CN110826791A - Hybrid wind power prediction method based on long-time and short-time memory neural network - Google Patents

Hybrid wind power prediction method based on long-time and short-time memory neural network Download PDF

Info

Publication number
CN110826791A
CN110826791A CN201911050486.9A CN201911050486A CN110826791A CN 110826791 A CN110826791 A CN 110826791A CN 201911050486 A CN201911050486 A CN 201911050486A CN 110826791 A CN110826791 A CN 110826791A
Authority
CN
China
Prior art keywords
data
model
prediction
wind power
time
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.)
Pending
Application number
CN201911050486.9A
Other languages
Chinese (zh)
Inventor
王恭
张群
唐振浩
童瑶
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.)
Northeast Electric Power University
Original Assignee
Northeast Dianli University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Northeast Dianli University filed Critical Northeast Dianli University
Priority to CN201911050486.9A priority Critical patent/CN110826791A/en
Publication of CN110826791A publication Critical patent/CN110826791A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • 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)
  • General Physics & Mathematics (AREA)
  • Economics (AREA)
  • General Health & Medical Sciences (AREA)
  • Biophysics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • Strategic Management (AREA)
  • Human Resources & Organizations (AREA)
  • Evolutionary Computation (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • Artificial Intelligence (AREA)
  • General Business, Economics & Management (AREA)
  • Marketing (AREA)
  • Tourism & Hospitality (AREA)
  • Operations Research (AREA)
  • Primary Health Care (AREA)
  • Development Economics (AREA)
  • Game Theory and Decision Science (AREA)
  • Water Supply & Treatment (AREA)
  • Public Health (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Quality & Reliability (AREA)
  • Wind Motors (AREA)

Abstract

The invention discloses a mixed wind power prediction method based on a long-time and short-time memory neural network, which belongs to the technical field of wind power generation power prediction.A long-time and short-time memory neural network is introduced into wind power prediction; performing wavelet threshold denoising on the selected characteristic attribute value, and acquiring an original signal from a signal mixed with strong noise; the LSTM method is adopted to train the hybrid wind power prediction model, and the weight matrix of the prediction model is updated by increasing the iteration times, so that the prediction precision is improved; and finally, correcting the prediction error of the LSTM model by using a least square method, and further improving the prediction precision of the LSTM hybrid wind power prediction method.

