CN110738253A - A short-term wind power prediction method based on FCM and AFSA-Elman - Google Patents

Abstract

The invention discloses a short-term wind power prediction method based on FCM and AFSA-Elman, which relates to the field of wind power prediction and includes the following steps: cleaning and standardizing the historical data of the wind farm; selecting the input vector of the clustering model for the first time, using The fuzzy C clustering algorithm trains the fan samples and selects the appropriate clustering index; constructs a new input set according to the clustering index, and trains it again with FCM to obtain the division results of different clusters; The method of weighted aggregation of fan capacity is used to obtain the parameter value of the equivalent units of each cluster, and this parameter value can represent the corresponding cluster; according to the equivalent parameters, the AFSA-Elman prediction model of different clusters can be established, and the prediction results of different clusters can be obtained; The total predicted power of the entire wind farm can be achieved by capacity-weighting the predicted power of each fleet. It has achieved the effect of avoiding the occurrence of the "dimension disaster" of the power system and accurately and effectively predicting the short-term wind power.

Description

A short-term wind power prediction method based on FCM and AFSA-Elman

technical field

The invention relates to the field of wind power prediction, in particular to a short-term wind power prediction method based on FCM and AFSA-Elman.

Background technique

With the rapid economic development and tight energy supply, the world's energy structure has been transformed from a fossil energy system to a sustainable new energy system based on renewable energy. Therefore, all countries in the world attach great importance to the development of sustainable new energy. As a kind of renewable energy, wind energy has the characteristics of wide distribution, inexhaustibility and large reserves. Wind power plays a prominent role in reducing greenhouse gas emissions and slowing down global warming. Today, more than 100 countries in the world are vigorously promoting the use of wind energy. The wind power industry is developing rapidly, and the global wind power industry will continue to develop rapidly in the future. According to statistics from the World Wind Energy Association, the world's newly installed capacity reached 52,492 MW in 2017, 3.8% lower than the 54,642 MW newly installed capacity in 2016.

China is vigorously developing wind power generation under the promotion of the policy of developing clean energy. In order to promote the development of renewable energy and provide guidance for the operation and management of the power system, the work of high-precision wind power forecasting is very necessary. Wind power prediction is a short-term forecast of the output wind speed of the wind farm by using physical simulation calculation and scientific statistical methods according to the relevant data of the wind farm meteorological information, so as to predict the power generation of the wind farm.

Because the random fluctuation of wind will make wind energy have strong uncertainty, the research significance of wind power prediction is increasingly prominent under the background of the expanding wind farm scale and the difficulty of power grid dispatching. For the power sector, the extreme value of power forecast can help to modify and formulate grid dispatching strategies in a timely manner, which can not only effectively reduce the consumption of wind energy resources, but also effectively ensure the safe and stable operation of the national power system; from the power market, efficient power forecasting can Improve the evaluation index of wind power in the power market to ensure that more wind energy resources can be used; for the wind farm itself, maintenance personnel can selectively maintain the wind turbines according to the power curve predicted and simulated, which not only ensures that the wind turbines are protected from harsh conditions Weather damage can also minimize the loss of wind resources, thereby further improving the economic benefits of wind farms.

Therefore, in view of the complex terrain of the target wind farm and the numerous units in the field, a wind power prediction method is urgently needed to predict the wind power.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide a short-term wind power prediction method based on FCM and AFSA-Elman, which can better avoid the occurrence of the "dimension disaster" of the power system, and is useful for evaluating the mutual influence between large-capacity wind farms and power systems. It is of great significance and provides an accurate and effective method for short-term wind power prediction.

The above-mentioned technical purpose of the present invention is achieved through the following technical solutions:

A short-term wind power prediction method based on FCM and AFSA-Elman, including the following steps:

Step 1: Clean and standardize the historical data of the wind farm, and establish the observation data of each wind turbine in the target wind farm;

Step 2: Select the input vector of the clustering model for the first time, use the fuzzy C clustering algorithm to train the fan samples, and select the appropriate clustering index;

Step 3: Construct a new input set according to the clustering index, and train it with FCM again to obtain the division results of different clusters;

Step 4: Use the weighted aggregation method of fan capacity to obtain the parameter values of the equivalent units of each cluster of the various clusters of the cluster, and this parameter value can represent the corresponding cluster;

Step 5: According to the equivalent parameters, establish the AFSA-Elman prediction model of different fleets, and then the prediction results of different fleets can be obtained;

Step 6: Perform capacity weighting on the predicted power of each fleet to achieve the total predicted power of the entire wind farm.

