CN104504466A - Wind power plant power prediction method considering atmospheric disturbance effect - Google Patents

Abstract

A wind farm power prediction method considering atmospheric disturbance effects, the method regards the atmospheric nonlinear Lorenz system as a disturbance model, defines atmospheric nonlinear disturbance variables to eliminate the nonlinear factors contained in the original wind speed data, and optimizes the power prediction model In order to achieve higher precision wind power prediction. On the basis of the conventional power prediction method, the present invention uses the power prediction disturbance formula to correct the wind speed prediction result by nonlinear interference, eliminates the influence of some nonlinear disturbance factors existing in the original wind speed on the prediction result, and achieves accurate prediction of wind power Purpose. Experimental results prove that this method can greatly improve the prediction accuracy of wind energy and greatly promote the development of wind power industry.

Description

Wind Farm Power Prediction Method Considering Atmospheric Disturbance Effect

technical field

The invention relates to a method capable of accurately predicting the wind speed of a wind farm, belonging to the technical field of power generation.

Background technique

The development and utilization of new energy is an important way to solve the current global energy shortage and the deterioration of the ecological environment. Wind energy is a clean renewable energy with abundant resources and wide distribution. Wind power generation can realize large-scale and effective utilization of wind energy resources. According to statistics from the Global Wind Energy Association, by the end of 2013, the cumulative installed wind power capacity in the world had reached 318,117MW, an increase of more than five times compared with the statistical data 10 years ago. During the operation of wind farms, wind is one of the most critical meteorological factors affecting power changes. The unique random fluctuation and intermittency of wind energy make the wind power generated by it have similar instability. With the continuous increase of wind power penetration power, wind power grid integration will bring risks to the stability and security of the power system, and then affect people's normal production and life. Therefore, the research and development of high-precision wind speed and power prediction technology has become an urgent task for the development of wind energy resources.

Existing wind power forecasting models can be divided into physical models, statistical models, artificial intelligence models and hybrid models according to different modeling methods. The physical model is to use some physical variables and geographical factors to improve the resolution of numerical weather prediction, which is suitable for long-term wind speed prediction. Statistical models are based on a large amount of historical data to establish the relationship between model input and output, generally including continuous method models, time series models, Kalman filter models, etc. Artificial intelligence is currently widely used wind energy prediction technology, usually including wavelet neural network (WNN), error backpropagation neural network (BP), radial basis neural network (RBF), support vector machine (SVM), fuzzy logic ( FL) and other methods. Since a single forecasting model has limitations to varying degrees, more and more combined models of multiple forecasting methods have been proposed and applied in recent years.

Most of the existing wind power prediction technologies rely on improving various numerical algorithms to improve the prediction accuracy. At present, there is no prediction method that takes into account the influence of the nonlinear characteristics of the atmospheric system on the power prediction results, so it is impossible to obtain higher wind power prediction results. prediction accuracy.

Contents of the invention

The purpose of the present invention is to provide a wind farm power prediction method considering the effect of atmospheric disturbance, so as to improve the prediction accuracy of wind energy.

Problem described in the present invention is realized with following technical scheme:

A wind farm power prediction method considering atmospheric disturbance effects, the method regards the atmospheric nonlinear Lorenz system as a disturbance model, defines atmospheric nonlinear disturbance variables to eliminate the nonlinear factors contained in the original wind speed data, and optimizes the power prediction model Input, so as to achieve higher precision wind power prediction, the method includes the following steps:

a. Collect the wind speed and power data of the wind farm at the set frequency, divide the collected data into training set and test set, and use the data of the training set to analyze the wavelet neural network (WNN) and error backpropagation network (BP) respectively. ) and support vector machine (SVM) three prediction models for training, and predict and verify the test set data;

b. Given initial conditions and parameter values, numerically solve the Lorenz equation;

c. The modulus length of each point in the phase space of the atmospheric nonlinear system is taken as the comprehensive disturbance variable L of the three disturbance variables in the Lorenz system, and the atmospheric nonlinear disturbance variable L is given by the following formula:

                                                                                    

Indicates that the atmospheric nonlinear disturbance system Momentary state of motion;

d. Using wavelet neural network (WNN), error backpropagation network (BP) and support vector machine (SVM) to predict wind speed respectively, the initial prediction results of the three prediction models are recorded as ;

e. Define the power prediction disturbance formula:

in, For the preliminary wind speed prediction results, is the correction value of the wind speed prediction result, is the disturbance coefficient.