Description

Hybrid wind power prediction method based on long-time and short-time memory neural network
Technical Field
The invention relates to the field of active power prediction of wind turbine generators in the wind power generation process of a wind power plant, in particular to a hybrid wind power prediction method based on a long-time memory neural network.
Background
The installed capacity of wind power in China is increasing year by year, and the proportion of wind power in the power industry is continuously expanded. The volatility, intermittency and chaos of wind power bring a serious challenge to the safe and stable operation of a power grid, and meanwhile, a wind power plant is required to keep high rotating reserve capacity to stabilize the voltage of the power grid, so that serious economic loss is caused. In order to reduce economic loss, the power system requires the wind power of the wind power plant to be predicted in an ultra-short period (15 min-4 h) and a short period (5 h-48 h), and the prediction error generally should not exceed 20%.
Although the traditional statistical method and machine learning method can predict wind power of a wind power plant, the two methods still have the following defects:
1) the time dependence information in the sample information is not considered in the traditional machine learning algorithm, so that the calculation cost is high, and the learning efficiency of the algorithm is influenced.
2) The traditional statistical method is difficult to mine deep characteristic information in data, and the prediction performance still has a larger space for improvement.
Disclosure of Invention
Aiming at the defects in the prior art, the invention provides the hybrid wind power prediction method based on the long-time memory neural network, which has the advantages of simple steps, lower calculation cost and higher prediction precision and is suitable for practical prediction application.
The technical scheme adopted by the invention is as follows:
a mixed wind power prediction method based on a long-time memory neural network comprises the following steps:
step 1: obtaining historical data sets of fan systems from wind farms
Figure BDA0002255210720000011
To representThe method comprises the following steps of selecting alternative input variables, wherein N represents the number of the alternative input variables, N belongs to N, y represents a current-time wind power sequence, ranking the importance of the alternative input variables relative to the current-time wind power sequence by utilizing a decision tree algorithm, and eliminating input features with low importance degree to achieve the purpose of selecting the data features of the fan system;
step 1-1: calculating each input variable
Figure BDA0002255210720000013
For the importance of the input variable complete set V relative to the decision attribute set D, the formula is:
Figure BDA0002255210720000014
wherein, U is a non-empty finite set of decision variables of the decision tree, and is called a domain of discourse; v is a complete set of input variables; d is a decision attribute set; the positive V domain of D is denoted posV(D) (ii) a D of
Figure BDA0002255210720000015
Positive field is marked as
Figure BDA0002255210720000016
sig () represents importance (significance), n is the total number of input features of the wind turbine system;
step 1-2, setting an importance threshold, selecting a data set with importance greater than the threshold as an input parameter of a prediction model, and reordering the selected input characteristic data sets into
Figure BDA0002255210720000021
Step 2: denoising and normalizing the selected data, and dividing the processed data set into a training set and a test set according to a proportion;
step 2-1: the wavelet threshold denoising method is adopted to obtain an original signal from a signal mixed with strong noise conveniently and quickly, and 8-layer wavelet decomposition and denoising processing are carried out on a noisy signal by using Daubiche 4 wavelet:
1) setting the detail coefficients of 1 layer, 7 layers and the approximate coefficients of 8 layers to 0;
2) processing the 3-6 layers by using a soft threshold function;
3) and reconstructing the processed 8 layers of signals to obtain a denoised data vector x'.
Step 2-2: normalizing the denoised data by adopting a Min-Max method to eliminate dimensional difference between different data, wherein the Min-Max method has the following calculation formula:
Figure BDA0002255210720000022
wherein, x'minIs the minimum value, x 'in the data of a single row'maxIs the maximum value in the single-column data, and x is the normalized data;
step 2-3: and selecting a data segment from the normalized data to establish a data set, and dividing the data set into a training set and a testing set according to a proportion.
