CN111191815B - Ultra-short-term output prediction method and system for wind power cluster - Google Patents

Abstract

Embodiments of the present invention provide an ultra-short-term output forecasting method and system for a wind power cluster. The method includes: dividing a wind farm of a wind power cluster to be predicted into sub-regions to obtain a combination of sub-region divisions of the wind power cluster to be predicted; obtaining the sub-regions Divide and combine the real power value of each sub-region in the first preset historical time period and the power real value of the second preset historical time period; according to the power real value of the second preset historical time period, the first preset historical Predict the power of each sub-region in the time period, and obtain the cluster power prediction value under each division and combination of the first preset historical period; obtain the real power value of the first preset historical period and the cluster power under each division and combination. For the error between the predicted values, the sub-regional division and combination corresponding to the smallest error are used as the optimal sub-regional division and combination to obtain the ultra-short-term output forecast of the wind power cluster. The embodiments of the present invention improve the accuracy of ultra-short-term output prediction of wind power clusters.

Description

An ultra-short-term output forecasting method and system for wind power clusters

technical field

The invention relates to the technical field of wind power data processing, in particular to a method and system for forecasting ultra-short-term output for wind power clusters.

Background technique

The existing ultra-short-term forecasting method for wind power clusters is mainly the "accumulation method", which is derived from the power prediction of a single wind farm. The forecast results of the wind farms are summed to obtain the forecast results of the wind power cluster.

The "accumulation method" is a very intuitive prediction method, and its calculation principle is exactly the same as that of wind farm prediction, but there are two defects. First of all, wind farms are more volatile than wind power clusters, and the prediction accuracy of wind power clusters based on the "accumulation method" is easily limited by the prediction accuracy of wind farms, especially some wind farms have output data with strong volatility due to sudden changes in weather processes. , the reduction of its prediction quality will greatly affect the prediction quality of the overall cluster. Secondly, the "accumulation method" separates each wind farm in the wind power cluster and makes predictions separately, without considering the correlation of the output of the wind farms in the cluster and the smoothness of the regional wind power output, and abandoning the original rich data in the wind power cluster information.

Therefore, there is an urgent need for an ultra-short-term output forecasting method and system for wind power clusters to solve the above problems.

SUMMARY OF THE INVENTION

Aiming at the problems existing in the prior art, embodiments of the present invention provide a method and system for predicting ultra-short-term output for a wind power cluster.

In a first aspect, an embodiment of the present invention provides an ultra-short-term output forecasting method for a wind power cluster, including:

Sub-regional division is performed on the wind farm of the to-be-predicted wind power cluster, and all sub-region division combinations of the to-be-predicted wind power cluster are obtained;

Obtain the real power value of each sub-region in the first preset historical time period and the power real value of the second preset historical time period in all sub-region division combinations, and the first preset historical time period is the latest at the current moment. The historical period of time, the second preset historical period is the most recent historical period of the first preset historical period;

According to the actual power value of the second preset historical time period, the power of each sub-region of the first preset historical time period is predicted to obtain the different division and combination forms of the first preset historical time period. The predicted value of cluster power;

Obtain the error between the actual power value of the first preset historical time period and the predicted value of the cluster power under various division and combination forms, and use the sub-region division and combination corresponding to the minimum error as the optimal sub-region division and combination, so that according to the The optimal sub-area division and combination are used to obtain the ultra-short-term output forecast value of the wind power cluster to be predicted.

Further, according to the actual power value of the second preset historical time period, predict the power of each sub-region of the first preset historical time period, and obtain the power of the first preset historical time period. Before the cluster power prediction value under various division and combination forms, the method further includes:

obtaining a training data set, the training data set including sample power data of each sub-region of the wind power cluster;

Through the training data set, the support vector machine model is trained to obtain a fitting model function for output prediction of each sub-region, so as to obtain the fitting model function according to the actual power value of the second preset historical time period, through the fitting The model function predicts the power of each sub-region in the first preset historical time period, and obtains the cluster power prediction value under various division and combination forms of the first preset historical time period.

Further, the acquisition of the training data set specifically includes:

Divide the sample power data through the sliding time window, and obtain the sample power output characteristic data of the sample power data based on the sliding time window;

According to the sample power data and the sample power output characteristic data corresponding to the sample power data, a training data set is constructed.

