CN117709082B - Wind generating set power curve prediction method - Google Patents
Wind generating set power curve prediction method Download PDFInfo
- Publication number
- CN117709082B CN117709082B CN202311645622.5A CN202311645622A CN117709082B CN 117709082 B CN117709082 B CN 117709082B CN 202311645622 A CN202311645622 A CN 202311645622A CN 117709082 B CN117709082 B CN 117709082B
- Authority
- CN
- China
- Prior art keywords
- wind turbine
- detection
- data
- turbine generator
- power
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 50
- 238000007781 pre-processing Methods 0.000 claims abstract description 7
- 238000012795 verification Methods 0.000 claims abstract description 4
- 238000001514 detection method Methods 0.000 claims description 184
- 230000002159 abnormal effect Effects 0.000 claims description 44
- 238000012549 training Methods 0.000 claims description 43
- 238000012360 testing method Methods 0.000 claims description 39
- 239000013598 vector Substances 0.000 claims description 12
- 238000002955 isolation Methods 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 8
- 238000012544 monitoring process Methods 0.000 claims description 8
- 238000002922 simulated annealing Methods 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 5
- 238000005070 sampling Methods 0.000 claims description 5
- 230000005856 abnormality Effects 0.000 claims description 4
- 238000010276 construction Methods 0.000 claims description 2
- 238000010248 power generation Methods 0.000 abstract description 6
- 230000008901 benefit Effects 0.000 description 5
- 238000010586 diagram Methods 0.000 description 3
- 238000005311 autocorrelation function Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000010998 test method Methods 0.000 description 2
- OKTJSMMVPCPJKN-UHFFFAOYSA-N Carbon Chemical compound [C] OKTJSMMVPCPJKN-UHFFFAOYSA-N 0.000 description 1
- 230000002547 anomalous effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 229910052799 carbon Inorganic materials 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 239000000446 fuel Substances 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- Water Supply & Treatment (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- Human Resources & Organizations (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- Public Health (AREA)
- Computer Hardware Design (AREA)
- Evolutionary Computation (AREA)
- Geometry (AREA)
- General Engineering & Computer Science (AREA)
- Wind Motors (AREA)
Abstract
The invention relates to the technical field of fan power prediction, and discloses a wind generating set power curve prediction method, which comprises the following steps: acquiring operation history data of the wind turbine generator under different time dimensions and preprocessing; drawing a historical power curve trend chart of the wind turbine generator and carrying out stability verification; constructing a wind turbine power curve prediction model, optimizing, and predicting wind turbine power by using the optimized wind turbine power curve prediction model; and drawing a wind turbine predicted power curve trend graph according to the predicted wind turbine power, and fitting the wind turbine predicted power curve trend graph with a wind turbine historical power curve trend graph to evaluate a power prediction result. The method can effectively solve the problem of low accuracy of the existing wind power prediction method, improves the stability and accuracy of the power generation of the wind turbine, and ensures the normal operation of the wind turbine to the greatest extent.
Description
Technical Field
The invention relates to the technical field of fan power prediction, in particular to a wind generating set power curve prediction method.
Background
With the global pursuit of the "two carbon" goal, renewable energy is becoming an important choice for replacing fossil energy. Among them, wind energy is considered as one of the most potential renewable energy sources due to its unique advantages. Wind energy is an inexhaustible natural resource, and its renewable nature determines its importance in future energy structures. Moreover, wind energy is a clean energy source, and the wind energy is used for generating electricity without burning fuel or polluting the environment, so that the wind energy fully accords with the sustainable development goal of people.
Wind power generation has been widely used worldwide. However, although wind power generation has many advantages, it also has a certain problem. First, uncertainty in wind energy is a significant challenge. Because the change of wind speed cannot be accurately predicted, the power output of wind power often has larger fluctuation, which not only affects the economic benefit of the electric power market, but also threatens the stability of the power grid. Thus, accurate prediction of wind power becomes critical. By predicting wind power, the power resource can be better scheduled, the operation of the wind power plant is optimized, and the economic benefit and stability of the power system are improved. Meanwhile, accurate wind power prediction is also helpful for better understanding and managing wind energy resources, and powerful support is provided for the sustainable development of wind energy. Therefore, the invention provides a wind generating set power curve prediction method.
Disclosure of Invention
Aiming at the problems in the related art, the invention provides a wind generating set power curve prediction method, solves the problem of lower accuracy of the existing wind power prediction method, can improve the stability and accuracy of the power generation of the wind generating set, can furthest ensure the normal operation of the wind generating set, and can overcome the technical problems existing in the related art.
For this purpose, the invention adopts the following specific technical scheme:
a wind generating set power curve prediction method comprises the following steps:
S1, acquiring wind turbine generator operation history data under different time dimensions, and preprocessing the wind turbine generator operation history data to obtain power history sequence data under normal working conditions;
S2, drawing a historical power curve trend chart of the wind turbine generator according to the power historical sequence data under the normal working condition, and performing stability verification on the power historical sequence data;
S3, constructing a wind turbine generator power curve prediction model based on power history sequence data under normal working conditions, optimizing the model, and predicting the wind turbine generator power by using the optimized wind turbine generator power curve prediction model;
and S4, drawing a wind turbine predicted power curve trend graph according to the predicted wind turbine power, and fitting the wind turbine predicted power curve trend graph with a wind turbine historical power curve trend graph to evaluate a power prediction result.