Further, in step 1, the observation data includes the time series formed by the wind speed, wind direction, active power and temperature data of each fan.

计算基于空间度量特征的距离，其中，ρ_xy为相关系数，d_xy为基于空间度量特征的距离。Further, in step 2, the distance based on the spatial metric feature is used as the similarity measure of the clustering algorithm, and the correlation coefficient is appropriately transformed by introducing the idea of distance, and according to the formula

Calculate the distance based on the spatial metric feature, where ρ _xy is the correlation coefficient, and d _xy is the distance based on the spatial metric feature.

Further, in step 3, the three elements of wind speed, wind direction and active power are integrated as the grouping index.

Further, in step 5, the AFSA-Elman algorithm is used as the wind power prediction model, and the weight threshold of the Elman neural network is optimized by using the parallel processing of the artificial fish swarm algorithm and the characteristics of automatic global optimization.

To sum up, the present invention has the following beneficial effects:

1. The distance based on the spatial metric feature fully reflects the close relationship between the correlation and the distance. Through the transformation of the correlation coefficient, the distance of the spatial metric feature that can reflect the correlation between variables can be obtained. The research on the correlation of wind turbines more application value.

2. Compared with the unoptimized Elman algorithm and the commonly used BP algorithm, the AFSA-Elman algorithm has significantly smaller relative root mean square error (rRMSE) and relative mean absolute error (rMAE), and the error value is relatively stable, and the prediction curve is closer. on the actual power curve. Therefore, the effect of the proposed FCM-AFSA-Elman short-term wind power model is ideal.

Description of drawings

Fig. 1 is the fan distribution position diagram of the wind farm in the embodiment of the present invention;

Fig. 2 is the AFSA-Elman algorithm and the BP algorithm prediction curve contrast in the embodiment of the present invention;

Fig. 3 is AFSA-Elman algorithm and Elman algorithm prediction curve contrast in the embodiment of the present invention;

Fig. 4 is the absolute error diagram of BP algorithm prediction in the embodiment of the present invention;

Fig. 5 is the Elman algorithm prediction absolute error diagram in the embodiment of the present invention;

FIG. 6 is a graph of the absolute prediction error of the AFSA-Elman algorithm in an embodiment of the present invention.

Detailed ways

The specific embodiments of the present invention will be further described below with reference to the accompanying drawings, and this embodiment does not constitute a limitation to the present invention.

The invention discloses a short-term wind power forecasting method based on FCM and AFSA-Elman, which includes using FCM to cluster wind turbines, establishing an AFSA-Elman model for each cluster after the clustering, and combining the prediction results of each cluster. The final short-term wind power forecast result is obtained by superposition, which includes the following steps:

Step 1: Clean and standardize the wind farm historical data;

Among them, the observed data includes the time series formed by the wind speed, wind direction, active power and temperature data of each fan.

Step 2: Select the input vector (all feature quantities) of the clustering model for the first time, use the fuzzy C clustering algorithm based on the distance of the spatial metric feature to train the fan samples, and select the clustering results according to the influence of different clustering indicators on the clustering results. Appropriate clustering metrics;

Among them, the distance based on the spatial metric feature is used as the similarity measure of the clustering algorithm, and the correlation coefficient is appropriately transformed by introducing the idea of distance. The reference formula of the distance formula is:

The above formula mainly calculates the distance based on the spatial metric feature, where ρ _xy is the correlation coefficient, and d _xy is the distance based on the spatial metric feature. The application of d _xy to FCM can better solve the problem of Euclidean distance, which also solves the problem that the FCM clustering algorithm is only suitable for processing data with tight intra-class and good separation between classes and spherical data, but cannot process non-linear data. Problems with convexly shaped data.