Further, in the obtaining of the error between the actual power value of the first preset historical time period and the predicted value of the cluster power under various division and combination forms, the sub-region division and combination corresponding to the minimum error are used as the optimal sub-region. After dividing and combining, the method further includes:

Obtain the optimal sub-region division combination of the prediction time scale corresponding to each historical time point in the first preset historical time period;

According to the actual power value of each sub-region under the first historical time point and the fitted model function, obtain the power prediction value of each sub-region corresponding to each historical time point in the future time;

According to the power prediction value of each sub-region at the future time, the optimal sub-region division and combination corresponding to the prediction time scale at each historical time point is used as the division and combination method at the current prediction time, so as to obtain the ultra-short-term output of the wind power cluster to be predicted. Predictive value.

Further, the number of historical moment points in the first preset historical time period is not more than 16, and the time interval of each historical moment point is 15 minutes.

Further, the wind farms of the to-be-predicted wind power cluster are divided into sub-regions, and all sub-region division combinations of the to-be-predicted wind power cluster are obtained, including:

Based on the Bell number formula, the wind farms of the wind power cluster to be predicted are divided into sub-regions, and all sub-region division combinations are obtained.

In a second aspect, an embodiment of the present invention provides an ultra-short-term output forecasting system for a wind power cluster, including:

The wind power cluster sub-region division module is used for sub-region division of the wind farms of the wind power cluster to be predicted, to obtain all the sub-region division combinations of the to-be-predicted wind power cluster;

The acquisition module is used to acquire the real power value of each sub-region in the first preset historical time period and the real power value of the second preset historical time period of each sub-region in all sub-region division combinations. The first preset historical time period The segment is the most recent historical time segment at the current moment, and the second preset historical time segment is the most recent historical time segment of the first preset historical time segment;

A power prediction module, configured to predict the power of each sub-region of the first preset historical time period according to the actual power value of the second preset historical time period, and obtain the Cluster power prediction value under various division and combination forms;

The ultra-short-term output prediction module is used to obtain the error between the real power value of the first preset historical time period and the cluster power prediction value under various division and combination forms, and the sub-region division and combination corresponding to the minimum error is regarded as the optimal Sub-area division and combination, so as to obtain the ultra-short-term output predicted value of the wind power cluster to be predicted according to the optimal sub-area division and combination.

Further, the system also includes:

a training data set building module for obtaining a training data set, the training data set including sample power data of each sub-region of the wind power cluster;

The fitting model training module is used to train the support vector machine model through the training data set, so as to obtain the fitting model function for prediction of each sub-region, so that the power of the second preset historical time period can be real value, predict the power of each sub-region in the first preset historical time period by using the fitting model function, and obtain the cluster power prediction value under various division and combination forms of the first preset historical time period.

In a third aspect, an embodiment of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, the processor implementing the program as described in the first aspect when the processor executes the program Steps of the provided method.

In a fourth aspect, an embodiment of the present invention provides a non-transitory computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, implements the steps of the method provided in the first aspect.

An ultra-short-term output forecasting method and system for a wind power cluster provided by an embodiment of the present invention performs prediction modeling based on sub-regions composed of multiple wind farms, and fully considers the relationship between the wind farms and the internal area of the wind power cluster. Based on the characteristic nature of the wind power output, the ultra-short-term output of the wind power cluster is predicted, which improves the accuracy of the output forecast.

Description of drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the accompanying drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description These are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained according to these drawings without creative efforts.

FIG. 1 is a schematic flowchart of an ultra-short-term output forecasting method for a wind power cluster according to an embodiment of the present invention;

FIG. 2 is a schematic diagram of a sequence of optimal division provided by an embodiment of the present invention;

3 is a schematic time sequence diagram of continuous division under ultra-short-term prediction provided by an embodiment of the present invention;

4 is a schematic structural diagram of an ultra-short-term output prediction system for wind power clusters provided by an embodiment of the present invention;

FIG. 5 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.

Detailed ways

In order to make the purposes, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments These are some embodiments of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

In the field of wind power output forecasting, wind power output is forecasted through three basic classification directions, namely forecasting method, forecasting object and forecasting time range. From the perspective of forecasting methods, wind power forecasting can be divided into physical model forecasting and statistical model forecasting. Among them, physical model forecasting is based on the idea of fluid mechanics, through the wind speed, air pressure and temperature on the boundary of a certain area, according to partial differential equations Iteratively obtains the relevant weather information of the wind turbines in the area, thereby obtaining the future wind power output; the statistical model prediction uses the past historical data of the prediction object to learn, and uses the basic regression model or artificial intelligence algorithm for model training, and input in the real-time prediction. The historical output data for a period of time before the current moment can be used to obtain future forecast results in a data-driven manner. Physical model prediction is not easy to obtain accurate results due to the influence of terrain and data accuracy. In contrast, statistical model prediction often has higher accuracy and faster computing speed on short time scales.