Further, the step of obtaining the wind turbine generator operation history data in different time dimensions and preprocessing the wind turbine generator operation history data to obtain the power history sequence data under the normal working condition comprises the following steps:
S11, acquiring wind turbine running history data in different time dimensions, wherein the wind turbine running history data are SCADA data of a wind turbine;
S12, eliminating error values and zero values in abnormal working conditions in the operation history data of the wind turbine generator to obtain an initial data set;
S13, identifying abnormal data in the operation history data of the wind turbine generator by using a data detection algorithm based on an isolation forest, and removing the abnormal data in the initial data set to obtain a data set;
S14, dividing the wind turbine operation history data in the data set into a training set and a testing set, analyzing the correlation between wind power and other influencing factors by using cosine similarity, and selecting a monitoring quantity with high correlation as a model input characteristic variable.
Further, the method for identifying the abnormal data in the wind turbine generator operation history data by using the data detection algorithm based on the isolation forest, and removing the abnormal data in the initial data set to obtain the data set comprises the following steps:
S131, setting the maximum height of the detection tree, generating the number of the detection trees and the sub-sampling size, and initializing a detection forest;
s132, constructing L detection trees to form an initial detection forest;
s133, training an initial detection forest by using a training set, calculating the precision value of each detection tree, and calculating the difference value between the detection trees by using a statistic method;
S134, selecting L detection trees with fitness values meeting requirements from the initial detection forests by using a simulated annealing algorithm according to the difference and the accuracy of each detection tree to form a new detection forest;
S135, performing anomaly detection on the operation history data of the wind turbine generator by using a new detection forest, and calculating the path length of each data in each detection tree to obtain an anomaly score;
S136, judging whether the operation history data of the wind turbine generator is abnormal according to the abnormal score, and eliminating abnormal data in the initial data set to obtain a cleaned data set.
Further, the calculating the precision value of each detection tree includes the following steps:
Dividing the original training data in the training set into N subsets which are equal in number and mutually disjoint, training by using the N-1 subsets each time, and testing by using the rest one subset;
and training and testing N data in the N subsets one by one as a test set, wherein the average value of the final N metric values is the precision value of the test tree.
Further, the calculation formula of the difference value is as follows:
the calculation formula of the fitness value is as follows:
Where Q i,j represents a difference value between the detection tree T i and the detection tree T j, N ab represents a detection result of a sample in the detection training set of the detection tree T i and the detection tree T j, a=1 if the detection tree T i can correctly detect data in the training set, a=0 if the detection tree Tj can correctly detect data in the training set, b=1 if the detection tree Tj can correctly detect data in the training set, b= 0,F (T j) represents an fitness function of the detection number Tj, P j represents an accuracy value of T j, and W 1 and W 2 represent weights corresponding to accuracy and difference respectively.
Further, according to the difference and accuracy of each detection tree, selecting L detection trees with fitness values meeting requirements from the initial detection forest by using a simulated annealing algorithm to form a new detection forest, wherein the method comprises the following steps:
S1341, taking the difference and the precision of each detection tree as an objective function, and defining an initial detection tree;
S1342, setting an initial temperature and setting a termination condition;
S1343, randomly generating a part of new detection trees in each iteration, and calculating the objective function value and the difference value between the new detection trees and the current detection tree;
S1344, judging whether the new detection tree is better than the current detection tree, if so, accepting the new detection tree, otherwise, accepting the new detection tree with a preset probability, wherein the probability depends on the difference between the current temperature and the objective function values of the two detection trees;
s1345, judging whether the termination condition is met, if not, returning to S1343 until the termination condition is met, if so, selecting a detection tree with the best objective function value from all the received new detection trees, and obtaining the best L detection trees to form a new detection forest.
Further, the method for performing anomaly detection on the operation history data of the wind turbine generator by using the new detection forest, calculating the path length of each data in each detection tree, and obtaining the anomaly score comprises the following steps:
S1351, inputting operation history data of a wind turbine to be detected into a new detection forest, and classifying the data according to the structure and rules of each detection tree;
S1352, calculating the path length of the operation history data of each wind turbine to be detected in the detection tree, and calculating the anomaly score.
Further, the calculation formula of the anomaly score is as follows:
Where E (h (d)) represents an average value of path lengths h (d) in the detection tree set, C (m) represents an average value of path lengths h (d) at a given m time, path lengths h (d) are the number of edges experienced from the root node to the external node, d represents data, and m is the number of leaf nodes.
