind direction is mainly divided into two types, namely 0-180 degrees and 180-360 degrees, which are mainly related to the arrangement mode of wind field fans. In the axial direction of wind speed, the wind speed can be roughly divided into three types, namely breeze, stroke and gale, and the three types correspond to three sections of a power curve generated by a fan. In a word, the wind type clustering model can well identify the characteristics of the power generation process under different wind conditions.
And step 3: calculating according to the format of an input sample and the structure of a BP neural network to obtain a coding length of 41, defining the maximum iteration number of 1000, an initial population size of 50, a temporary sub-population size of 20 and a winning sub-population size of 5, inputting sample data, training, calculating the score condition of each individual and population, continuously iterating to obtain an optimal weight and a threshold, and storing the obtained parameters.
And 4, step 4: the wind power prediction is carried out by using the combined prediction method in the embodiment, and the prediction result is compared with the prediction results of other prediction models. As shown in fig. 7, it can be seen that the variation trend of the predicted value and the actual value of the combined prediction model in the embodiment is substantially the same, and the predicted result is closer to the actual output than the predicted result of other models.
Through calculation, the prediction relative errors of the BP neural network prediction model based on K-means cluster analysis and optimized by the thought evolution algorithm in the embodiment are mainly distributed between-5% and 5%, the errors of all the prediction points are distributed between-15% and 20%, and the prediction error frequency distribution histogram is as shown in FIG. 8 and basically conforms to normal distribution.
The prediction error indexes of each prediction model are calculated and shown in table 1, and table 1 specifically includes:
TABLE 1 prediction error index for each model
Figure BDA0002303685590000091
Wherein: e.g. of the typeNMAETo normalize the absolute mean error, eNRMSEIs the normalized root mean square error.
The error index in the table 1 is analyzed to find the normalized absolute average error e of the K-means-MAE-BP combined prediction modelNMAEAnd normalized root mean square error eNRMSECompared with K-means-PSO-BP, the weight is respectively reduced by 2.06% and 1.78%, which shows that the initial weight and the threshold of the BP neural network are optimized by using a thought evolution algorithm, so that the method can be more effectively applied to wind power prediction and improve the prediction precision. Compared with the prediction models of the MAE-BP and the K-means-MAE-BP, the prediction error indexes after the wind types are clustered by the K-means algorithm are respectively reduced by 4.76% and 5.49%, which shows that the K-means algorithm can more accurately simulate different wind types after the wind types are clustered by the K-means algorithmThe overall prediction accuracy of the prediction model is improved in the field power generation process under the wind condition type.
The present invention and its embodiments have been described in an illustrative manner, and are not to be considered limiting, as illustrated in the accompanying drawings, which are merely exemplary embodiments of the invention and not limiting of the actual constructions and methods. Therefore, if the person skilled in the art receives the teaching, the structural modes and embodiments similar to the technical solutions are not creatively designed without departing from the spirit of the invention, and all of them belong to the protection scope of the invention.

Claims (6)