From the perspective of the forecast object, wind power forecast can be divided into wind power cluster forecast, wind farm forecast and wind turbine forecast, and the corresponding wind power installed capacity gradually decreases. Since the wind power output is affected by the spatial distribution, when the installed capacity is large, the output volatility will be reduced due to the diversity of the space weather system, achieving a "smoothing effect". Therefore, the output of wind power clusters tends to have a lower level of fluctuation, and its prediction accuracy tends to be higher than the latter two.

From the perspective of forecasting time range, wind power forecasting can be divided into ultra-short-term forecasting, short-term forecasting and mid- and long-term forecasting. According to the existing wind power forecasting standards, the ultra-short-term forecast refers to the forecast for 0-4 hours in the future, and the forecast interval is 15 minutes, that is, a forecast outputs 16-step forecast results; Prediction. Since the weather system is a chaotic system, with the increase of the prediction time range, the prediction accuracy decreases accordingly. Therefore, wind power prediction based on ultra-short term has high prediction accuracy, and its main purpose is to provide wind power output basis for real-time dispatching system.

Support vector machine regression is an ∈-insensitive loss regression method. Different from calculating the error between the regression value and the actual value in the traditional regression method, when the absolute value of the error specified by the support vector machine is less than ∈, the error is considered to be zero, and it is ignored. Compared with the traditional regression method, the support vector machine regression is greatly improved. The sparsity of the solution can reduce the computational difficulty, improve the computational speed, and ensure the generalization ability of the model. The objective function corresponding to the support vector machine adopted in the embodiment of the present invention is:

st|y _i -(w ^T x _i +b)|≤ε+ξ _i ;

ξ _i ≥ 0

表示误差项，该误差项只考虑绝对值大于∈的情况；

表示正则化项，用于保证预测模型的泛化能力；C表示目标函数两项之间的取舍关系，w，b表示待求解目标函数的权重参数；∈表示误差带的大小，为待优化参数；x_i，y_i表示样本特征。本发明实施例基于支持向量机回归构建的预测模型，对风电集群中各种组合下的风电集群子区域历史功率进行预测，分析判断得到不同历史时刻的最优子区域划分组合，从而根据这些最优子区域划分组合对风电集群进行超短期出力预测。in,

represents the error term, which only considers the case where the absolute value is greater than ∈;

Represents the regularization term, which is used to ensure the generalization ability of the prediction model; C represents the trade-off relationship between the two terms of the objective function, w, b represent the weight parameters of the objective function to be solved; ∈ represents the size of the error band, which is the parameter to be optimized ; x _i , y _i represent sample features. The embodiment of the present invention predicts the historical power of the sub-regions of the wind power cluster under various combinations in the wind power cluster based on the prediction model constructed by the support vector machine regression, and analyzes and judges to obtain the optimal sub-region division and combination at different historical moments, so as to obtain the optimal sub-region division and combination at different historical times, so as to obtain the optimal sub-region division and combination at different historical times. The sub-regional division and combination are used to forecast ultra-short-term output of wind power clusters.

FIG. 1 is a schematic flowchart of an ultra-short-term output prediction method for wind power clusters provided by an embodiment of the present invention. As shown in FIG. 1 , an embodiment of the present invention provides an ultra-short-term output prediction method for wind power clusters, including:

Step 101 , sub-region division is performed on the wind farm of the wind power cluster to be predicted, and all sub-region division combinations of the to-be-predicted wind power cluster are obtained;

In the embodiment of the present invention, the wind farm is affected by the terrain and the regional weather system, and in the wind power cluster formed by the wind farm, there is a regional feature characterized by the sub-regional wind power output as the main body. Since the weather system in a larger area has a certain inertia, the total wind power output in such sub-regions often has relatively regular output cycle characteristics. precision. Considering the limitations of actual wind power cluster construction and data communication, the embodiment of the present invention takes a wind farm as a basic unit, a sub-area is expressed as a combination of multiple wind farms, and a wind power cluster is expressed as a combination of multiple sub-areas, so that the wind farm, Sub-regions and wind power clusters constitute a three-level main body, and constitute various division methods of wind power clusters. Specifically, in the embodiment of the present invention, it is defined that the sub-region ci is formed by the wind farm _{j ip} , and is expressed as:

c _i ={j _i1 ,j _i2 ,...,j _im },j _ip ∈N;

