CN104899656A - Wind power combined predication method based on ensemble average empirical mode decomposition and improved Elman neural network - Google Patents

Abstract

The invention discloses a wind power combined predication method based on ensemble average empirical mode decomposition and an improved Elman neural network, and belongs to the technical field of wind power prediction. The method comprises the following steps: step 1, extracting wind speed sequence historical data and normalizing the data; step 2, making sequence analysis of the extracted wind speed sequence historical data through empirical mode decomposition; step 3, reconstructing a phase space for the sequences obtained through empirical mode decomposition; step 4, cyclically selecting the number of nodes in a hidden layer to train an Elman neural network, and superposing the prediction results of the sequences to obtain a wind speed prediction result; and step 5, making error analysis of the wind speed prediction result. The modeling process of the method is simple and practical. By adopting the method, the wind power can be predicted quickly and effectively. The method is of great significance to the security and stability and dispatching operation of a power system under the condition of wind power grid-connection.

Description

A kind of wind power combination Forecasting Methodology based on population mean empirical mode decomposition and Modified Elman Neural Network

Technical field

A kind of wind power combination Forecasting Methodology based on population mean empirical mode decomposition and Modified Elman Neural Network of the present invention, belongs to wind power prediction technical field.

Background technology

In today that energy problem becomes increasingly conspicuous, wind energy gets more and more people's extensive concerning as the widespread regenerative resource of one.Along with the day by day maturation of wind generating technology, the ratio that wind-powered electricity generation accounts for electric system generating total amount also increases gradually.But, the randomness of wind energy and the feature such as intermittent cause harmful effect to the safety of the quality of power supply and electrical network, stable operation, and wind power is predicted effectively to be the operating cost reducing above-mentioned impact and system, improve the effective means of wind power penetration limit, therefore, the short term prediction method studying wind power has important theoretical and practical significance.

The short-term forecasting of wind power generally adopts physical method and statistical method.Physical method mainly considers the technical parameter of weather data and Wind turbines, seek the wind speed optimal estimation value that Wind turbines takes turns firm At The Height, the last output power calculating wind energy turbine set according to the powertrace of wind energy turbine set, but the method too relies on perfect numerical weather forecast information, and therefore engineering practicability is lower.Statistical method does not consider the physical process that wind speed changes, but predicts Power Output for Wind Power Field according to a large amount of measured data.Statistics Forecasting Methodology conventional at present has lasting method, time series method, artificial neural network method, support vector machine method etc.This type of statistical method can meet accuracy requirement for the wind power prediction result of several hours in advance, but for shifting to an earlier date predicting the outcome of longer time, precision is inadequate.These methods deeply expose along with wind power technology the defect being difficult to overcome, and as precision of prediction is poor, speed of convergence is slow, has the shortcomings such as limitation.

Summary of the invention

For the deficiencies in the prior art, the invention provides a kind of wind power combination Forecasting Methodology based on population mean empirical mode decomposition and Modified Elman Neural Network, EEMD can by the time series tranquilization of complexity, obtain several intrinsic mode functions (intrinic model function, IMF) component, and it is separate between each component, simultaneously, as a kind of EMD method of improvement, white noise sequence adds in original series by the method, then carry out repeatedly EMD to it to decompose, to the average of the many groups IMF obtained be decomposed as its true component, thus avoid chaos overlapping phenomenon.

The technical scheme that the present invention takes is:

Based on a wind power combination Forecasting Methodology for population mean empirical mode decomposition and Modified Elman Neural Network, comprise the following steps:

Step 1: the acquisition of data and normalized.The present invention adopts the air speed data in front several moment or numerical weather forecast information to train, for predicting the wind speed of subsequent time as input data.The data obtained are normalized:

$\stackrel{\‾}{x_i}=\frac{x_i-x_{\min }}{x_{\max }-x_{\min }}---\left(1\right)$

In formula: it is the result after certain data normalization in training sample T; x _maxand x _minmaximal value and the minimum value of this group variable data in training sample T respectively;

Step 2: population mean empirical mode decomposition.White noise sequence is added in original series, then EMD is carried out repeatedly to it and decompose, will the average of the many groups IMF obtained be decomposed as its true component, thus avoid chaos overlapping phenomenon;

The determination of step 3:Elman neural network structure;

The non-linear state space expression formula of Elman neural network is:

y(k)＝g(w ³x(k)+b ₂) (2)

x(k)＝f(w ¹x _c(k)+w ²(u(k-1))+b ₁) (3)

x _c(k)＝x(k-1) (4)

In formula: k represents the moment, y, x, u, x _crepresent 1 dimension output node vector respectively, m ties up hidden layer node unit vector, and n ties up input vector and m ties up feedback states vector, w ³, w ², w ¹represent that hidden layer is to output layer, input layer to hidden layer, undertaking layer to the connection weight value matrix of hidden layer respectively, P _ifor the transport function of hidden layer neuron, b ₁and b ₂be respectively the threshold value of input layer and hidden layer;

Step 4: choose optimum node in hidden layer and predict;

Step 5: finally superpose predicting the outcome of each component and obtain final predicted data.