The wind speed prediction result in step d according to the power prediction disturbance formula The nonlinear interference correction is carried out respectively, and the corrected prediction results are recorded as ;

f. Wind speed prediction data in steps d and e As input to the WNN model, As the input of the BP model, As the input of the SVM model, three sets of corresponding power prediction results are respectively obtained.

g. Specify the error index and perform error analysis on each prediction result.

The above-mentioned wind farm power prediction method considering the effect of atmospheric disturbance, when numerically solving the Lorenz equation, the given initial value condition is , the parameter value is .

The above wind farm power prediction method considering the effect of atmospheric disturbance specifies error indicators and performs error analysis on each prediction result. The error indicators include mean absolute error (MAE), mean square error (MSE) and absolute percentage error (MAPE), The calculation formula is respectively:

,

,

;

in and Respectively Observed and predicted values of wind speed or power at any time, Indicates the number of predicted samples;

Record the wind speed sample data in the forecast period as ; Separately predict the value and sample values The error results of WNN, BP and SVM models are calculated by bringing in the above three error indicators; similarly, the predicted values and sample values Bring the error results of LSWNN, LSBP and LSSVM models into the formula. See Table 1 for specific error statistics.

From the power prediction process in step f, three groups of different power prediction results can be obtained, which are denoted as , the power sample data in the forecast period is recorded as , the predicted value and sample values The error results of power prediction can be obtained by correspondingly bringing into the above three error formulas. The specific error statistics are shown in Table 2.

Based on the conventional power prediction method, the present invention uses the power prediction disturbance formula to correct the wind speed prediction result by nonlinear interference, eliminates the influence of some nonlinear disturbance factors existing in the original wind speed on the prediction result, and achieves accurate prediction of wind power Purpose. Experimental results prove that this method can greatly improve the prediction accuracy of wind energy and greatly promote the development of wind power industry.

Description of drawings

The present invention will be further described below in conjunction with accompanying drawing.

Figure 1 is the 4028 wind speed data of Sotavento Wind Farm in February 2014, the abscissa Indicates the observation time, and the ordinate indicates the wind speed data;

Fig. 2 is the distribution curve of atmospheric nonlinear disturbance quantity defined by formula (1);

Fig. 3 is the wind speed prediction curve of the WNN model and the LSWNN model, the abscissa is the prediction time, and the ordinate is the wind speed value;

Fig. 4 is the wind speed prediction curve of BP model and LSBP model, the abscissa is the prediction time, and the ordinate is the wind speed value;

Fig. 5 is the wind speed prediction curve of the SVM model and the LSSVM model, the abscissa is the prediction time, and the ordinate is the wind speed value;

Figure 6 is composed of the wind speed sequence The power prediction result of the WNN model as input;

Figure 7 is composed of the wind speed sequence The power prediction result of the BP model as the input quantity;

Figure 8 is composed of the wind speed sequence The power prediction result of the SVM model as input;

Figure 9 shows the distribution of atmospheric nonlinear disturbance variables applied in the prediction model LSWNN (disturbance coefficient is 0.0253);

Figure 10 shows the distribution of atmospheric nonlinear disturbance variables imposed in the prediction model LSBP (the disturbance coefficient is -0.0384);

Figure 11 shows the distribution of atmospheric nonlinear disturbance variables applied in the prediction model LSSVM (the disturbance coefficient is -0.0131);

Fig. 12 is a flowchart of the present invention.

The symbols in this paper are: L, atmospheric nonlinear disturbance variable, , Atmospheric nonlinear disturbance system No. momentary state of motion, , the wind speed prediction results of wavelet neural network (WNN), , the wind speed prediction results of the error backpropagation network (BP), , the wind speed prediction results of support vector machine (SVM), , the correction value of the wind speed prediction result of the wavelet neural network (WNN), , the correction value of the wind speed prediction result of the error backpropagation network (BP), , the correction value of the wind speed prediction result of the support vector machine (SVM), , Preliminary wind speed prediction results, , Correction value of wind speed prediction results, MAE, mean absolute error, MSE, mean square error, MAPE, absolute percentage error, , Observations of wind speed or power at any time, , The predicted value of wind speed or power at any time, , Disturbance coefficient, Indicates the number of forecast samples.