Among them, c _i represents the ith sub-region, j _ip represents the p-th wind farm in the ith sub-region with a total of m wind farms, 0<p<m; N is the wind power cluster, indicating that there are N wind farms in the wind power cluster . Then the wind power output of a sub-region is the sum of the outputs of all wind farms in the sub-region, which is expressed as:

表示子区域c_i在t时刻的风电出力。在本发明实施例中，风电集群的子区域属于风电集群的一个非空子集，对于有N个风电场的风电集群，其可能组成的子区域个数为2^N－1个。in,

represents the wind power output of sub-region c _i at time t. In the embodiment of the present invention, the sub-regions of the wind power cluster belong to a non-empty subset of the wind power cluster. For a wind power cluster with N wind farms, the number of possible sub-regions is 2 ^N −1.

Further, in the embodiment of the present invention, the sub-region division of the wind power cluster is defined. The wind power cluster is composed of sub-regions, and the division of the wind power cluster is the different combinations of the sub-regions. In a combination of divisions, these sub-regions in the wind power cluster do not overlap each other, that is, each sub-region does not contain the same wind power. The wind farms in all sub-regions include all wind farms in the wind power cluster N, which are expressed as:

Among them, C ^S represents the division of the s-th seed area, and c _si represents the i-th sub-area in the division of the s-th seed area of the wind power cluster, then the wind power output of the wind power cluster is the sum of the outputs of all sub-areas within the wind power cluster, expressed as:

Among them, P ^t represents the wind power output of the wind power cluster at time t. Specifically, in the embodiment of the present invention, based on the Bell number formula, the wind farms of the to-be-predicted wind power cluster are divided into sub-regions, and all sub-region division combinations are obtained. Since all the numbers of wind power clusters are divided into a Bell number, showing order divergence, for a wind power cluster including N wind farms, the Bell number formula is expressed as B _N . For example, there are 3 wind farms in a wind power cluster in a certain area. , the division form of the wind power cluster can be referred to as shown in Figure 1.

Table 1

Cluster partition number Cluster division form 1 {[1],[2],[3]} 2 {[1,2],[3]} 3 {[1],[2,3]} 4 {[1,3],[2]} 5 {[1,2,3]}

From the cluster division methods in Table 1, it can be seen that there are multiple ways to divide the sub-regions in the wind power cluster. Each division can add the processing predicted values of the sub-regions to obtain the overall predicted value of the wind power cluster. Since the selection of different sub-regions represents the utilization of different cluster-related characteristics, only selecting the most appropriate clustering method can improve the accuracy. Therefore, the embodiment of the present invention transforms the problem of wind power output prediction into selecting an appropriate cluster division method. In the embodiment of the present invention, based on the prediction algorithm of the division of wind power clusters, a prediction model _fci is constructed for the output of sub-regions in the wind power cluster, and then the predicted wind power output values of the sub-regions are added to obtain the output prediction of the wind power cluster. , the specific formula is:

表示t₀时刻第i个子区域在未来第k时刻的风电出力预测值，表示t₀时刻风电集群在未来第k时刻的风电集群预测值，k表示预测时间范围。in,

Represents the predicted value of wind power output of the i-th sub-region at time k in the future at time t ₀ , and represents the predicted value of the wind power cluster at time k in the future for the wind power cluster at time t ₀ , and k represents the prediction time range.

The embodiment of the present invention fully considers the output characteristics of the sub-regions of the wind power cluster, and the division based on the sub-regions reflects the regular periodicity, that is, the regional characteristics, and the smooth processing effect is achieved through the larger spatial range of the sub-regions. Compared with a single wind farm Prediction reduces the volatility of output and improves the accuracy of prediction.

Step 102: Obtain the real power value of each sub-region in the first preset historical time period and the power real value of the second preset historical time period in all sub-region division combinations, and the first preset historical time period is: The most recent historical time period at the current moment, the second preset historical time period is the most recent historical time period of the first preset historical time period;

Step 103: Predict the power of each sub-region of the first preset historical time period according to the real power value of the second preset historical time period, and obtain various divisions of the first preset historical time period Predicted value of cluster power in combined form;

Step 104: Obtain the error between the actual power value of the first preset historical time period and the cluster power prediction value under various division and combination forms, and use the sub-region division and combination corresponding to the minimum error as the optimal sub-region division and combination. , so as to obtain the ultra-short-term output predicted value of the wind power cluster to be predicted according to the optimal sub-region division and combination.