In step 2, EEMD concrete steps are as follows:

Step 2.1: initialization procedure, initialization EMD performs number of times M time, and each amplitude of adding noise is k;

Step 2.2: aweather power time series x (t) adds normal distribution white noise n (t), performs the m time EMD process:

1), find out all minimal values and the maximum value of sequence x (t), utilize its upper and lower envelope of cubic spline function difference matching;

2), mean value m (t) of upper and lower envelope, h (t)=x (t)-m (t) is calculated;

3), judge whether h (t) meets the definition of IMF, satisfied then using h (t) as first IMF, otherwise h (t) is repeated 1 as original series) and 2) until meet the definition of IMF;

4), IMF component is separated from original signal, r (t)=x (t)-h (t), and r (t) is repeated above-mentioned steps as new sequence, and obtain remaining IMF component and surplus, surplus is monotonic quantity;

Step 2.3: population mean computing: each IMF computation of mean values of obtaining is decomposed as net result to M EMD.

Comprise the following steps in step 3:

Step 3.1: the phase space reconfiguration of wind power time series:

To wind power time series { p (t) }, t=1,2,3 ... N, select a suitable Embedded dimensions m and time delay t, the new state space of formula (4) can be met, the wind power time series phase space namely reconstructed;

P(t)＝{p(t),p(t+τ),p(t+(m-1)τ)}(5)

Wherein: t=1,2,3 ... M, M=N-(m-1) τ;

Step 3.2: use C-C method to ask for delay time T and the Embedded dimensions m of phase space reconfiguration, choose the input number of nodes of the saturated embedding m in phase space reconfiguration as neural network;

Step 3.3: because the node in hidden layer of Elman neural network has larger impact to Elman Prediction Accuracy, uses the minimum method of training error to choose optimum nodes.

Step 3.3 concrete steps are as follows:

1), utilize phase space reconfiguration method to be reconstructed by wind power sequence, obtain input data and input number of nodes m thereof;

2), the span of hidden node number to be optimized is decided to be n ∈ [2m-4,2m+6];

3), circulation chooses node in hidden layer and trains Elman neural network, and choosing the minimum node of training error is optimum node.

Compared with the conventional method, a kind of wind power combination Forecasting Methodology based on population mean empirical mode decomposition and Modified Elman Neural Network of the present invention, has the following advantages and beneficial effect:

(1), for the non-linear and non-stationary feature of wind power time series, utilize EEMD to decompose the wind power sequence of complexity decomposed and obtain a series of relatively simple subcomponent, effectively can improve precision of prediction.

(2), Elman neural network has stronger nonlinear fitting ability, and obtains the parameter of model structure according to phase space reconfiguration and the minimum method of training error, can further improve precision of prediction.

(3), as can be seen from model of the present invention predict the outcome and with the comparative analysis of other models, prediction modeling approach of the present invention has certain advance, achieves good prediction effect.

Accompanying drawing explanation

Fig. 1 is Elman neural network structure of the present invention;

Fig. 2 is EEMD-Modified Elman Neural Network of the present invention prediction process flow diagram;

Fig. 3 is power in the embodiment of the present invention;

Fig. 4 is that in the embodiment of the present invention, EEMD-Modified Elman Neural Network predicts the outcome figure.

Embodiment

A kind of based on quadtatic low signal-to-noise ratio sinusoidal signal High Precision Frequency method, comprise the following steps:

Step 1: the acquisition of data and normalized: the present invention adopts the air speed data in front several moment or numerical weather forecast information to train, for predicting the wind speed of subsequent time as input data.The data obtained are normalized:

$\stackrel{\‾}{x_i}=\frac{x_i-x_{\min }}{x_{\max }-x_{\min }}---\left(6\right)$

In formula: it is the result after certain data normalization in training sample T; x _maxand x _minmaximal value and the minimum value of this group variable data in training sample T respectively.

Step 2: population mean empirical mode decomposition.White noise sequence is added in original series, then EMD is carried out repeatedly to it and decompose, will the average of the many groups IMF obtained be decomposed as its true component, thus avoid chaos overlapping phenomenon.

EEMD concrete steps are as follows:

Step 2.1: initialization procedure.Initialization EMD performs number of times M time, and each amplitude of adding noise is k.

Step 2.2: aweather power time series x (t) adds normal distribution white noise n (t), performs the m time EMD process:

1), find out all minimal values and the maximum value of sequence x (t), utilize its upper and lower envelope of cubic spline function difference matching.