Detailed ways

The invention proposes a new idea and a new perspective on the wind farm power prediction method. Considering the influence of nonlinear factors in the atmospheric system on wind speed changes, the atmospheric nonlinear Lorenz system is regarded as a disturbance model, and the atmospheric nonlinear disturbance variables are defined to eliminate the nonlinear factors contained in the original wind speed, and optimize the input of the power prediction model. This enables higher-precision power prediction. Below in conjunction with embodiment the present invention is described in detail:

Step 1: This embodiment uses the wind speed and power data recorded every 10 minutes from Sotavento wind farm in Galicia from February 1, 2014 to February 28, and its sample size is 4028 (total sample size is 4032, due to There are four missing values in the original data caused by the wind tower or wind conditions, which are discarded and rearranged into 4028 data), which are divided into training set and test set, and the data of the training set are used to analyze the wavelet neural network respectively. (WNN), error backpropagation network (BP) and support vector machine (SVM) three prediction models are trained, and the test set data are predicted and verified;

Step 2: Initial Conditions , the parameter value , under the condition of numerically solving the Lorenz equation, a set of suitable atmospheric disturbance sequences can be obtained;

Step 3: Propose the concept of atmospheric nonlinear disturbance variable and give its expression form:

The modulus length of each point in the phase space of the atmospheric nonlinear system is taken as the comprehensive disturbance variable L of the three disturbance variables in the Lorenz system, and the atmospheric nonlinear disturbance variable L is given by the following formula:

(1)

Indicates that the atmospheric nonlinear disturbance system Momentary state of motion;

Step 4: Use the conventional prediction model wavelet neural network (WNN), error backpropagation network (BP) and support vector machine (SVM) to predict the wind speed in February 2014, and the prediction results are recorded as ;

Step 5: Define the power prediction disturbance formula:

(2)

in, For the preliminary wind speed prediction results, is the correction value of the wind speed prediction result, is the disturbance coefficient,

According to the power prediction disturbance formula, the wind speed prediction results in step 4 The nonlinear interference correction is carried out respectively, and the corrected prediction results are recorded as ;

Step 6: The above wind speed forecast data As input to the WNN model, As the input of the BP model, As the input of the SVM model, three sets of corresponding power prediction results are obtained respectively, so as to compare the prediction level and the accuracy of the prediction results of each model;

Step 7: Given a suitable error index, compare and analyze the wind speed and power prediction results of each model:

The error indicators include mean absolute error (MAE), mean square error (MSE) and absolute percentage error (MAPE), and the calculation formulas are respectively:

(3)

(4)

(5)

in and Respectively Observed and predicted values of wind speed or power at any time, Indicates the number of forecast samples.

Analysis of results

In this study, the method proposed by the present invention is verified through the wind speed and power data of Sotavento wind farm in Galicia in February 2014. The accompanying drawings show the main experimental results of the present invention. It is hereby noted that the following experimental analysis is only for demonstration, rather than limiting this method to a specific application environment.

First, Figure 1 shows the wind speed distribution curve of the Sotavento Wind Farm in February 2014. It can be seen that the wind speed fluctuates greatly, and the average wind speed reaches 12m/s. With the help of the numerical solution of the atmospheric nonlinear equation, the calculation is carried out according to the atmospheric nonlinear disturbance quantity formula defined in formula (1). As shown in Figure 2, the distribution of atmospheric nonlinear disturbances presents the characteristics of random fluctuations.

Secondly, the 4028 wind speed data in February are divided into training set and test set, and the WNN, BP and SVM models are trained respectively, and the test set data are predicted and verified. The prediction results are shown in Figure 3-Figure 5.