An ultra-short-term output prediction method for a wind power cluster provided by an embodiment of the present invention performs prediction modeling based on sub-regions composed of multiple wind farms, and fully considers the relationship between wind farms and the regional output within the wind power cluster. The characteristic properties of the wind power cluster can be used to predict the ultra-short-term output of the wind power cluster, which improves the accuracy of the output prediction.

On the basis of the above-mentioned embodiment, according to the actual power value of the second preset historical time period, the power of each sub-region in the first preset historical time period is predicted to obtain the first predicted power. Before setting the cluster power prediction value under various division and combination forms of the historical time period, the method further includes:

obtaining a training data set, the training data set including sample power data of each sub-region of the wind power cluster;

Through the training data set, the support vector machine model is trained to obtain a fitting model function for output prediction of each sub-region, so as to obtain the fitting model function according to the actual power value of the second preset historical time period, through the fitting The model function predicts the power of each sub-region in the first preset historical time period, and obtains the cluster power prediction value under various division and combination forms of the first preset historical time period.

On the basis of the foregoing embodiment, the obtaining of the training data set specifically includes:

Divide the sample power data through the sliding time window, and obtain the sample power output characteristic data of the sample power data based on the sliding time window;

According to the sample power data and the sample power output characteristic data corresponding to the sample power data, a training data set is constructed.

In the embodiment of the present invention, firstly, by constructing a sample data set, the support vector machine model is trained, and the specific steps are:

其中，

表示模型训练时期的

时刻；Step S10: Obtain the sample power historical data of each wind farm in the wind power cluster, and calculate the sample power data of the sub-regions c _i under different division forms

in,

represents the model training epoch

time;

时刻的子区域c_i的样本功率数据

滚动向后划分样本数据集，得到

时刻的样本

其中，

表示输入到滑动时间窗的特征数据，

表示滑动时间窗输出的样本功率特征数据；Step S11, set the length of the sliding time window to L, and set the length of the sliding time window to

The sample power data of the sub-region c _i at the moment

Rolling backwards to divide the sample dataset, we get

sample of moments

in,

represents the feature data input to the sliding time window,

Represents the sample power characteristic data output by the sliding time window;

在1到T₀之间所有子区域的样本，其中子区域c_i的所有数据集表示为

Step S12, by traversing the sample power historical data according to time and all possible divisions of sub-regions, to obtain the time

The samples of all subregions between 1 and _T0 , where all datasets of subregion _ci are denoted as

得到的拟合模型函数为

In step _S13 , all possible sub-region sample data are respectively input into the support vector machine model for training, so as to obtain a trained model, that is, a fitting model function for the sub-region output prediction is obtained; data set

The resulting fitted model function is

In the embodiment of the present invention, a data set of historical power feature data is constructed by using a sliding time window. The length of the sliding time window represents the number of features of the input model, and the model outputs 16 points, representing that the ultra-short-term prediction is ahead of 0- 4h corresponds to 16 steps. Through the fitting model function, the L output data before the sub-region time t are used as the input of the model, and the prediction results of the next 16 steps at the time t are obtained. The forecast results of the corresponding forecast time ranges of each sub-region in the cluster division are added to obtain the ultra-short-term forecast results of the wind power cluster.

On the basis of the above-mentioned embodiment, according to the error between the obtained actual power value of the first preset historical time period and the predicted value of cluster power under various division and combination forms, the sub-regions corresponding to the minimum errors are divided and combined. After the optimal sub-region division and combination, the method further includes:

Obtain the optimal sub-region division combination of the prediction time scale corresponding to each historical time point in the first preset historical time period;

According to the actual power value of each sub-region in the first preset historical time period and the fitted model function, the power prediction value of each sub-region in the future corresponding to each historical time point is obtained;

According to the power prediction value of each sub-region at the future time, the optimal sub-region division and combination corresponding to the prediction time scale at each historical time point is used as the division and combination method at the current prediction time, so as to obtain the ultra-short-term output of the wind power cluster to be predicted. Predictive value.