Further, the method for constructing and optimizing the wind turbine power curve prediction model based on the power history sequence data under the normal working condition, and predicting the wind turbine power by using the optimized wind turbine power curve prediction model comprises the following steps:
S31, constructing a wind turbine generator power curve prediction model based on power history sequence data under normal working conditions, wherein the expression of the wind turbine generator power curve prediction model is as follows:
y=Xβ+ε;
Wherein y represents a wind power vector, beta represents a model parameter vector, X represents a monitoring quantity matrix with high correlation with wind power, epsilon represents a random error term vector with a mean value of 0 and a variance of 1;
s32, optimizing a model on a training set by using a least square method, and searching for optimal parameters Minimizing a residual objective function to obtain an optimized wind turbine generator power curve prediction model, wherein the expression of the optimized wind turbine generator power curve prediction model is/>
Further, the method for estimating the power prediction result includes the following steps:
S41, drawing a wind turbine generator predicted power curve trend chart according to predicted wind turbine generator power;
and S42, fitting the wind turbine predicted power curve trend graph with the wind turbine historical power curve trend graph, judging whether the deviation value of the wind turbine predicted power curve trend graph and the wind turbine historical power curve trend graph is smaller than a preset threshold, if so, judging that the power prediction is normal, and if not, judging that the power prediction is abnormal.
The beneficial effects of the invention are as follows:
1) The invention provides a wind turbine generator system power curve prediction method, which can effectively solve the problem of lower accuracy of the existing wind turbine generator system power prediction method, improve the stability and accuracy of the wind turbine generator system power generation, and ensure the normal operation of the wind turbine generator system to the maximum extent.
2) Considering that the data detection algorithm of the isolated forest performs well for a data set of high dimensionality, but does not perform well for a data set of low latitude, in addition, when the proportion of outliers in the data set is high, the effect of isolating the forest may be degraded. The method introduces normal data, and performs classification fusion with the input wind turbine running history data according to the time sequence and the serial numbers of the wind turbines, so that the proportion of abnormal points is reduced, and meanwhile, high-dimensional data is formed according to the time gradient, so that the generated detection tree has stronger robustness.
3) According to the method, the abnormal data in the operation history data of the wind turbine generator can be detected and identified by utilizing the data detection algorithm based on the isolation forest, and the abnormal data can be removed based on the detection result, so that the influence on the prediction of the power curve of the wind turbine generator due to data abnormality can be effectively avoided.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are needed in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a method of predicting a power curve of a wind turbine generator set in accordance with an embodiment of the present invention;
Fig. 2 is a trend chart of wind power prediction curves and a trend chart of actual power curves in a method for predicting power curves of a wind generating set according to an embodiment of the present invention.
Detailed Description
For the purpose of further illustrating the various embodiments, the present invention provides the accompanying drawings, which are a part of the disclosure of the present invention, and which are mainly used to illustrate the embodiments and, together with the description, serve to explain the principles of the embodiments, and with reference to these descriptions, one skilled in the art will recognize other possible implementations and advantages of the present invention, wherein elements are not drawn to scale, and like reference numerals are generally used to designate like elements.
According to the embodiment of the invention, a wind generating set power curve prediction method is provided.
The invention will now be further described with reference to the accompanying drawings and detailed description, as shown in fig. 1-2, according to one embodiment of the invention, there is provided a method for predicting a power curve of a wind turbine generator set, comprising the steps of:
S1, acquiring wind turbine generator operation history data under different time dimensions, and preprocessing the wind turbine generator operation history data to obtain power history sequence data under normal working conditions;
the method comprises the steps of acquiring wind turbine generator operation history data in different time dimensions, preprocessing the wind turbine generator operation history data, and obtaining power history sequence data under normal working conditions, wherein the steps of:
S11, acquiring wind turbine running history data in different time dimensions, wherein the wind turbine running history data are SCADA data of a wind turbine;
Specifically, wind farm historical data are obtained, and wind farm SCADA data are obtained according to wind farm business scenes.
S12, eliminating error values and zero values in abnormal working conditions in the operation history data of the wind turbine generator to obtain an initial data set;
S13, identifying abnormal data in the operation history data of the wind turbine generator by using a data detection algorithm based on an isolation forest, and removing the abnormal data in the initial data set to obtain a data set;
Specifically, the method for identifying abnormal data in the wind turbine generator operation history data by using a data detection algorithm based on an isolation forest, removing the abnormal data in the initial data set, and obtaining the data set comprises the following steps:
S131, setting the maximum height of the detection tree, generating the number of the detection trees and the sub-sampling size, and initializing a detection forest;
s132, constructing L detection trees to form an initial detection forest;
For the data detection algorithm of the isolated forest, whether the initial detection forest is robust has a great influence on the final result. The data detection algorithm considering the isolated forest performs well on high-dimensional data, but may perform poorly on low-dimensional data. On the other hand, because the abnormal point data is not dense in the wind turbine generator system operation history data, constructing L detection trees based on the abnormal point data specifically comprises the following steps:
s1321, acquiring normal wind turbine generator system operation data.
The normal wind turbine generator operation data may be a plurality of wind turbine generator operation history data screened out when a data detection algorithm of the isolated forest was previously executed.
S1322, calculating fitting data corresponding to each wind turbine generator set based on the normal wind turbine generator set operation data.