Step 3: Construct a new input set according to the obtained clustering index, and train it with FCM again to obtain the division results of different clusters;

In this embodiment, since the influence of the temperature element on the clustering result is small, the temperature element is omitted, and the three elements of wind speed, wind direction and active power are integrated as the clustering index;

Step 4: Use the weighted aggregation method of fan capacity to obtain the equivalent parameters of the equivalent units of each cluster, and the parameter value can represent the corresponding cluster;

The clustering result judgment criterion uses the variance idea to define the measurement of intra-class distance and inter-class distance. The larger the inter-class distance, the better, and the smaller the intra-class distance, the better. The internal evaluation index reference formula of the clustering result is:

where STDI is the ratio of the inter-class distance to the intra-class distance, c _k is the centroid of cluster k, x ^t is the centroid of all samples, x _i is the ith sample of cluster k, n _k is the centroid of cluster k The number of samples, K is the number of clusters of the dataset.

Step 5: According to the equivalent parameters, establish AFSA-Elman prediction models of different fleets, and then the forecast results of different fleets can be obtained. Among them, the AFSA-Elman algorithm is used as the wind power prediction model, and the parallel processing and parallel processing of the artificial fish swarm algorithm are used. Automatically realize the characteristics of global optimization to optimize the weight threshold of Elman neural network;

In order to improve the prediction accuracy of the Elman algorithm, the artificial fish swarm algorithm (AFSA) is introduced to optimize the weight threshold of the Elman algorithm. The optimization process satisfies the following two equations:

为某时刻视点所在位置状态，Rand函数产生0到1之间的随机数，Step为步长，Visual为视野范围。Among them, M=(m ₁ , m ₂ ,..., m _n ) is the current state of the virtual artificial fish,

is the position state of the viewpoint at a certain moment, the Rand function generates a random number between 0 and 1, Step is the step size, and Visual is the field of view.

Step 6: Based on the AFSA-Elman prediction model of different clusters established in Step 5, the prediction results of different clusters can be obtained, and the predicted power of each cluster can be weighted by capacity to achieve the total predicted power of the entire wind farm.

As shown in Figure 1, taking Yunnan Modoushan Wind Farm as an example: Modoushan Wind Farm is located in low-latitude and high-altitude terrain. There are 24 wind turbines of the same model in the farm. Each wind turbine has an installed capacity of 2MW. The total farm capacity is 48MW. The cut-in wind speed of the fan is 3m/s, the rated wind speed is 12m/s, the cut-out wind speed is 25m/s, and the rated power is 2MW.

Because the wind farm is located in a special mountainous terrain, the location distribution of wind turbines and the spatial distribution of wind speed have a greater impact on the wind power output. It can be seen that it is necessary to conduct detailed modeling and analysis of the wind turbines in the wind farm.

The geographical distribution of wind turbines is shown in Figure 1, where No. 0 is the location of the wind measuring tower, and No. 1 to No. 24 represent the positions of No. 1 to No. 24 fans respectively. Due to the failure of No. 7 fan itself, it is impossible to obtain valid data accurately. , so the fan is not used as the research object. The present invention is further described below:

1. After removing invalid observation data, removing data outliers and normalizing the data of wind turbines, establish the observation data of each wind turbine in the target wind farm, which is expressed as D _i =[D _i1 ,D _i2 ,D _i3 , D _i4 ], where i∈[1,24], represents the fan number, and D _i1 , D _i2 , D _i3 , and D _i4 represent the time series formed by the fan wind speed, wind direction, active power and temperature data, respectively.

2. The wind farm is located in mountainous terrain, and the layout of wind turbines in the farm is irregular. Due to factors such as terrain, altitude, and the influence of other units, the operating data captured by each fan in the field is quite different, making the wind speed highly fluctuating and intermittent. analysis of the fan. The key to cluster analysis lies in the clustering index and similarity measurement.

(1) In the selection of wind turbine characteristic quantities, it should be ensured that not only can significantly affect the wind turbine power generation process, but also these characteristic indicators can be easily obtained before power generation occurs.

其中，ρ_xy为相关系数，d_xy为基于空间度量特征的距离，其对风机相关性的研究更具有应用价值。以基于空间度量特征的距离为相似性度量，分类为4类，采用模糊C均值聚类(FCM)算法对不同分群指标的情况进行分析，分析结果如表1所示。In order to fully reflect the close relationship between correlation and distance, by transforming the correlation coefficient, the distance of the spatial metric feature that can reflect the correlation between variables can be obtained.

Among them, ρ _xy is the correlation coefficient, and d _xy is the distance based on the spatial metric feature, which has more application value for the research on the correlation of wind turbines. Taking the distance based on the spatial metric feature as the similarity measure, it is classified into four categories, and the fuzzy C-means clustering (FCM) algorithm is used to analyze the situation of different clustering indicators. The analysis results are shown in Table 1.