And step 3: establishing a mixed wind power prediction model based on a long-time memory LSTM neural network, and establishing the mixed wind power prediction model according to the training set obtained in the step 2;
step 3-1: building an LSTM network structure;
the LSTM network includes 1 input layer, 5 hidden layers, and 1 output layer. The output layer adopts an identity function as a regression function; the operation principle of the LSTM network is shown in equations (3) to (7):
ft＝sigmoid(Wfxxt+Whfht-1+bf) (3)
kt＝sigmoid(Wkxxt+Wkhht-1+bk) (4)
gt＝tanh(Wgxxt+Wghht-1+bg) (5)
ot＝sigmoid(Woxxt+Wohht-1+bo) (6)
ht＝tanh(gt⊙kt+St-1⊙ft)⊙ot(7)
where ⊙ denotes the element multiplication, t denotes the sampling time when the subscript is t, xtFor data sampled at time t, ft,kt,gt,otAnd htRespectively the states of the forgetting gate, the input node, the output gate and the state unit at the time ttFor the updated state of the memory cell at time t, St-1For the memory cell state at time t-1, Wfx,Whf,Wkx,Wkh,Wgx,Wgh,WoxAnd WohRespectively corresponding to the gate and input xtAnd state unit intermediate output ht-1Weight matrices of the products, the weight matrices being diagonal matrices; bf,bk,bg,boRespectively the offset of the corresponding door; tanh represents an activation function;
finally, obtaining an LSTM model output formula:
G＝f(xt,Whf,Wkh,Woh,B) (8)
wherein G represents the long-short time memory neural network model output, Whf、Wkh、WohRespectively representing an input gate weight vector, a forgetting gate weight vector and an output gate weight vector, and B represents a bias;
step 3-2: optimizing parameters of a long-time and short-time memory neural network by adopting an interval enumeration method;
and carrying out preliminary experiments on the number of hidden layer nodes and the batch scale by adopting an interval enumeration method to obtain a preselected interval of the two parameters. Then, carrying out wind power prediction experiments by using a long-time memory neural network with different combinations of two parameters of the number of hidden layer nodes and the batch scale, and obtaining the optimal number of the hidden layer nodes and the batch scale modulus according to a prediction result;
step 3-3: and (3) bringing the preprocessed data set into an LSTM prediction model for training to obtain a predicted value G of the sample in the training set, wherein G is vector representation of the output predicted value.
And 4, step 4: comparing the result of the model training of the fan system data with the actual value to obtain a correction formula, and correcting the mixed wind power prediction model based on the long-time memory neural network by using the formula to improve the accuracy of model prediction;
step 4-1: calculating a prediction error of a fan system data set;
Ei'＝Xi'-Gi(9)
wherein, Xi' is a preprocessed data set, GiObtaining a predicted value of the ith sample after training the LSTM model, Ei' is the prediction error of the LSTM model;
step 4-2: establishing a prediction model of a prediction error;
Figure BDA0002255210720000031
wherein E isi' error value for model prediction, C constant, AiRepresenting the model coefficient, Xi' represents input parameters of a training set, and l is the number of data in a data vector;
step 4-3: fitting constant C and model coefficient A by least square methodiAnd obtaining a relational expression of the model error value with respect to the input parameter.
Step 4-4: predicting the prediction error of the model by using the relational expression obtained in the step 4-3;
and 4-5: correcting and calculating the predicted value of the model by using the error predicted value obtained in the step 4-4,
Gi'＝Gi-Ei' (11)
wherein G isiShows the predicted value of the ith sample obtained after LSTM model training, Gi' denotes a corrected predicted value, i.e., a corrected LSTM model output vector.
And 5: and testing the test set of the fan system data by using the corrected prediction model, calculating to obtain the prediction accuracy, and verifying the prediction accuracy of the prediction model.
Drawings
Table 1 is an alternative input variable table selected in the embodiment of the present invention;
FIG. 1 is a flow chart of a hybrid wind power prediction method based on a long-time and short-time memory neural network according to the present invention;
FIG. 2 is a graph illustrating an importance ranking of feature attributes of variables using a decision tree algorithm in an embodiment of the present invention;
FIG. 3 is a comparison graph before and after denoising wind power data and wind speed data by using a wavelet threshold denoising method in the embodiment of the present invention;
FIG. 4 is a scatter plot of the comparison of prediction accuracies before and after the LSTM model is modified in an embodiment of the present invention;
FIG. 5 is a graph illustrating wind power prediction using a modified LSTM model in accordance with an embodiment of the present invention;