Specifically, the normal wind turbine generator set operation data are classified according to each wind turbine motor, and fitting data corresponding to each wind turbine motor are further constructed.
S1323, calculating a distance value corresponding to the operation history data of each wind turbine generator based on the fitting data.
Assuming that the wind turbine generator operation history data are two-dimensional vectors, the fitting data are straight lines, and the distance value is the distance from each wind turbine generator operation history data to the straight line; and assuming that the wind turbine generator operation history data are three-dimensional vectors, the fitting data are planes, and the distance value is the distance from each wind turbine generator operation history data to the plane. It will be appreciated that the distance value may be positive or negative, for example, when a certain vector is above the fitting data, the convention is positive, otherwise the convention is negative, and vice versa. In some embodiments, the wind turbine operation history data may be six-dimensional data, including: wind power, temperature, humidity, pressure, wind speed, wind direction. Fitting data can be constructed by least squares.
S1324, constructing L detection trees based on the distance values to form an initial detection forest.
In some embodiments, L may be the number of wind turbines, and the construction method of each detection tree is as follows:
S13241, arranging absolute values of distance values in order from small to large to obtain a data sequence, selecting wind turbine generator set operation history data with minimum absolute values of distance values from the data sequence as a root node of the detection tree, removing the root node from the data sequence, and marking the root node as a current node;
S13242, selecting wind turbine generator system operation history data with the minimum absolute value of distance value from the data sequence, if the wind turbine generator system operation history data is opposite to the sign of the current node, using the wind turbine generator system operation history data as a left child node of the current node, otherwise, using the wind turbine generator system operation history data as a right child node of the current node, and marking the wind turbine generator system operation history data as the current node;
S13243, repeatedly executing step S13242 until the data sequence is null.
S133, training an initial detection forest by using a training set, calculating the precision value of each detection tree, and calculating the difference value between the detection trees by using a statistic method;
specifically, the calculating the precision value of each detection tree includes the following steps:
dividing the original training data in the training set into N subsets which are equal in number and mutually disjoint, training by using the N-1 subsets each time, and testing by using the rest one subset; and training and testing N data in the N subsets one by one as a test set, wherein the average value of the final N metric values is the precision value of the test tree.
Specifically, the calculating the difference value between the detection trees by using the statistic method includes the following steps:
Given a training set, if the detection tree T i can correctly detect data in the training set, a=1, otherwise a=0, and the list of detection results of T i and T j is listed in table 1;
Table 1 Table i and Table j are listed in the Table
Ti=1 | Tj=0 | |
Tj=1 | N11 | N10 |
Ti=0 | N01 | N00 |
The calculation formula of the difference value is as follows:
Where Q i,j represents a difference value between the test tree T i and the test tree T j, N ab represents a test result of the test tree T i and the test tree T j for testing samples in the training set, a=1 if the test tree T i can correctly test data in the training set, a=0 if the test tree T j can correctly test data in the training set, b=1 if the test tree T3962 can correctly test data in the training set, b=0 if the test tree T j can correctly test data in the training set, and the statistical belief that the Q statistic value of the two test trees is 0 if the two test trees are independent. The value of the Q statistic varies between [ -1,1], the larger the value the smaller the degree of difference between the two detection trees.
S134, selecting L detection trees with fitness values meeting requirements from the initial detection forests by using a simulated annealing algorithm according to the difference and the accuracy of each detection tree to form a new detection forest;
specifically, according to the difference and accuracy of each detection tree, selecting L detection trees with fitness values meeting requirements from the initial detection forest by using a simulated annealing algorithm to form a new detection forest, wherein the method comprises the following steps of:
S1341, taking the difference and the precision of each detection tree as an objective function, and defining an initial detection tree;
S1342, setting an initial temperature and setting a termination condition;
S1343, randomly generating a part of new detection trees in each iteration, and calculating the objective function value and the difference value between the new detection trees and the current detection tree;
S1344, judging whether the new detection tree is better than the current detection tree, if so, accepting the new detection tree, otherwise, accepting the new detection tree with a preset probability, wherein the probability depends on the difference between the current temperature and the objective function values of the two detection trees;
s1345, judging whether the termination condition is met, if not, returning to S1343 until the termination condition is met, if so, selecting a detection tree with the best objective function value from all the received new detection trees, and obtaining the best L detection trees to form a new detection forest.
The calculation formula of the fitness value is as follows:
Where F (T j) represents the fitness function of the detection number T j, P j represents the precision value of T j, and W 1 and W 2 represent weights corresponding to precision and variability, respectively.
S135, performing anomaly detection on the operation history data of the wind turbine generator by using a new detection forest, and calculating the path length of each data in each detection tree to obtain an anomaly score;
Specifically, the abnormal detection is performed on the operation history data of the wind turbine generator by using a new detection forest, the path length of each data in each detection tree is calculated, and the abnormal score is obtained, which comprises the following steps:
S1351, inputting operation history data of a wind turbine to be detected into a new detection forest, and classifying the data according to the structure and rules of each detection tree;
s1352, calculating the path length of the operation history data of each wind turbine generator to be detected, that is, the number of edges passing from the root node to the node where classification ends, in the detection tree, generally, the path length of the abnormal data will be longer than that of the normal data, because the abnormal data is generally more difficult to be accurately classified, and calculating an abnormal score, and the higher the abnormal score, the higher the probability that the data is considered to be abnormal.