2), mean value m (t) of upper and lower envelope, h (t)=x (t)-m (t) is calculated.

3), judge whether h (t) meets the definition of IMF, satisfied then using h (t) as first IMF, otherwise h (t) is repeated 1 as original series) and 2) until meet the definition of IMF.

4), IMF component is separated from original signal, r (t)=x (t)-h (t), and r (t) is repeated above-mentioned steps as new sequence, and obtain remaining IMF component and surplus, surplus is monotonic quantity.

Step 2.3: population mean computing: each IMF computation of mean values of obtaining is decomposed as net result to M EMD.

The determination of step 3:Elman neural network structure.

The non-linear state space expression formula of Elman neural network is:

y(k)＝g(w ³x(k)+b ₂) (7)

x(k)＝f(w ¹x _c(k)+w ²(u(k-1))+b ₁) (8)

x _c(k)＝x(k-1) (9)

In formula: k represents the moment, y, x, u, x _crepresent 1 dimension output node vector respectively, m ties up hidden layer node unit vector, and n ties up input vector and m ties up feedback states vector.W ³, w ², w ¹represent that hidden layer is to output layer, input layer to hidden layer, undertaking layer to the connection weight value matrix of hidden layer respectively.P _ifor the transport function of hidden layer neuron, b ₁and b ₂be respectively the threshold value of input layer and hidden layer.

Step 3.1: the phase space reconfiguration of wind power time series:

To wind power time series { p (t) }, t=1,2,3 ... N, select a suitable Embedded dimensions m and time delay t, the new state space of formula (4) can be met, the wind power time series phase space namely reconstructed.

P(t)＝{p(t),p(t+τ),p(t+(m-1)τ)} (10)

Wherein: t=1,2,3 ... M, M=N-(m-1) τ.

Step 3.2: use C-C method to ask for delay time T and the Embedded dimensions m of phase space reconfiguration.Choose the input number of nodes of the saturated embedding m in phase space reconfiguration as neural network.

Step 3.3: because the node in hidden layer of Elman neural network has larger impact to Elman Prediction Accuracy, uses the minimum method of training error to choose optimum nodes.

Its concrete steps are as follows:

1), utilize phase space reconfiguration method to be reconstructed by wind power sequence, obtain input data and input number of nodes m thereof.

2), the span of hidden node number to be optimized is decided to be n ∈ [2m-4,2m+6].

3), circulation chooses node in hidden layer and trains Elman neural network, and choosing the minimum node of training error is optimum node.

Step 4: choose optimum node in hidden layer and predict.

Step 5: finally superpose predicting the outcome of each component and obtain final predicted data.

Embodiment:

The present invention is with the actual measurement wind power data of certain wind energy turbine set unit, and the sampling period of these data is 10min, and the rated power of unit is 850kW.For studying conveniently, choosing and putting the least possible data segment stop time and carry out simulation analysis, choosing 360 continuous power data points, front 300 for training, rear 60 for testing and analysis.Its powertrace as shown in Figure 3.

Carrying out quantitative evaluation for the precision predicted the outcome and reliability is the important component part that prediction effect is analyzed.Conventional multiple prediction index is evaluated predicting the outcome, herein the following several method of main employing:

(1) square error (MSE)

$e_{\mathrm{MSN}}=\frac{1}{N}\sqrt{\underset{t=1}{\overset{N}{\mathrm{\Σ}}}{(A_i-P_i)}^2}---\left(11\right)$

(2) mean absolute percentage error (MAPE)

$e_{\mathrm{MAPE}}=\frac{1}{N}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\left|\frac{A_i-P_i}{A_i}\right|---\left(12\right)$

(3) all square percentage error (MSPE)

$e_{\mathrm{MSPE}}=\sqrt{\frac{1}{N}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{((A_i-P_i)/A_i)}^2}---\left(13\right)$

In formula: A _ibe the measured value of i-th future position, P _ibe the predicted value of i-th future position, N is the number of future position.

As seen from Figure 4, predicted value herein immediately following the variation tendency of actual value, can have higher fitting precision, thus demonstrate this paper put forward the validity of forecast model.

For further comparative study, the present invention also utilizes single BP, Elman neural network prediction model and EMD-Elman forecast model to carry out wind power prediction.Its predicated error index is as shown in table 2.

Table 2 model performance index

As shown in Table 2, compared with the performance index of other each forecast model, institute's model accuracy of carrying is higher herein, has certain advance.

The wind power sequence of non-stationary property is converted into a series of subsequence by EEMD, then predicts the subsequence with certain rule, reduces prediction difficulty, and therefore relative to single Forecasting Methodology, combination forecasting precision of prediction is higher herein; EEMD utilizes noisiness to avoid the aliasing of EMD, utilizes EEMD to decompose, and prediction effect also can be better.