FIG. 6 is a diagram illustrating the comparison of the prediction errors of the modified LSTM model and the BP, RBF, MLP, and LSSVM models in accordance with an embodiment of the present invention.
Detailed Description
The following describes in further detail embodiments of the present invention with reference to the accompanying drawings. The following examples are intended to illustrate the invention but are not intended to limit the scope of the invention.
As shown in fig. 1, the method of the present embodiment is as follows.
Step 1: selecting 29 characteristic attributes including historical wind speed (instantaneous wind speed, average wind speed of the first 30 seconds and average wind speed of the first 10 minutes), active power data (the first 1 moment to the first 20 moments) and grid phase voltages (A phase, B phase and C phase), generator rotating speed, wind wheel rotating speed and gearbox oil temperature from a wind power plant SCADA monitoring system as alternative input variables
Figure BDA0002255210720000042
As shown in table 1;
TABLE 1 wind-power-related variable table
Figure BDA0002255210720000041
Figure BDA0002255210720000051
Step 1-1: computing a single input variable using a decision tree algorithm
Figure BDA0002255210720000052
For the importance of the input variable corpus V relative to the decision variable D, the formula is:
Figure BDA0002255210720000053
wherein, U is a non-empty finite set of decision variables of the decision tree, and is called a domain of discourse; v is a complete set of input variables; d is a decision attribute set; the positive V domain of D is denoted posV(D) (ii) a D of
Figure BDA0002255210720000054
Positive field is marked as
Figure BDA0002255210720000055
sig () represents importance (significance), n is the total number of input features of the wind turbine system;
obtaining the importance ranking of the feature attributes shown in FIG. 2;
step 1-2, selecting an importance threshold value of 0.8, and selecting 23 fan system characteristic parameters with the importance greater than the threshold value of 0.80 in the graph 2 as model input, wherein the method comprises the following steps: active power from the first 1 to the first 20 moments, instantaneous wind speed at the first 1 moment, average wind speed 30 seconds before the first 1 moment and average wind speed 10 minutes before the first 1 moment are removed, 6 characteristic variables of grid phase voltage (A phase, B phase and C phase), generator rotating speed, wind wheel rotating speed and gearbox oil temperature with importance lower than 0.80 in the graph 2 are removed, and the selected characteristic data sets are reordered to obtain
Figure BDA0002255210720000056
To
Figure BDA0002255210720000057
As a predictive modelThe input parameters of (1);
step 2: preprocessing the selected 23 characteristic data sets, and dividing the processed data sets into a training set and a testing set according to a ratio of 4: 1;
step 2-1: the wavelet threshold denoising method is adopted to obtain an original signal from a signal mixed with strong noise conveniently and quickly, and 8-layer wavelet decomposition and denoising processing are carried out on a noisy signal by using Daubiche 4 wavelet:
1) setting the detail coefficients of 1 layer, 7 layers and the approximate coefficients of 8 layers to 0;
2) processing the layers 3 to 6 by using a soft threshold algorithm;
3) and reconstructing the processed 8 layers of signals to obtain a denoised data vector x'.
As shown in FIG. 3, a comparison graph before and after denoising wind speed data is shown.
Step 2-2: normalizing the denoised data by adopting a Min-Max method to eliminate dimensional difference between different data, wherein the Min-Max method has the following calculation formula:
wherein x ' is the denoised data set signal, x ' in step 2-2 'minIs the minimum value, x 'in the data of a single row'maxIs the maximum value in the single-column data, and x is the normalized data;
step 2-3: and sequentially establishing 2000 groups of data sets of the processed data selection data segments, wherein the first 1600 groups are taken as a training set and the last 400 groups are taken as a test set according to the ratio of 4: 1.
And step 3: establishing a mixed wind power prediction model based on a long-time memory neural network, and establishing the mixed wind power prediction model according to the training set obtained in the step 2;
step 3-1: building an LSTM network structure;
the LSTM network includes 1 input layer, 5 hidden layers, and 1 output layer. The output layer adopts an identity function as a regression function; the operation principle of the LSTM network is shown in formula (3) -formula (7):
ft＝sigmoid(Wfxxt+Whfht-1+bf) (3)
kt＝sigmoid(Wkxxt+Wkhht-1+bk) (4)
gt＝tanh(Wgxxt+Wghht-1+bg) (5)
ot＝sigmoid(Woxxt+Wohht-1+bo)(6)