Specifically, a subset of the extracted dataset is randomly sampled to construct each detection tree to ensure diversity of the detection tree. The path length of the data d in each tree is calculated by traversing each detection tree in the detection forest, and then the anomaly score of d is calculated according to the path length, so that whether d is anomalous is judged, and the calculation formula of the anomaly score is as follows:
Wherein E (h (d)) represents the average value of path lengths h (d) in the detection tree set, and when E (h (d)). Fwdarw.C (m), S.fwdarw.0.5, namely when S.apprxeq.0.5 is returned by all data, no obvious outlier is present in all samples; when E (h (d)). Fwdarw.0, S.fwdarw.1, i.e., when the S to which the data is returned is very close to 1, they are outliers; when E (h (d)). Fwdarw.m-1, S.fwdarw.0, i.e., when the S returned by the data is much smaller than 0.5, they are highly likely to be normal values;
C (m) is expressed as an average value of path length h (d) at a given m, path length h (d) being the number of edges experienced from the root node to the external node, d representing data, and m being the number of leaf nodes.
S136, judging whether the operation history data of the wind turbine generator is abnormal according to the abnormal score, and eliminating abnormal data in the initial data set to obtain a cleaned data set.
S14, dividing the wind turbine operation history data in the data set into a training set and a testing set, analyzing the correlation between wind power and other influencing factors by using cosine similarity, and selecting a monitoring quantity with high correlation as a model input characteristic variable.
S2, drawing a historical power curve trend chart of the wind turbine generator according to the power historical sequence data under the normal working condition, and performing stability verification on the power historical sequence data;
Specifically, drawing a power trend chart of the wind turbine generator according to the power history sequence data, and performing stationarity check on the power history sequence data trend chart comprises: drawing an autocorrelation function diagram of power history sequence data according to a time sequence, judging whether the trend diagram of the power history sequence data is stable or not according to the waveform of the autocorrelation function diagram, carrying out stability check on the power history sequence by adopting an ADF (automatic frequency correction) test method, and carrying out stability check on the power history sequence by adopting the ADF test method comprises the following steps: determining the proper order of the power history sequence, and determining through a data trend graph; performing differential processing on the power history sequence; ADF test statistics are calculated, and for a given level of significance, a test conclusion is drawn from the value of the test statistics, and if the test statistics are less than a threshold, the power history sequence is considered stationary.
S3, constructing a wind turbine generator power curve prediction model based on power history sequence data under normal working conditions, optimizing the model, and predicting the wind turbine generator power by using the optimized wind turbine generator power curve prediction model;
The method for predicting the power of the wind turbine by using the optimized wind turbine power curve prediction model comprises the following steps of:
S31, constructing a wind turbine power curve prediction model based on the monitoring quantity with high correlation with wind power in the power history sequence data under the normal working condition, wherein the expression of the wind turbine power curve prediction model is as follows:
y=Xβ+ε;
Wherein y represents a wind power vector, beta represents a model parameter vector, X represents a monitoring quantity matrix with high correlation with wind power, epsilon represents a random error term vector with a mean value of 0 and a variance of 1;
s32, optimizing a model on a training set by using a least square method, and searching for optimal parameters Minimizing a residual objective function to obtain an optimized wind turbine generator power curve prediction model, wherein the expression of the optimized wind turbine generator power curve prediction model is/>
Specifically, on a training set, searching optimal parameters by utilizing a least square optimization modelMinimizing a residual objective function;
the matrix after derivative arrangement is expressed as:
Wherein X' represents the transpose of the matrix X, n represents the number of model input variables, and the optimal parameters obtained by the least square method The expression of the optimized wind turbine generator power curve prediction model is/>
And S4, drawing a wind turbine predicted power curve trend graph according to the predicted wind turbine power, and fitting the wind turbine predicted power curve trend graph with a wind turbine historical power curve trend graph to evaluate a power prediction result.
The method comprises the steps of drawing a predicted power curve trend graph of the wind turbine according to predicted power of the wind turbine, fitting the predicted power curve trend graph of the wind turbine with a historical power curve trend graph of the wind turbine, and evaluating a power prediction result, wherein the method comprises the following steps:
S41, drawing a wind turbine generator predicted power curve trend chart according to predicted wind turbine generator power;
and S42, fitting the wind turbine predicted power curve trend graph with the wind turbine historical power curve trend graph, judging whether the deviation value of the wind turbine predicted power curve trend graph and the wind turbine historical power curve trend graph is smaller than a preset threshold, if so, judging that the power prediction is normal, and if not, judging that the power prediction is abnormal.
Specific application examples are as follows:
Taking wind power prediction of a certain wind farm as an example, table 2 is an analysis result of cosine similarity between wind power and influence factors of the present invention, and fig. 2 is a trend chart of wind power curves predicted by the present invention and a trend chart of actual power curves.