In the embodiment of the present invention, since the correlation of wind farms in a wind power cluster is relatively complex, it is generally difficult to directly obtain an optimal cluster division method through external conditions. The embodiment of the present invention adopts a continuous division method, and on the basis of known actual cluster output, calculates the optimal cluster division method at the previous moment, and uses this optimal division as the optimal division at the current moment. Since this method uses the result of the previous moment as the current prediction, which is similar to the continuous method, it is called continuous division. According to this method, the weather system often maintains a certain inertia, and the correlation properties of the wind farms in the cluster corresponding to the two closer time points are relatively close, so that the corresponding optimal divisions are basically the same.

定义为风电集群在t时刻时，超前预测k个时间范围的最优划分。图2为本发明实施例提供的最优划分的时序示意图，可参考图2所示，通过将第一预设历史时间段预测得到的最优子区域划分作为本时刻的子区域划分，从用于后续的风电集群出力预测。Specifically, among many cluster division methods, a group of division methods with the smallest prediction error in the first preset historical time period is used as the current optimal cluster division. In the ultra-short-term cluster forecasting of wind power, there are two scales of optimization for optimal division, namely, sample optimization and prediction time range optimization. Sample optimization means that at each moment, for the samples at each time point, an optimal division calculation should be performed, and the optimal division at two adjacent time points may be different; the prediction time range optimization is It means that in the ultra-short-term 16-step forecast, each step of the forecast corresponds to an optimal division, and the division corresponding to different forecast time ranges at the same time point may be different. Therefore, at one point in time, 16 divisions are to be optimized, and the

It is defined as the optimal division of the k time horizons predicted in advance for the wind power cluster at time t. FIG. 2 is a schematic time sequence diagram of an optimal division provided by an embodiment of the present invention. Referring to FIG. 2 , the optimal sub-area division predicted by the first preset historical time period is used as the sub-area division at this moment. In the follow-up wind power cluster output forecast.

并将该子区域最优划分

作为当前时刻预测t在超前k步预测中所采用的集群划分

需要说明的是，在本发明实施例中，对于不同的预测时间范围k，所谓的最近历史时刻并不相同，因为预测时间范围越长，需要更长时间后的真实集群出力才能判断，因此k越大，对应的历史时刻距离当前时刻越远，但是选取历史时刻子区域最优划分与当前集群划分的超前时刻k相同。图3为本发明实施例提供的超短期预测下持续划分的时序示意图，可参考图3所示，每一行代表不同的预测时间范围。在每一行中，绿色代表已知对应时间范围下最优划分的历史时刻，对于超前k步的预测，其历史时刻截止到t－k处，即持续性方法就是将最近的已知对应时间范围下最优划分作为当前时刻对应预测时间范围的集群划分。Further, because the optimal division method provided by the above embodiment is a concept of a posteriori calculation, that is, only after knowing the real cluster output at the predicted time, the error can be calculated to obtain the optimal division at the corresponding time. In order to meet practical application requirements, the embodiment of the present invention provides a continuous division prediction method.

and divide the subregion optimally

As the cluster division used in the prediction of t at the current moment in the prediction of k steps ahead

It should be noted that, in this embodiment of the present invention, for different prediction time ranges k, the so-called recent historical moments are not the same, because the longer the prediction time range, the longer the actual cluster output is needed to judge, so k The larger the value is, the farther the corresponding historical time is from the current time, but the optimal sub-region division of the historical time is selected to be the same as the advance time k of the current cluster division. FIG. 3 is a schematic time sequence diagram of continuous division under ultra-short-term prediction provided by an embodiment of the present invention. Referring to FIG. 3 , each row represents a different prediction time range. In each row, green represents the historical time of the optimal division in the known corresponding time range. For predictions ahead of k steps, the historical time is up to t-k, that is, the persistence method is to use the latest known corresponding time range. The lower optimal division is used as the cluster division of the prediction time range corresponding to the current moment.

On the basis of the above embodiment, the number of historical time points in the first preset historical time period is not more than 16, and the time interval of each historical time point is 15 minutes.

In the embodiment of the present invention, the length of the advance forecast can be preset according to the actual output forecast demand of the wind power cluster. The sub-area division corresponding to the minimum error between the actual power value and the predicted power value in the historical time period is set as the optimal sub-area division combination for ultra-short-term output prediction of wind power clusters. For example, it is only necessary to predict the output of the wind power cluster one hour in the future at the current moment, and then it is only necessary to perform ultra-short-term output forecasting according to the four historical time points of the most recent historical time at the current moment, so as to obtain the predicted output value of the current moment one hour in the future.