ht＝tanh(gt⊙kt+St-1⊙ft)⊙ot(7)
where ⊙ denotes the element multiplication, t denotes the sampling time when the subscript is t, xtFor data sampled at time t, ft,kt,gt,otAnd htRespectively the states of the forgetting gate, the input node, the output gate and the state unit at the time ttFor the updated state of the memory cell at time t, St-1For the memory cell state at time t-1, Wfx,Whf,Wkx,Wkh,Wgx,Wgh,WoxAnd WohRespectively corresponding to the gate and input xtAnd state unit intermediate output ht-1Weight matrices of the products, the weight matrices being diagonal matrices; bf,bk,bg,boRespectively the offset of the corresponding door; tanh represents an activation function;
finally, obtaining an LSTM model output formula:
G＝f(xt,Whf,Wkh,Woh,B) (8)
wherein G represents the long-short time memory neural network model output, Whf、Wkh、WohRespectively representing an input gate weight vector, a forgetting gate weight vector and an output gate weight vector, and B represents a bias;
step 3-2: optimizing parameters of a long-time and short-time memory neural network by adopting an interval enumeration method;
and carrying out preliminary experiments on the number of hidden layer nodes and the batch scale by adopting an interval enumeration method to obtain a preselected interval of the two parameters. On the premise that other parameters are not changed, one of the parameters is changed, the trend of an error curve is observed, and finally when the number of hidden layer nodes is determined to be between 110 and 160, the model error is within an acceptable range, the error values of the parameters at the two ends of the range are similar, the model error of the batch scale modulus is within the acceptable range within the range of 1-26, then the two parameters are subjected to combination experiments by respectively taking 6 groups of parameters at equal intervals within a preselected interval range, finally, the number s of the hidden layer nodes is selected to be 130, and the number j of the batch scale is selected to be 1;
and 3-3, substituting 1600 groups of training set data obtained in the step 2 into an LSTM prediction model for training to obtain a predicted value G of the training set after model training, wherein G is the vector representation of the output predicted value.
And 4, step 4: comparing the result of the model training of the fan system data with the actual value to obtain a correction formula, and correcting the mixed wind power prediction model based on the long-time memory neural network by using the formula to improve the accuracy of model prediction;
step 4-1: calculating a prediction error of a fan system data set;
Ei'＝Xi'-Gi(9)
wherein, Xi' for the preprocessed training set 1600 groups of data, GiFor the predicted values obtained after training of the LSTM model, Ei' is the prediction error of the LSTM model, i ═ 1,2, …, 1600;
step 4-2: establishing a prediction model of a prediction error;
wherein E isi' represents an error value of the model prediction in step 4-1, C represents a constant, AiRepresenting the input coefficient, Xi' represents the input parameters of the training set, and l is the number 1600 of data in the data vector;
step 4-3:fitting constant C and input coefficient A by least square methodi，C＝-8.8，AiArranging according to the input sequence of the step 1-2: a [ -400, -273,661,791, -374, -1692,2361, -1737,637,1111, -2129,2309, -3675,4530, -3721,2562, -1278,932, -792,133, -1266,0,1561]And obtaining a relational expression of the model error value with respect to the input parameter.
Step 4-4: predicting the prediction error of the model by using the relational expression obtained in the step 4-3;
and 4-5: and 4, carrying out correction calculation on the predicted value of the model by using the error predicted value obtained in the step 4-4
Gi'＝Gi-Ei' (11)
Wherein G isiRepresenting the predicted value of the model before correction, i.e. the output vector of the LSTM model before correction, Gi' denotes a corrected predicted value, i.e., a corrected LSTM model output vector, as shown in FIG. 4Gi' value and GiAnd comparing the values, so that the corrected prediction curve is closer to an ideal prediction curve.
And 5: and testing 400 groups of data in the test set of the fan system data by using the corrected prediction model, wherein the comparison between the test value and the true value is shown in FIG. 5.
Finally, prediction experiments were performed using the LSTM model and the comparative models (BP, RBF, MLP, and LSSVM), respectively. FIG. 6 is a graph of 5 model error comparisons. Through experimental analysis, the solution proposed by each link in the modeling process plays a positive role in improving the final modeling precision, and the constructed hybrid wind power prediction model based on the long-time and short-time neural network can meet the requirements of the wind power plant on wind power prediction.