Specifically, the method for predicting the wind power curve of a certain wind power plant comprises the following steps:
Step 1, acquiring historical operation data of 2021 years 1 to 12 months from a SCADA system of a certain wind farm, wherein the data sampling interval is 15 minutes. The data comprises wind power, temperature, humidity, pressure, wind speed and wind direction, wherein the wind speed and wind direction comprise 7 measuring points.
And 2, dividing the acquired historical data into a training set and a testing set, removing error values and zero values under abnormal working conditions, identifying abnormal data in the operation historical data of the wind turbine by using a data detection algorithm based on an isolation forest, removing the abnormal data in an initial data set, analyzing the correlation between wind power and other influencing factors by using cosine similarity, and selecting a monitoring quantity with high correlation as a model input characteristic variable. The analysis results of cosine similarity between wind power and influence factors in this example are shown in table 2. And analyzing the correlation between wind power and parameters such as V1 wind speed, V1 wind direction, V2 wind speed, V2 wind direction, V3 wind speed, V3 wind direction, V5 wind speed, V5 wind direction, V6 wind speed, V6 wind direction, V7 wind speed, V7 wind direction, temperature, humidity, pressure and the like by utilizing cosine similarity, and selecting parameters such as V1 wind speed, V2 wind speed, V3 wind speed, V5 wind speed, V6 wind speed, V7 wind speed, temperature, humidity, pressure and the like with the correlation being more than 0.95 as input characteristic variables of the model.
TABLE 2 analysis of cosine similarity between wind power and influence factors
And 3, constructing a wind power prediction model according to parameters with high correlation between the sampling interval and the wind power, and training an optimization model on a training set.
And step 4, testing the trained model on the test set, and drawing a power curve trend graph predicted by the model and an actual power curve trend graph. Fig. 2 is a graph showing the trend of the wind power curve predicted by the present invention and the trend of the actual power curve.
In summary, by means of the technical scheme, the invention provides the wind turbine generator system power curve prediction method, which can effectively solve the problem of low accuracy of the existing wind power prediction method, improve the stability and accuracy of the wind turbine generator system power generation, and ensure the normal operation of the wind turbine generator system to the greatest extent. In addition, the method can realize detection and identification of abnormal data in the wind turbine generator operation history data by utilizing the data detection algorithm based on the isolation forest, and can realize rejection of the abnormal data based on the detection result, so that the influence on wind turbine generator power curve prediction caused by data abnormality can be effectively avoided.
In the present invention, the above-mentioned embodiments are only preferred embodiments of the present invention, and are not intended to limit the present invention, but any modifications, equivalent substitutions, improvements, etc. within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (7)
1. The wind generating set power curve prediction method is characterized by comprising the following steps of:
S1, acquiring wind turbine generator operation history data under different time dimensions, and preprocessing the wind turbine generator operation history data to obtain power history sequence data under normal working conditions;
S2, drawing a historical power curve trend chart of the wind turbine generator according to the power historical sequence data under the normal working condition, and performing stability verification on the power historical sequence data;
S3, constructing a wind turbine generator power curve prediction model based on power history sequence data under normal working conditions, optimizing the model, and predicting the wind turbine generator power by using the optimized wind turbine generator power curve prediction model;
S4, drawing a predicted power curve trend graph of the wind turbine according to the predicted power of the wind turbine, and fitting the predicted power curve trend graph of the wind turbine with a historical power curve trend graph of the wind turbine to evaluate a power prediction result;
The method for obtaining the wind turbine generator operation history data under different time dimensions and preprocessing the wind turbine generator operation history data to obtain the power history sequence data under the normal working condition comprises the following steps:
S11, acquiring wind turbine running history data in different time dimensions, wherein the wind turbine running history data are SCADA data of a wind turbine;
S12, eliminating error values and zero values in abnormal working conditions in the operation history data of the wind turbine generator to obtain an initial data set;
S13, identifying abnormal data in the operation history data of the wind turbine generator by using a data detection algorithm based on an isolation forest, and removing the abnormal data in the initial data set to obtain a data set;
S14, dividing wind turbine operation history data in the data set into a training set and a testing set, analyzing the correlation between wind power and other influencing factors by using cosine similarity, and selecting a monitoring quantity with high correlation as a model input characteristic variable;
The method for identifying the abnormal data in the wind turbine generator operation history data by using the data detection algorithm based on the isolation forest, and removing the abnormal data in the initial data set to obtain the data set comprises the following steps:
S131, setting the maximum height of the detection tree, generating the number of the detection trees and the sub-sampling size, and initializing a detection forest;
s132, constructing L detection trees to form an initial detection forest;
s133, training an initial detection forest by using a training set, calculating the precision value of each detection tree, and calculating the difference value between the detection trees by using a statistic method;
S134, selecting L detection trees with fitness values meeting requirements from the initial detection forests by using a simulated annealing algorithm according to the difference and the accuracy of each detection tree to form a new detection forest;
S135, performing anomaly detection on the operation history data of the wind turbine generator by using a new detection forest, and calculating the path length of each data in each detection tree to obtain an anomaly score;