In another embodiment of the embodiment of the present invention, a prediction model is first established based on the historical output of all possible sub-regions in the wind power cluster, and this part can be calculated and obtained at the off-line time; in the online operation of the system, the optimal division algorithm is used to calculate The optimal division of the recent historical moment at the current moment; finally, the prediction method of continuous division is adopted, and the optimal division of the most recent historical moment is used as the division adopted for the prediction of the current moment, and the prediction power of each sub-region is calculated, and then the most The sub-region powers in the optimal division are summed to obtain the predicted value of the cluster power. Specific steps are as follows:

其中，

表示模型训练时期的

时刻；Step S20: Obtain the sample power historical data of each wind farm in the wind power cluster, and calculate the sample power data of the sub-regions c _i under different division forms

in,

represents the model training epoch

time;

时刻的子区域c_i的样本功率数据

滚动向后划分样本数据集，得到

时刻的样本

其中，

表示输入到滑动时间窗的特征数据，

表示滑动时间窗输出的样本功率特征数据；Step S21, the length of the sliding time window is set to L, and the training set is

The sample power data of the sub-region c _i at the moment

Rolling backwards to divide the sample dataset, we get

sample of moments

in,

represents the feature data input to the sliding time window,

Represents the sample power characteristic data output by the sliding time window;

在1到T₀之间所有子区域的样本，其中子区域c_i的所有数据集表示为

Step S22, by traversing the sample power historical data according to time and all possible divisions of sub-regions, to obtain the time

The samples of all subregions between 1 and _T0 , where all datasets of subregion _ci are denoted as

得到的拟合模型函数为

Step S23, input all possible sub-region sample data into the support vector machine model for training, so as to obtain a trained model, that is, obtain a fitting model function for sub-region output prediction; wherein for the sample of sub-region c _i data set

The resulting fitted model function is

输入到拟合模型函数中，并进行初始化，将超前预测范围设置为k’，第一预设历史时间段内的历史时刻点t’＝T-k’，子区域编号i设置为1；In step S24, the previous sliding time window (that is, the second preset historical time period) of each sub-region in the last 16 historical time points of the current time T of the wind power cluster (that is, the historical time points within the first preset historical time period) historical power data of

Input into the fitting model function, and initialize, set the advance prediction range to k', the historical time point t'=T-k' in the first preset historical time period, and the sub-region number i is set to 1;

Step S25, by fitting the model function, and obtaining the power prediction values of 16 time points of all sub-regions in the first preset historical time period through the historical power data of the second preset historical time period

在获取划分组合下的子区域功率预测值之和之后，进行步骤S27；Step S26, obtaining the sum of the predicted power values of the sub-regions in each division of the wind power cluster under the s-seed region division C ^S ,

After obtaining the sum of the sub-region power prediction values under the division and combination, go to step S27;

In step S27, the division with the smallest error, that is, the optimal sub-region division combination, is obtained by the following formula, and the formula is:

Step S28, dividing and combining the optimal sub-regions of each historical time point in the first preset historical time period as the predicted division corresponding to the future k' time at the current time, that is,

计算对应子区域当前时刻T的未来k’时刻的功率预测值，即

Step S29, according to the optimal sub-region division and combination of each historical time point in the first preset historical time period, and the historical power data of the corresponding sub-region in a sliding time window before the current time T

Calculate the power prediction value of the current time T in the corresponding sub-region at the future k' time, that is

Step S30: Calculate the sum of the predicted power values of the wind power cluster at time k' in the future according to the predicted power value at time k' in the future at the current time T of the sub-region, and the formula is:

Step S31, obtain the output forecast of the current time T of the wind power cluster in the next 0-4 hours

In an embodiment of the present invention, by predicting the wind power cluster data in a certain region, the method for predicting the ultra-short-term output of a wind power cluster provided by the embodiment of the present invention is compared and explained with the existing method. Specifically, two time points in spring and summer in a certain region are used, and 1000 points are selected as training samples and 500 points as test samples for each calculation example. The length L of the sliding time window is taken as 7, and the root mean square error of the error is set as the evaluation index according to the existing wind power labeling regulations. Since the ultra-short-term forecast is a rolling forecast of multiple time horizons from 0 to 4h, here are given the kth time horizon forecast error root mean square E _{RMSE,k forward} and the 16 time horizon comprehensive forecast error root mean square E _{RMSE,k all} , the specific definition is as follows:

For the description of the wind power cluster in spring, reference can be made to Table 2. The prediction error of the prediction method provided by the embodiment of the present invention is smaller than that of the existing accumulation method in each time range and comprehensive range, and the prediction accuracy is improved.