Claims (5)

1. A mixed wind power prediction method based on a long-time memory neural network is characterized by comprising the following steps:
step 1: obtaining historical data sets of fan systems from wind farms
Figure FDA0002255210710000015
Figure FDA0002255210710000016
Representing alternative input variables, N representing the number of the alternative input variables, N belonging to N, y representing a current-time wind power sequence, sequencing the importance of the alternative input variables relative to the current-time wind power sequence by utilizing a decision tree algorithm, and eliminating input features with low importance degree to achieve the purpose of selecting the data features of the fan system;
step 2: denoising and normalizing the selected data, and dividing the processed data set into a training set and a test set according to a proportion;
and step 3: establishing a mixed wind power prediction model based on a long-time memory LSTM neural network, and establishing the mixed wind power prediction model according to the training set obtained in the step 2;
and 4, step 4: comparing the result of the model training of the fan system data with the actual value to obtain a correction formula, and correcting the mixed wind power prediction model based on the long-time memory neural network by using the formula to improve the accuracy of model prediction;
and 5: and testing the test set of the fan system data by using the corrected prediction model, calculating to obtain the prediction accuracy, and verifying the prediction accuracy of the prediction model.
2. The method for predicting the hybrid wind power based on the long-time and short-time memory neural network according to claim 1, wherein the process of the step 1 is as follows:
step 1-1: calculating each input variable
Figure FDA0002255210710000011
For the importance of the input variable complete set V relative to the decision attribute set D, the formula is:
Figure FDA0002255210710000012
wherein U is a non-null limit of decision variables of the decision treeCollections, called discourse domains; v is a complete set of input variables; d is a decision attribute set; the positive V domain of D is denoted posV(D) (ii) a D of
Figure FDA0002255210710000013
Positive field is marked as
Figure FDA0002255210710000014
sig () represents importance (significance), n is the total number of input features of the wind turbine system;
step 1-2, setting an importance threshold, selecting a data set with importance greater than the threshold as an input parameter of a prediction model, and reordering the selected input characteristic data sets into
3. The method for predicting the hybrid wind power based on the long-and-short-term memory neural network as claimed in claim 1, wherein in the step 2, the selected data is preprocessed as follows:
step 2-1: the wavelet threshold denoising method is adopted to obtain an original signal from a signal mixed with strong noise conveniently and quickly, and 8-layer wavelet decomposition and denoising processing are carried out on a noisy signal by using Daubiche 4 wavelet:
1) setting the detail coefficients of 1 layer, 7 layers and the approximate coefficients of 8 layers to 0;
2) processing the 3-6 layers by using a soft threshold function;
3) reconstructing the processed 8 layers of signals to obtain a denoised data vector x';
step 2-2: normalizing the denoised data by adopting a Min-Max method to eliminate dimensional difference between different data, wherein the Min-Max method has the following calculation formula:
Figure FDA0002255210710000021
wherein, x'minIn a single rowMinimum value in data, x'maxIs the maximum value in the single-column data, and x is the normalized data;
step 2-3: and selecting a data segment from the normalized data to establish a data set, and dividing the data set into a training set and a testing set according to a proportion.
4. The method for predicting the hybrid wind power based on the long-and-short-term memory neural network according to claim 1, wherein the process of establishing the hybrid wind power prediction model based on the long-and-short-term memory neural network in the step 3 is as follows:
step 3-1: building an LSTM network structure;
the LSTM network comprises 1 input layer, 5 hidden layers and 1 output layer, and the output layer adopts an identity function as a regression function; the operation principle of the LSTM network is shown in equations (3) to (7):
ft＝sigmoid(Wfxxt+Whfht-1+bf) (3)
kt＝sigmoid(Wkxxt+Wkhht-1+bk) (4)
gt＝tanh(Wgxxt+Wghht-1+bg) (5)
ot＝sigmoid(Woxxt+Wohht-1+bo) (6)
ht＝tanh(gt⊙kt+St-1⊙ft)⊙ot(7)
where ⊙ denotes the element multiplication, t denotes the sampling time when the subscript is t, xtFor data sampled at time t, ft,kt,gt,otAnd htRespectively the states of the forgetting gate, the input node, the output gate and the state unit at the time ttFor the updated state of the memory cell at time t, St-1For the memory cell state at time t-1, Wfx,Whf,Wkx,Wkh,Wgx,Wgh,WoxAnd WohRespectively corresponding to the gate and input xtAnd state unit intermediate output ht-1Weight matrices of the products, the weight matrices being diagonal matrices; bf,bk,bg,boRespectively the offset of the corresponding door; tanh represents an activation function;
finally, obtaining an LSTM model output formula:
G＝f(xt,Whf,Wkh,Woh,B) (8)
wherein G represents the long-short time memory neural network model output, Whf、Wkh、WohRespectively representing an input gate weight vector, a forgetting gate weight vector and an output gate weight vector, and B represents a bias;
step 3-2: optimizing parameters of a long-time and short-time memory neural network by adopting an interval enumeration method;
carrying out preliminary experiments on two parameters of the number of hidden layer nodes and the batch size by adopting an interval enumeration method to obtain a preselected interval of the two parameters; then, carrying out wind power prediction experiments by using a long-time memory neural network with different combinations of two parameters of the number of hidden layer nodes and the batch scale, and obtaining the optimal number of the hidden layer nodes and the batch scale modulus according to a prediction result;
step 3-3: and (3) bringing the preprocessed data set into an LSTM prediction model for training to obtain a predicted value G of the sample in the training set, wherein G is vector representation of the output predicted value.
5. The method for predicting the hybrid wind power based on the long-and-short-term memory neural network according to claim 1, wherein the model modification of the predicted output of the long-and-short-term memory neural network in the step 4 is as follows:
step 4-1: calculating a prediction error of a fan system data set;
Ei'＝Xi'-Gi(9)
wherein, Xi' is a preprocessed data set, GiObtaining the ith sample after training the LSTM modelPredicted value of (E), Ei' is the prediction error of the LSTM model;
step 4-2: establishing a prediction model of a prediction error;
Figure FDA0002255210710000031
wherein E isi' error value for model prediction, C constant, AiRepresenting the model coefficient, Xi' represents input parameters of a training set, and l is the number of data in a data vector;
step 4-3: fitting constant C and model coefficient A by least square methodiObtaining a relational expression of the model error value with respect to the input parameter;
step 4-4: predicting the prediction error of the model by using the relational expression obtained in the step 4-3;
and 4-5: and (4) carrying out correction calculation on the predicted value of the model by using the error predicted value obtained in the step (4-4):
Gi'＝Gi-Ei' (11)
wherein G isiShows the predicted value of the ith sample obtained after LSTM model training, Gi' denotes a corrected predicted value, i.e., a corrected LSTM model output vector.
CN201911050486.9A 2019-10-31 2019-10-31 Hybrid wind power prediction method based on long-time and short-time memory neural network Pending CN110826791A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911050486.9A CN110826791A (en) 2019-10-31 2019-10-31 Hybrid wind power prediction method based on long-time and short-time memory neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911050486.9A CN110826791A (en) 2019-10-31 2019-10-31 Hybrid wind power prediction method based on long-time and short-time memory neural network