S136, judging whether the operation history data of the wind turbine generator is abnormal according to the abnormal score, and eliminating abnormal data in the initial data set to obtain a cleaned data set;
constructing L detection trees and forming an initial detection forest comprises the following steps:
S1321, acquiring normal wind turbine generator system operation data;
s1322, calculating fitting data corresponding to each wind turbine generator set based on normal wind turbine generator set operation data;
S1323, calculating a distance value corresponding to the operation history data of each wind turbine generator based on the fitting data;
S1324, constructing L detection trees based on the distance value to form an initial detection forest;
Constructing L detection trees based on the distance values, wherein the forming of the initial detection forest specifically comprises the following steps: l is the number of wind turbines, and the construction method of each detection tree is as follows:
S13241, arranging absolute values of distance values in order from small to large to obtain a data sequence, selecting wind turbine generator set operation history data with minimum absolute values of distance values from the data sequence as a root node of the detection tree, removing the root node from the data sequence, and marking the root node as a current node;
S13242, selecting wind turbine generator system operation history data with the minimum absolute value of distance value from the data sequence, if the wind turbine generator system operation history data is opposite to the sign of the current node, using the wind turbine generator system operation history data as a left child node of the current node, otherwise, using the wind turbine generator system operation history data as a right child node of the current node, and marking the wind turbine generator system operation history data as the current node;
S13243, repeating step S13242 until the data sequence is empty;
the wind turbine generator power prediction model is constructed and optimized based on the power history sequence data under the normal working condition, and the wind turbine generator power prediction by utilizing the optimized wind turbine generator power prediction model comprises the following steps:
S31, constructing a wind turbine generator power curve prediction model based on power history sequence data under normal working conditions, wherein the expression of the wind turbine generator power curve prediction model is as follows:
y=Xβ+ε;
Wherein y represents a wind power vector, beta represents a model parameter vector, X represents a monitoring quantity matrix with high correlation with wind power, epsilon represents a random error term vector with a mean value of 0 and a variance of 1;
s32, optimizing a model on a training set by using a least square method, and searching for optimal parameters Minimizing a residual objective function to obtain an optimized wind turbine generator power curve prediction model, wherein the expression of the optimized wind turbine generator power curve prediction model is/>
2. A method of predicting a power curve of a wind turbine generator system according to claim 1, wherein said calculating the accuracy value of each of the test trees comprises the steps of:
Dividing the original training data in the training set into N subsets which are equal in number and mutually disjoint, training by using the N-1 subsets each time, and testing by using the rest one subset;
and training and testing N data in the N subsets one by one as a test set, wherein the average value of the final N metric values is the precision value of the test tree.
3. The method for predicting a power curve of a wind generating set according to claim 2, wherein the calculation formula of the difference value is:
the calculation formula of the fitness value is as follows:
Where Q i,j represents a difference value between the detection tree T i and the detection tree T j, N ab represents a detection result of a sample in the detection training set of the detection tree T i and the detection tree T j, a=1 if the detection tree T i can correctly detect data in the training set, a=0 if the detection tree Tj can correctly detect data in the training set, b=1 if the detection tree Tj can correctly detect data in the training set, b= 0,F (T j) represents an fitness function of the detection number Tj, P j represents an accuracy value of T j, and W 1 and W 2 represent weights corresponding to accuracy and difference respectively.
4. The method for predicting the power curve of the wind generating set according to claim 1, wherein the step of selecting the new detection forest from the initial detection forests by using the simulated annealing algorithm according to the difference and the accuracy of each detection tree, wherein the L detection trees meet the requirement of the fitness value of the detection trees are combined into the new detection forests comprises the following steps:
S1341, taking the difference and the precision of each detection tree as an objective function, and defining an initial detection tree;
S1342, setting an initial temperature and setting a termination condition;
S1343, randomly generating a part of new detection trees in each iteration, and calculating the objective function value and the difference value between the new detection trees and the current detection tree;
S1344, judging whether the new detection tree is better than the current detection tree, if so, accepting the new detection tree, otherwise, accepting the new detection tree with a preset probability, wherein the probability depends on the difference between the current temperature and the objective function values of the two detection trees;
s1345, judging whether the termination condition is met, if not, returning to S1343 until the termination condition is met, if so, selecting a detection tree with the best objective function value from all the received new detection trees, and obtaining the best L detection trees to form a new detection forest.
5. The method for predicting the power curve of the wind turbine generator set according to claim 1, wherein the abnormality detection is performed on the operation history data of the wind turbine generator set by using a new detection forest, the path length of each data in each detection tree is calculated, and the abnormality score is obtained by the following steps:
S1351, inputting operation history data of a wind turbine to be detected into a new detection forest, and classifying the data according to the structure and rules of each detection tree;
S1352, calculating the path length of the operation history data of each wind turbine to be detected in the detection tree, and calculating the anomaly score.