The present invention is illustrated according to the preferred embodiment, but above-described embodiment does not limit the present invention in any form, the technical scheme that the form that all employings are equal to replacement or equivalent transformation obtains, and all drops in the scope of technical solution of the present invention.

Claims (4)

1., based on a wind power combination Forecasting Methodology for population mean empirical mode decomposition and Modified Elman Neural Network, it is characterized in that comprising the following steps:

Step 1: the acquisition of data and normalized: adopting the air speed data in front several moment or numerical weather forecast information to train as input data, for predicting the wind speed of subsequent time, the data obtained being normalized:

$\stackrel{\‾}{x_i}=\frac{x_i-x_{\min }}{x_{\max }-x_{\min }}---\left(1\right)$

In formula: it is the result after certain data normalization in training sample T; x _maxand x _minmaximal value and the minimum value of this group variable data in training sample T respectively;

Step 2: population mean empirical mode decomposition: added to by white noise sequence in original series, then carries out repeatedly EMD to it and decomposes, and will decompose the average of the many groups IMF obtained as its true component, thus avoids chaos overlapping phenomenon;

The determination of step 3:Elman neural network structure:

The non-linear state space expression formula of Elman neural network is:

y(k)＝g(w ³x(k)+b ₂) (2)

x(k)＝f(w ¹x _c(k)+w ²(u(k-1))+b ₁) (3)

x _c(k)＝x(k-1) (4)

In formula: k represents the moment, y, x, u, x _crepresent 1 dimension output node vector respectively, m ties up hidden layer node unit vector, and n ties up input vector and m ties up feedback states vector, w ³, w ², w ¹represent that hidden layer is to output layer, input layer to hidden layer, undertaking layer to the connection weight value matrix of hidden layer respectively, P _ifor the transport function of hidden layer neuron, b ₁and b ₂be respectively the threshold value of input layer and hidden layer;

Step 4: choose optimum node in hidden layer and predict;

Step 5: finally superpose predicting the outcome of each component and obtain final predicted data.

2. a kind of wind power combination Forecasting Methodology based on population mean empirical mode decomposition and Modified Elman Neural Network according to claim 1, it is characterized in that, in step 2, EEMD concrete steps are as follows:

Step 2.1: initialization procedure, initialization EMD performs number of times M time, and each amplitude of adding noise is k;

Step 2.2: aweather power time series x (t) adds normal distribution white noise n (t), performs the m time EMD process:

1), find out all minimal values and the maximum value of sequence x (t), utilize its upper and lower envelope of cubic spline function difference matching;

2), mean value m (t) of upper and lower envelope, h (t)=x (t)-m (t) is calculated;

3), judge whether h (t) meets the definition of IMF, satisfied then using h (t) as first IMF, otherwise h (t) is repeated 1 as original series) and 2) until meet the definition of IMF;

4), IMF component is separated from original signal, r (t)=x (t)-h (t), and r (t) is repeated above-mentioned steps as new sequence, and obtain remaining IMF component and surplus, surplus is monotonic quantity;

Step 2.3: population mean computing: each IMF computation of mean values of obtaining is decomposed as net result to M EMD.

3. a kind of wind power combination Forecasting Methodology based on population mean empirical mode decomposition and Modified Elman Neural Network according to claim 1, is characterized in that, comprise the following steps in step 3:

Step 3.1: the phase space reconfiguration of wind power time series:

To wind power time series { p (t) }, t=1,2,3 ... N, select a suitable Embedded dimensions m and time delay t, the new state space of formula (4) can be met, the wind power time series phase space namely reconstructed;

P(t)＝{p(t),p(t+τ),p(t+(m-1)τ)} (5)

Wherein: t=1,2,3 ... M, M=N-(m-1) τ;

Step 3.2: use C-C method to ask for delay time T and the Embedded dimensions m of phase space reconfiguration, choose the input number of nodes of the saturated embedding m in phase space reconfiguration as neural network;

Step 3.3: because the node in hidden layer of Elman neural network has larger impact to Elman Prediction Accuracy, uses the minimum method of training error to choose optimum nodes.

4. a kind of wind power combination Forecasting Methodology based on population mean empirical mode decomposition and Modified Elman Neural Network according to claim 3, it is characterized in that, step 3.3 concrete steps are as follows:

1), utilize phase space reconfiguration method to be reconstructed by wind power sequence, obtain input data and input number of nodes m thereof;

2), the span of hidden node number to be optimized is decided to be n ∈ [2m-4,2m+6];

3), circulation chooses node in hidden layer and trains Elman neural network, and choosing the minimum node of training error is optimum node.