Publications (1)

Publication Number Publication Date
CN110826791A true CN110826791A (en) 2020-02-21

Family

ID=69551772

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911050486.9A Pending CN110826791A (en) 2019-10-31 2019-10-31 Hybrid wind power prediction method based on long-time and short-time memory neural network

Country Status (1)

Country Link
CN (1) CN110826791A (en)

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111259570A (en) * 2020-02-26 2020-06-09 沈阳工业大学 Wind power plant active standby quantification method based on information physical system
CN111915007A (en) * 2020-07-29 2020-11-10 厦门大学 Magnetic resonance spectrum noise reduction method based on neural network
CN112365040A (en) * 2020-11-03 2021-02-12 哈尔滨工业大学 Short-term wind power prediction method based on multi-channel convolution neural network and time convolution network
CN112379274A (en) * 2020-11-16 2021-02-19 河南科技大学 Method for predicting residual life of power battery
CN112418406A (en) * 2020-12-09 2021-02-26 甘肃靖远航天风力发电有限公司 Wind power tower inclination angle missing data supplementing method based on SSA-LSTM model
CN113255219A (en) * 2021-05-28 2021-08-13 浙江浙能技术研究院有限公司 Generator bar vibration optimization method based on correction long-time and short-time memory neural network
CN113567836A (en) * 2021-07-23 2021-10-29 电子科技大学长三角研究院(湖州) Segmented prediction circuit aging system and method
CN113902228A (en) * 2021-12-09 2022-01-07 北京东润环能科技股份有限公司 Wind power prediction method based on stepwise regression and deep learning and related equipment
CN114035468A (en) * 2021-11-08 2022-02-11 山东理工大学 Predictive monitoring method and system for fan overhaul process based on XGboost algorithm
CN114897245A (en) * 2022-05-12 2022-08-12 国网湖北省电力有限公司电力科学研究院 Large-scale wind power ultra-short term power prediction error correction method based on long-short term memory neural network
CN115271225A (en) * 2022-08-05 2022-11-01 中国矿业大学(北京) Wind power-wind power modeling method based on wavelet denoising and neural network
CN116960960A (en) * 2023-07-24 2023-10-27 青岛哈尔滨工程大学创新发展中心 Short-term wind power prediction method for wind turbine

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103117546A (en) * 2013-02-28 2013-05-22 武汉大学 Ultrashort-term slide prediction method for wind power
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
CN109214575A (en) * 2018-09-12 2019-01-15 河海大学 A kind of super short-period wind power prediction technique based on small wavelength short-term memory network
CN109242143A (en) * 2018-07-31 2019-01-18 中国电力科学研究院有限公司 A kind of neural network wind power forecasting method and system
CN109615146A (en) * 2018-12-27 2019-04-12 东北大学 A kind of wind power prediction method when ultrashort based on deep learning
CN109802430A (en) * 2018-12-29 2019-05-24 上海电力学院 A kind of wind-powered electricity generation power grid control method based on LSTM-Attention network
CN110276472A (en) * 2019-05-21 2019-09-24 南方电网调峰调频发电有限公司 A kind of offshore wind farm power ultra-short term prediction method based on LSTM deep learning network
CN110309603A (en) * 2019-07-05 2019-10-08 华北电力大学（保定） A kind of short-term wind speed forecasting method and system based on wind speed characteristics

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103117546A (en) * 2013-02-28 2013-05-22 武汉大学 Ultrashort-term slide prediction method for wind power
CN109242143A (en) * 2018-07-31 2019-01-18 中国电力科学研究院有限公司 A kind of neural network wind power forecasting method and system
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
CN109214575A (en) * 2018-09-12 2019-01-15 河海大学 A kind of super short-period wind power prediction technique based on small wavelength short-term memory network
CN109615146A (en) * 2018-12-27 2019-04-12 东北大学 A kind of wind power prediction method when ultrashort based on deep learning
CN109802430A (en) * 2018-12-29 2019-05-24 上海电力学院 A kind of wind-powered electricity generation power grid control method based on LSTM-Attention network
CN110276472A (en) * 2019-05-21 2019-09-24 南方电网调峰调频发电有限公司 A kind of offshore wind farm power ultra-short term prediction method based on LSTM deep learning network
CN110309603A (en) * 2019-07-05 2019-10-08 华北电力大学（保定） A kind of short-term wind speed forecasting method and system based on wind speed characteristics

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
朱乔木 等: "基于长短期记忆网络的风电场发电功率超短期预测", 《电网技术》 *
王朋 等: "基于小波长短期记忆网络的风电功率超短期概率预测", 《南京信息工程大学学报(自然科学版)》 *