6. The method for predicting a power curve of a wind turbine generator system according to claim 5, wherein the anomaly score is calculated according to the formula:
Where E (h (d)) represents an average value of path lengths h (d) in the detection tree set, C (m) represents an average value of path lengths h (d) at a given m time, path lengths h (d) are the number of edges experienced from the root node to the external node, d represents data, and m is the number of leaf nodes.
7. The method for predicting the power curve of the wind turbine generator set according to claim 1, wherein the steps of drawing a trend chart of the predicted power curve of the wind turbine generator set according to the predicted power of the wind turbine generator set, fitting the trend chart of the predicted power curve of the wind turbine generator set with a trend chart of the historical power curve of the wind turbine generator set, and evaluating the power prediction result comprise the following steps:
S41, drawing a wind turbine generator predicted power curve trend chart according to predicted wind turbine generator power;
and S42, fitting the wind turbine predicted power curve trend graph with the wind turbine historical power curve trend graph, judging whether the deviation value of the wind turbine predicted power curve trend graph and the wind turbine historical power curve trend graph is smaller than a preset threshold, if so, judging that the power prediction is normal, and if not, judging that the power prediction is abnormal.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311645622.5A CN117709082B (en) | 2023-12-04 | 2023-12-04 | Wind generating set power curve prediction method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311645622.5A CN117709082B (en) | 2023-12-04 | 2023-12-04 | Wind generating set power curve prediction method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN117709082A CN117709082A (en) | 2024-03-15 |
CN117709082B true CN117709082B (en) | 2024-06-18 |
Family
ID=90148976
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202311645622.5A Active CN117709082B (en) | 2023-12-04 | 2023-12-04 | Wind generating set power curve prediction method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN117709082B (en) |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110619360A (en) * | 2019-09-09 | 2019-12-27 | 国家电网有限公司 | Ultra-short-term wind power prediction method considering historical sample similarity |
CN117154680A (en) * | 2023-05-25 | 2023-12-01 | 中国科学院深圳先进技术研究院 | Wind power prediction method based on non-stationary transducer model |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US11531676B2 (en) * | 2020-07-29 | 2022-12-20 | Intuit Inc. | Method and system for anomaly detection based on statistical closed-form isolation forest analysis |
-
2023
- 2023-12-04 CN CN202311645622.5A patent/CN117709082B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110619360A (en) * | 2019-09-09 | 2019-12-27 | 国家电网有限公司 | Ultra-short-term wind power prediction method considering historical sample similarity |
CN117154680A (en) * | 2023-05-25 | 2023-12-01 | 中国科学院深圳先进技术研究院 | Wind power prediction method based on non-stationary transducer model |
Also Published As
Publication number | Publication date |
---|---|
CN117709082A (en) | 2024-03-15 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110596492B (en) | Transformer fault diagnosis method based on particle swarm optimization random forest model | |
US20230136352A1 (en) | Method and system for predicting a day-ahead wind power of wind farms | |
CN109873610B (en) | Photovoltaic array fault diagnosis method based on IV characteristic and depth residual error network | |
CN111709490B (en) | Fan health state assessment method based on GRU neural network | |
CN108336739B (en) | RBF neural network-based probability load flow online calculation method | |
CN110082136B (en) | Rotary machine fault diagnosis method based on cloud genetic algorithm optimization support vector machine | |
CN115018021A (en) | Machine room abnormity detection method and device based on graph structure and abnormity attention mechanism | |
CN112116058A (en) | Transformer fault diagnosis method for optimizing multi-granularity cascade forest model based on particle swarm algorithm | |
CN110263834B (en) | Method for detecting abnormal value of new energy power quality | |
CN116522268B (en) | Line loss anomaly identification method for power distribution network | |
CN115021679A (en) | Photovoltaic equipment fault detection method based on multi-dimensional outlier detection | |
CN115115090A (en) | Wind power short-term prediction method based on improved LSTM-CNN | |
CN112801350B (en) | Uncertainty-based deep learning ultra-short-time wind power prediction system and method | |
CN111323847B (en) | Method and apparatus for determining weight ratios for analog integration algorithms | |
CN117709082B (en) | Wind generating set power curve prediction method | |
CN117540317A (en) | Distribution network abnormal line change relation identification method based on cosine similarity and isolated forest | |
CN108038518A (en) | A kind of photovoltaic generation power based on meteorological data determines method and system | |
CN116051911B (en) | Small sample bearing vibration image data fault diagnosis method based on uncertainty learning | |
CN110717623B (en) | Photovoltaic power generation power prediction method, device and equipment integrating multiple weather conditions | |
CN113837096B (en) | Rolling bearing fault diagnosis method based on GA random forest | |
CN113496255B (en) | Power distribution network mixed observation point distribution method based on deep learning and decision tree driving | |
Jing et al. | A data expansion based piecewise regression strategy for incrementally monitoring the wind turbine with power curve | |
CN113554203B (en) | Wind power prediction method and device based on high-dimensional meshing and LightGBM | |
Jianli et al. | Wind power forecasting by using artificial neural networks and Grubbs criterion | |
CN110032981B (en) | Rotary machine fault identification method based on improved support vector machine |
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 |