Thirdly, according to the above prediction results, the disturbance models LSWNN, LSBP and LSSVM corresponding to each model are proposed, and the disturbance models are used to eliminate the nonlinear factors in the wind speed prediction data of the WNN, BP and SVM models respectively, so as to achieve more accurate prediction of the actual wind speed data. fit. Figures 9 to 11 show the disturbance coefficients and atmospheric nonlinear disturbances in the correction formulas of each disturbance model, which are the preliminary prediction results Several sets of optimal results obtained through repeated training with sample data.

Finally, use the above wind speed prediction results to predict the power in the prediction time period. The power prediction model adopts WNN, BP and SVM models, and the input quantities are vs , vs and vs ,. The power prediction results are shown in Figure 6-Figure 8. As can be seen from Figures 6-8, using the wind speed sequence The result of power prediction is far better than that of using wind speed sequence to fit the actual power distribution curve The result of power prediction. The errors defined by equations (3)-(5) were used to calculate the above-mentioned wind speed and power prediction errors, and the results are shown in Table 1 and Table 2, respectively.

Table 1

Table 2

All the experimental results of this method strongly demonstrate that the introduction of atmospheric nonlinear disturbance variables can achieve more accurate power prediction, and the introduction of atmospheric nonlinear disturbance models in the field of wind energy forecasting research is feasible and has very important theoretical research value and practical guidance significance.

Claims (3)

1. A wind power plant power prediction method considering atmospheric disturbance effect is characterized in that an atmospheric nonlinear Lorenz system is taken as a disturbance model, an atmospheric nonlinear disturbance variable is defined to eliminate nonlinear factors contained in original wind speed data, and the input of a power prediction model is optimized, so that wind power prediction with higher precision is realized, and the method comprises the following steps:

a. acquiring wind speed and power data of a wind power plant at a set frequency, equally dividing the acquired data into a training set and a test set, respectively training three prediction models, namely a Wavelet Neural Network (WNN), an error back propagation network (BP) and a Support Vector Machine (SVM), by using the data of the training set, and predicting and verifying the data of the test set;

b. giving initial value conditions and parameter values, and numerically solving a Lorenz equation;

c. taking the mode length of each point in the atmospheric nonlinear system phase space as a comprehensive disturbance variable L of three disturbance variables in the Lorenz system, wherein the atmospheric nonlinear disturbance variable L is given by the following formula:

system for representing atmospheric nonlinear disturbancesA motion state at a time;

d. respectively predicting the wind speed by utilizing a Wavelet Neural Network (WNN), an error back propagation network (BP) and a Support Vector Machine (SVM), and respectively recording the prediction results of the three prediction models as VD1, VD2 and VD 3;

e. defining a power prediction disturbance formula:

wherein,in order to be the result of the preliminary wind speed prediction,for the correction value of the wind speed prediction result,in order to be able to make the coefficients of the perturbations,

d, forecasting the wind speed in the step d according to a power forecasting disturbance formulaRespectively carrying out nonlinear interference correction, and respectively recording the corrected prediction results；

f. Predicting the wind speedAs an input to the WNN model,as an input to the BP model, there is,respectively obtaining three groups of corresponding power prediction results as the input of an SVM model;

g. and specifying an error index and carrying out error analysis on each prediction result.

2. The method for predicting the power of the wind power plant by considering the atmospheric disturbance effect as claimed in claim 1, wherein the given initial value condition when the Lorenz equation is solved numerically isThe parameter is taken as。

3. A wind farm power prediction method considering atmospheric disturbance effect according to claim 2, characterized in that error indicators are specified and error analysis is performed on each prediction result, the error indicators comprise Mean Absolute Error (MAE), Mean Square Error (MSE) and absolute percent error (MAPE), and the calculation formulas are as follows:

；

wherein,andrespectively representThe observed and predicted values of wind speed or power at the moment,representing the number of predicted samples;

recording wind speed sample data in a prediction periodRespectively predict the valuesSum sample valueCarry in the three error indicatorsCalculating error results of the WNN, the BP and the SVM models; in the same way, the predicted values are respectively calculatedSum sample valueSubstituting a formula to obtain error results of LSWNN, LSBP and LSSVM models;

f, three groups of different power prediction results can be obtained in the power prediction process in the step f and are respectively recorded asAnd power sample data in the prediction period is recorded asWill predict the valueSum sample valueAnd correspondingly substituting the three error formulas to obtain an error result of the power prediction.