1. A wind power short-term combination prediction method based on cluster analysis and an optimized neural network is characterized by comprising the following steps:
s1: determining the influence factors of the wind power output power according to the wind power generated by the fan, the fan efficiency under wind directions with different wind speeds and the wake effect;
s2: clustering input samples according to a K-means clustering algorithm, and classifying the input samples according to the influence factors of the wind power output power;
s3: establishing a BP neural network prediction model corresponding to each type of input sample according to the classified input samples, and simultaneously optimizing each BP neural network prediction model through a thought evolution algorithm;
s4: and inputting the input sample into a corresponding optimized BP neural network prediction model according to the category of the input sample, calling parameters of the optimized BP neural network prediction model, predicting the wind power, and obtaining a future fan output curve.
2. The wind power short-term combination prediction method based on cluster analysis and neural network optimization according to claim 1, wherein in step S1, the wind power generated by the wind turbine, the wind turbine efficiency at different wind speeds and wind directions, and the calculation formula of the wake effect are specifically:
Figure FDA0002303685580000011
wherein: p is the output power of the fan, CpIs the power coefficient of the fan, A is the swept area of the fan, ρ is the air density, v is the wind speed, η is the efficiency coefficient of the wind power, P is the efficiency coefficient of the wind powermFor actually measuring the output power, P, of the wind powerfThe output power of the wind power is under the condition of not being influenced by the wake flow.
3. The wind power short-term combination prediction method based on cluster analysis and optimization neural network as claimed in claim 1 or 2, wherein in the step S2, the input samples are classified as follows:
s2.1: determining the number of the types of clusters according to the influence factors of the wind power output power;
s2.2: inputting the number of the types of the clusters and a sample data object to be clustered into a wind type clustering model of the K-means;
s2.3: selecting initial clustering centers from the sample data objects to be clustered, wherein the number of the initial clustering centers is the same as the number of the types of clustering;
s2.4: calculating the distance between each sample data object to be clustered except the initial clustering center and each initial clustering center, selecting the minimum distance corresponding to each sample data object to be clustered from the distances, and taking the class to which the initial clustering center corresponding to the minimum distance belongs as the class to which the sample data object to be clustered belongs;
s2.5: calculating the average value of each type of data object according to the class of the sample data object to be clustered, and taking the average value as a new clustering center of each type of data;
s2.6: and (5) repeating the step (S2.4) to the step (S2.5) until the clustering center is not changed any more, and obtaining the best clustering result.
4. The wind power short-term combination prediction method based on cluster analysis and neural network optimization according to claim 3, characterized in that the objective function of the K-means clustering algorithm is specifically:
Figure FDA0002303685580000021
wherein: y is the target function of the K-means clustering algorithm, | | xij-cj||2For clustering sample points xijCluster center c to jth groupjJ is the cluster group number, k is the number of classes of the cluster, niThe number of sample points in the ith type of sample data object.
5. The wind power short-term combination prediction method based on cluster analysis and optimization neural network as claimed in claim 3, wherein in the step S3, each BP neural network prediction model is optimized as follows:
s3.1: according to the classified input samples, taking the wind speed and the wind direction in each class as the input of the BP neural network prediction model, taking the power in each class as the output of the BP neural network prediction model, and establishing the BP neural network prediction model corresponding to each class of input samples;
s3.2: calculating the coding length according to the structure of the BP neural network prediction model, wherein the calculation formula of the coding length specifically comprises the following steps:
s＝nl+ml+l+m
wherein: s is the coding length, n is the number of input nodes, m is the number of output nodes, and l is the number of hidden layer nodes;
s3.3: defining iteration times and an initial population size, pre-allocating a dominant sub-population and a temporary sub-population size, and determining the size of the sub-population, wherein a calculation formula of the size of the sub-population specifically comprises the following steps:
Figure FDA0002303685580000022
wherein: SG is the number of individuals in the sub-population, popsize is the initial population size, bestsize is the size of the dominant sub-population, tempsize is the size of the temporary sub-population;
s3.4: calculating score functions of each individual and each population, sequencing according to the score functions, and setting winning individuals and temporary individuals, wherein a calculation formula of the score functions of each individual and each population specifically comprises the following steps:
Figure FDA0002303685580000031
wherein: val is a score function of each individual and population, SE is mean square error, T is expected output, and A2 is an output value of an output layer after each iteration;
s3.5: performing convergence and differentiation operations on the sub-populations;
s3.6: judging whether the iteration times meet the preset maximum iteration times, if so, executing the next step, and if not, returning to the step S3.5;
s3.7: and analyzing the optimal individual obtained under the maximum iteration times to obtain an initial weight and a threshold of each BP neural network prediction model.
6. The wind power short-term combination prediction method based on cluster analysis and optimization neural network as claimed in claim 5, wherein in step S4, normalized absolute mean error and normalized root mean square error are used as evaluation indexes of prediction error, and the calculation formulas of normalized absolute mean error and normalized root mean square error are specifically:
Figure FDA0002303685580000032
wherein: e.g. of the typeNMAETo normalize the absolute mean error, eNRMSETo normalize root mean square error, PcapIs the rated capacity of the wind turbine, m is the number of samples,
Figure FDA0002303685580000033
to predict the output, yiIs the actual output.
CN201911231533.XA 2019-12-05 2019-12-05 Wind power short-term combination prediction method based on cluster analysis and neural network optimization Active CN110991743B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911231533.XA CN110991743B (en) 2019-12-05 2019-12-05 Wind power short-term combination prediction method based on cluster analysis and neural network optimization

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911231533.XA CN110991743B (en) 2019-12-05 2019-12-05 Wind power short-term combination prediction method based on cluster analysis and neural network optimization

Publications (2)

Publication Number Publication Date
CN110991743A true CN110991743A (en) 2020-04-10
CN110991743B CN110991743B (en) 2022-08-19

Family

ID=70090244

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911231533.XA Active CN110991743B (en) 2019-12-05 2019-12-05 Wind power short-term combination prediction method based on cluster analysis and neural network optimization

Country Status (1)

Country Link
CN (1) CN110991743B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114048930A (en) * 2022-01-13 2022-02-15 广东电网有限责任公司揭阳供电局 Ultra-short-term wind power prediction method and device