Cited By (20)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111259570A (en) * 2020-02-26 2020-06-09 沈阳工业大学 Wind power plant active standby quantification method based on information physical system
CN111259570B (en) * 2020-02-26 2023-10-27 沈阳工业大学 Wind farm active standby quantization method based on information physical system
CN111915007A (en) * 2020-07-29 2020-11-10 厦门大学 Magnetic resonance spectrum noise reduction method based on neural network
CN111915007B (en) * 2020-07-29 2022-06-14 厦门大学 Magnetic resonance spectrum noise reduction method based on neural network
CN112365040A (en) * 2020-11-03 2021-02-12 哈尔滨工业大学 Short-term wind power prediction method based on multi-channel convolution neural network and time convolution network
CN112379274A (en) * 2020-11-16 2021-02-19 河南科技大学 Method for predicting residual life of power battery
CN112418406B (en) * 2020-12-09 2022-06-07 甘肃靖远航天风力发电有限公司 Wind power tower inclination angle missing data supplementing method based on SSA-LSTM model
CN112418406A (en) * 2020-12-09 2021-02-26 甘肃靖远航天风力发电有限公司 Wind power tower inclination angle missing data supplementing method based on SSA-LSTM model
CN113255219A (en) * 2021-05-28 2021-08-13 浙江浙能技术研究院有限公司 Generator bar vibration optimization method based on correction long-time and short-time memory neural network
CN113567836A (en) * 2021-07-23 2021-10-29 电子科技大学长三角研究院(湖州) Segmented prediction circuit aging system and method
CN113567836B (en) * 2021-07-23 2023-12-26 电子科技大学长三角研究院(湖州) Segmented prediction circuit aging system and method
CN114035468A (en) * 2021-11-08 2022-02-11 山东理工大学 Predictive monitoring method and system for fan overhaul process based on XGboost algorithm
CN114035468B (en) * 2021-11-08 2024-05-28 山东理工大学 Method and system for predictively monitoring overhaul flow of fan based on XGBoost algorithm
CN113902228A (en) * 2021-12-09 2022-01-07 北京东润环能科技股份有限公司 Wind power prediction method based on stepwise regression and deep learning and related equipment
CN114897245A (en) * 2022-05-12 2022-08-12 国网湖北省电力有限公司电力科学研究院 Large-scale wind power ultra-short term power prediction error correction method based on long-short term memory neural network
CN114897245B (en) * 2022-05-12 2024-05-28 国网湖北省电力有限公司电力科学研究院 Method for correcting prediction error of ultra-short-term power of large-scale wind power based on long-term and short-term memory neural network
CN115271225A (en) * 2022-08-05 2022-11-01 中国矿业大学(北京) Wind power-wind power modeling method based on wavelet denoising and neural network
CN115271225B (en) * 2022-08-05 2023-07-21 中国矿业大学(北京) Wind power-wind power modeling method based on wavelet noise reduction and neural network
CN116960960A (en) * 2023-07-24 2023-10-27 青岛哈尔滨工程大学创新发展中心 Short-term wind power prediction method for wind turbine
CN116960960B (en) * 2023-07-24 2024-02-09 青岛哈尔滨工程大学创新发展中心 Short-term wind power prediction method for wind turbine

Similar Documents

Publication Publication Date Title
CN110826791A (en) Hybrid wind power prediction method based on long-time and short-time memory neural network
CN110705743B (en) New energy consumption electric quantity prediction method based on long-term and short-term memory neural network
CN110942194A (en) Wind power prediction error interval evaluation method based on TCN
CN111461463B (en) Short-term load prediction method, system and equipment based on TCN-BP
CN110766200A (en) Method for predicting generating power of wind turbine generator based on K-means mean clustering
CN111027775A (en) Step hydropower station generating capacity prediction method based on long-term and short-term memory network
CN109615124B (en) SCADA master station load prediction method based on deep learning
CN112669168B (en) Short-term prediction method for wind power
CN110555548A (en) ICS-ELM ultra-short-term wind power prediction method based on data mining original error correction
CN114861533A (en) Wind power ultra-short-term prediction method based on time convolution network
CN114757427A (en) Autoregressive corrected LSTM intelligent wind power plant ultra-short term power prediction method
CN111242353A (en) Wind power combined prediction modeling and prediction method
CN115511657A (en) Wind power output and photovoltaic output evaluation method based on combined prediction model
CN112712215A (en) Wind power plant power prediction method
CN116341717A (en) Wind speed prediction method based on error compensation
CN110458341B (en) Ultra-short-term wind power prediction method and system considering meteorological characteristics
CN116632820A (en) Wind power prediction method based on sequence extension VMD and wind speed fluctuation characteristics
CN111506868A (en) Ultrashort-term wind speed prediction method based on HHT weight optimization
CN113361782B (en) Photovoltaic power generation power short-term rolling prediction method based on improved MKPLS
CN114091647A (en) Solar 10.7 cm radio flow forecasting method based on BP neural network
CN112232570A (en) Forward active total electric quantity prediction method and device and readable storage medium
CN111476402A (en) Wind power generation capacity prediction method coupling meteorological information and EMD technology
CN116070768A (en) Short-term wind power prediction method based on data reconstruction and TCN-BiLSTM
CN114091782B (en) Medium-long term power load prediction method
CN115907131A (en) Method and system for building electric heating load prediction model in northern area

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
RJ01 Rejection of invention patent application after publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20200221