Table 1 FCM clustering results table under different clustering indicators

It can be seen from Table 1 that under different clustering indicators, the clustering situation of the clusters is quite different. Among them, the influence of temperature on the clustering results is minimal, so we omit this factor and only study the influence of the three major factors of wind speed, wind direction, and active power. The clustering effect was analyzed with STDI as the clustering result evaluation index.

Using STDI as the clustering result evaluation index to analyze the clustering effect, as shown in Table 2. The STDI value of the comprehensive clustering index combined with wind speed, wind direction and active power is 0.7325, which is greater than the STDI value of wind speed, wind speed + wind direction as the clustering index, which proves that the former has a better clustering effect. Therefore, it is determined to use wind speed, wind direction and active power. The three elements are integrated into the classification index.

Table 2 STDI values under different indicators

Clustering Metrics STDI value wind speed 0.4329 wind speed + wind direction 0.5663 Wind speed + wind direction + active power 0.7325

(2) Similarity measurement is also an important link in cluster analysis. The Euclidean distance, the Pearson coefficient and the distance based on the spatial metric feature (Space distance) were selected as similarity measures for comparison, and the fuzzy C-means clustering was used as the clustering algorithm to cluster the wind turbines in the wind farm. The clustering results and STDI values are shown in Table 3.

Table 3 FCM clustering results table under different similarity measures

It can be seen from Table 3 that although the clustering methods are the same, different similarity measures lead to different groupings of clusters, indicating that similarity measures have a great influence on the division between clusters. From the STDI value, the clustering effect based on the distance based on the spatial metric feature is better than the other two types.

3. In order to further evaluate the effect of different similarity measures, the AFSA-Elman prediction model is used to conduct experiments on the data of different clusters. The dynamic information processing capability of Elman neural network makes it widely used in time series forecasting. In order to improve the prediction accuracy of Elman algorithm, artificial fish swarm algorithm (AFSA) is introduced to optimize the weight threshold of Elman algorithm. The optimization process is as follows:

为某时刻视点所在位置状态，Rand函数产生0到1之间的随机数，Step为步长，Visual为视野范围。Among them, M=(m ₁ , m ₂ ,..., m _n ) is the current state of the virtual artificial fish,

is the position state of the viewpoint at a certain moment, the Rand function generates a random number between 0 and 1, Step is the step size, and Visual is the field of view.

Select the data of the first 27 days of the wind farm in January, that is, 2592 sets of data as training samples to train the AFSA-Elman prediction model, and then use the trained model to predict the last three days, that is, 288 sets of data, and compare the predicted data with the measured data of the wind farm. Compare and find each error value. In order to evaluate the clustering results under different similarity measures, the AFSA-Elman model was used for training and testing.

The present invention selects four error indexes of root mean square error (RMSE), mean absolute error (MAE), relative root mean square error (rRMSE) and relative mean absolute error (rMAE) to evaluate the short-term wind speed prediction result of the wind farm. Its calculation formula is as follows:

(1) Root Mean Square Error (RMSE)

(2) Average relative error (MAE)

(3) Relative root mean square error (rRMSE)

(4) Relative mean absolute error (rMAE)

表示的是第i个预测值，x_k(i)表示的是第i个实际值。Among them, N represents the number of samples,

represents the ith predicted value, and x _k (i) represents the ith actual value.

The corresponding prediction error results under different similarity measures are shown in Table 4.

Table 4 Comparison of prediction accuracy under different similarity measures

It can be seen from Table 4 that the rRMSE and rMAE values of the model errors when the distance based on the spatial metric feature is used as the similarity measure are 22.25% and 16.53%, respectively, and the model error values when the Euclidean distance is used as the similarity measure are 25.41% and 19.04%. %, the model error values with the Pearson coefficient as the similarity measure are 23.48% and 17.85%. In contrast, the model accuracy based on the distance based on the spatial metric feature as the similarity measure is higher, and the effect of power prediction is better. it is good. The feasibility of the proposed similarity measurement method, namely distance based on spatial measurement features, is verified.