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104573879A (en) * 2015-01-30 2015-04-29 河海大学 Photovoltaic power station output predicting method based on optimal similar day set
CN105631517A (en) * 2015-12-17 2016-06-01 河海大学 Photovoltaic power generation power short term prediction method based on mind evolution Elman neural network
CN106251001A (en) * 2016-07-18 2016-12-21 南京工程学院 A kind of based on the photovoltaic power Forecasting Methodology improving fuzzy clustering algorithm
CN108564192A (en) * 2017-12-29 2018-09-21 河海大学 A kind of short-term photovoltaic power prediction technique based on meteorological factor weight similar day
CN109214566A (en) * 2018-08-30 2019-01-15 华北水利水电大学 Short-term wind power prediction method based on shot and long term memory network
CN110334847A (en) * 2019-05-24 2019-10-15 广东智造能源科技研究有限公司 Based on the wind power prediction method for improving K-means cluster and support vector machines

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104573879A (en) * 2015-01-30 2015-04-29 河海大学 Photovoltaic power station output predicting method based on optimal similar day set
CN105631517A (en) * 2015-12-17 2016-06-01 河海大学 Photovoltaic power generation power short term prediction method based on mind evolution Elman neural network
CN106251001A (en) * 2016-07-18 2016-12-21 南京工程学院 A kind of based on the photovoltaic power Forecasting Methodology improving fuzzy clustering algorithm
CN108564192A (en) * 2017-12-29 2018-09-21 河海大学 A kind of short-term photovoltaic power prediction technique based on meteorological factor weight similar day
CN109214566A (en) * 2018-08-30 2019-01-15 华北水利水电大学 Short-term wind power prediction method based on shot and long term memory network
CN110334847A (en) * 2019-05-24 2019-10-15 广东智造能源科技研究有限公司 Based on the wind power prediction method for improving K-means cluster and support vector machines

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
关添升等: "基于k-means聚类的SVR短期风速预测", 《可再生能源》 *
刘永前等: "风电场输出功率预测中两种神经网络算法的研究", 《现代电力》 *
方市彬等: "基于NWP单点聚类分析与BP神经网络的短期风电功率预测", 《电气应用》 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114048930A (en) * 2022-01-13 2022-02-15 广东电网有限责任公司揭阳供电局 Ultra-short-term wind power prediction method and device

Also Published As

Publication number Publication date
CN110991743B (en) 2022-08-19

Similar Documents

Publication Publication Date Title
CN108197744B (en) Method and system for determining photovoltaic power generation power
CN111008504B (en) Wind power prediction error modeling method based on meteorological pattern recognition
CN106251001A (en) A kind of based on the photovoltaic power Forecasting Methodology improving fuzzy clustering algorithm
CN109978284B (en) Photovoltaic power generation power time-sharing prediction method based on hybrid neural network model
CN111753893A (en) Wind turbine generator power cluster prediction method based on clustering and deep learning
CN111369070A (en) Envelope clustering-based multimode fusion photovoltaic power prediction method
CN112186761B (en) Wind power scene generation method and system based on probability distribution
CN110766200A (en) Method for predicting generating power of wind turbine generator based on K-means mean clustering
CN109492748B (en) Method for establishing medium-and-long-term load prediction model of power system based on convolutional neural network
CN103489046A (en) Method for predicting wind power plant short-term power
CN113344288B (en) Cascade hydropower station group water level prediction method and device and computer readable storage medium
CN107609774B (en) Photovoltaic power prediction method for optimizing wavelet neural network based on thought evolution algorithm
CN110263834B (en) Method for detecting abnormal value of new energy power quality
CN113762387B (en) Multi-element load prediction method for data center station based on hybrid model prediction
CN110738232A (en) grid voltage out-of-limit cause diagnosis method based on data mining technology
CN111292124A (en) Water demand prediction method based on optimized combined neural network
CN115099296A (en) Sea wave height prediction method based on deep learning algorithm
CN112288157A (en) Wind power plant power prediction method based on fuzzy clustering and deep reinforcement learning
CN110570042B (en) Short-term electric vehicle charging load prediction method and system
CN115995810A (en) Wind power prediction method and system considering weather fluctuation self-adaptive matching
CN115759389A (en) Day-ahead photovoltaic power prediction method based on weather type similar day combination strategy
CN110991743B (en) Wind power short-term combination prediction method based on cluster analysis and neural network optimization
CN110991689A (en) Distributed photovoltaic power generation system short-term prediction method based on LSTM-Morlet model
CN110163437B (en) Day-ahead photovoltaic power generation power prediction method based on DPK-means
CN112330051A (en) Short-term load prediction method based on Kmeans and FR-DBN

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