In order to further evaluate the prediction effect of the AFSA-Elman algorithm, the AFSA-Elman algorithm is compared with the unoptimized Elman algorithm and the commonly used BP neural network algorithm. The comparison of the prediction curves is shown in Figure 2 and Figure 3, respectively.

Due to the large amount of data, 48 data were extracted from the 288 data in the test set for specific analysis. Figure 2 compares the prediction curve of the AFSA-Elman algorithm, the prediction curve of the BP algorithm and the actual power curve. Figure 3 compares the prediction curve of the AFSA-Elman algorithm, the prediction curve of the Elman algorithm and the actual power curve. The predicted curve of the AFSA-Elman algorithm is closer to the actual power curve than the predicted curve of the BP algorithm and the Elman algorithm, and the fitting effect of the AFSA-Elman algorithm is better. Figure 4, Figure 5, and Figure 6 are the comparison charts of the absolute error between each predicted power value and the measured value in the corresponding time of the BP algorithm, Elman algorithm, and AFSA-Elman algorithm model. It can be seen from the comparison of the three figures that the absolute error of the AFSA-Elman model is significantly smaller than that of BP and Elman, and the error value is relatively stable. In order to understand the prediction effect of the three algorithms more intuitively, Table 5 lists various prediction error indicators under different algorithms.

Table 5 Comparison of prediction accuracy of several prediction and prediction algorithms

It can be seen from Table 5 that the prediction accuracy of the AFSA-Elman algorithm is significantly higher than that of the BP neural network algorithm and the Elman algorithm. For summer and autumn when the error index is the largest, the algorithm also has a good prediction effect, and the predicted power sequence is closer to the actual power sequence. It shows the effectiveness of the algorithm in power prediction, and it also has high accuracy compared with the more widely used algorithms at present. The analysis of experimental results shows that the prediction error of AFSA-Elman model is significantly smaller than that of Elman and BP, which verifies that the prediction effect of AFSA-Elman is better than that of Elman and BP algorithms. Therefore, the short-term wind power prediction method based on cluster analysis and AFSA-Elman algorithm proposed in this paper can better improve the prediction accuracy.

The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Those skilled in the art can make various modifications or equivalent replacements to the present invention within the spirit and protection scope of the present invention. Or equivalent replacement should also be regarded as falling within the protection scope of the technical solution of the present invention.

Claims (5)

1. a short-term wind power prediction method based on FCM and AFSA-Elman, is characterized in that: Include the following steps: Step 1: Clean and standardize the historical data of the wind farm, and establish the observation data of each wind turbine in the target wind farm; Step 2: Select the input vector of the clustering model for the first time, use the fuzzy C clustering algorithm to train the fan samples, and select the appropriate clustering index; Step 3: Construct a new input set according to the clustering index, and train it with FCM again to obtain the division results of different clusters; Step 4: Use the weighted aggregation method of fan capacity to obtain the parameter values of the equivalent units of each cluster of the various clusters of the cluster, and this parameter value can represent the corresponding cluster; Step 5: According to the equivalent parameters, establish the AFSA-Elman prediction model of different fleets, and then the prediction results of different fleets can be obtained; Step 6: Perform capacity weighting on the predicted power of each fleet to achieve the total predicted power of the entire wind farm. 2. a kind of short-term wind power prediction method based on FCM and AFSA-Elman according to claim 1, is characterized in that: In step 1, the observation data includes the time series formed by the wind speed, wind direction, active power and temperature data of each fan. 3. a kind of short-term wind power prediction method based on FCM and AFSA-Elman according to claim 1, is characterized in that: in step 2, adopt the distance based on spatial metric feature to be the similarity measure of clustering algorithm, by introducing The idea of distance transforms the correlation coefficient appropriately, according to the formula

Calculate the distance based on the spatial metric feature, where ρ _xy is the correlation coefficient, and d _xy is the distance based on the spatial metric feature. 4. A kind of short-term wind power prediction method based on FCM and AFSA-Elman according to claim 1, it is characterized in that: in step 3, take wind speed, wind direction and active power three kinds of element synthesis as grouping index. 5. a kind of short-term wind power prediction method based on FCM and AFSA-Elman according to claim 1, is characterized in that: step 5 adopts AFSA-Elman algorithm as wind power prediction model, utilizes the parallel processing of artificial fish swarm algorithm And automatically realize the characteristics of global optimization to optimize the weight threshold of Elman neural network.

Images