Table 2

Existing accumulation method The method provided by the present invention E_{RMSE, 1h}_forward/% 8.98 7.92 E_{RMSE, 2h}_forward/% 11.78 10.17 E_{RMSE, 3h}_forward/% 13.52 11.81 E_{RMSE, 4h}_forward/% 14.73 12.85 E_{RMSE, 4h}_all/% 11.68 10.20

For the description of the summer wind power cluster, reference can be made to Table 3. The prediction errors of the prediction method provided by the embodiment of the present invention in each time range and comprehensive range are smaller than the existing accumulation method, and the prediction accuracy is improved. Since the prediction accuracy is improved in spring and summer, the generalization of the method provided by the embodiment of the present invention is illustrated, and it can be more suitable for multiple event scenarios.

table 3

Existing accumulation method The method provided by the present invention E_{RMSE, 1h}_forward/% 12.56 10.62 E_{RMSE, 2h}_forward/% 13.76 12.48 E_{RMSE, 3h}_forward/% 14.24 12.60 E_{RMSE, 4h}_forward/% 15.03 13.05 E_{RMSE, 4h}_all/% 13.57 11.86

FIG. 4 is a schematic structural diagram of an ultra-short-term output prediction system for wind power clusters provided by an embodiment of the present invention. As shown in FIG. 4 , an embodiment of the present invention provides an ultra-short-term output prediction system for wind power clusters, including wind power Cluster sub-area division module 401, acquisition module 402, power prediction module 403 and ultra-short-term output prediction module 404, wherein, the wind power cluster sub-area division module 401 is used for sub-area division of the wind farms of the wind power cluster to be predicted to obtain the to-be-predicted wind farms. Predicting all sub-area divisions and combinations of the wind power cluster; the acquiring module 402 is configured to acquire the real power value of each sub-area in the first preset historical time period and the power real value in the second preset historical time period for each sub-area in all sub-area division combinations value, the first preset historical time period is the most recent historical time period at the current moment, and the second preset historical time period is the most recent historical time period of the first preset historical time period; the power prediction module 403 uses According to the actual power value of the second preset historical time period, the power of each sub-region of the first preset historical time period is predicted, and various division and combination forms of the first preset historical time period are obtained. The cluster power prediction value under the ultra-short-term output prediction module 404 is used to obtain the error between the real power value of the first preset historical time period and the cluster power prediction value under various division and combination forms, and the minimum error corresponding to The sub-area division and combination are used as the optimal sub-area division and combination, so as to obtain the ultra-short-term output predicted value of the wind power cluster to be predicted according to the optimal sub-area division and combination.

An ultra-short-term output prediction system for a wind power cluster provided by an embodiment of the present invention performs prediction modeling based on sub-regions composed of multiple wind farms, and fully considers the relationship between wind farms and the regional output within the wind power cluster. The characteristic properties of the wind power cluster can be used to predict the ultra-short-term output of the wind power cluster, which improves the accuracy of the output prediction.

On the basis of the above embodiment, the system further includes:

a training data set building module for obtaining a training data set, the training data set including sample power data of each sub-region of the wind power cluster;

The fitting model training module is used to train the support vector machine model through the training data set, and obtain the fitting model function used for output prediction of each sub-region, so as to obtain the fitting model function according to the power of the second preset historical time period. The actual value is used to predict the power of each sub-region in the first preset historical time period by using the fitting model function, and the cluster power prediction value under various division forms of the first preset historical time period is obtained.

The system provided by the embodiments of the present invention is used to execute the above method embodiments. For specific processes and details, please refer to the above embodiments, which will not be repeated here.

FIG. 5 is a schematic structural diagram of an electronic device provided by an embodiment of the present invention. Referring to FIG. 5 , the electronic device may include: a processor (processor) 501, a communication interface (Communications Interface) 502, a memory (memory) 503, and a communication bus 504, wherein , the processor 501 , the communication interface 502 , and the memory 503 communicate with each other through the communication bus 504 . The processor 501 can call the logic instructions in the memory 503 to perform the following method: perform sub-region division on the wind farm of the wind power cluster to be predicted, and obtain all sub-region division combinations of the to-be-predicted wind power cluster; The real power value of each sub-region in the first preset historical time period, and the real power value in the second preset historical time period, the first preset historical time period is the most recent historical time period at the current moment, and the The second preset historical time period is the most recent historical time period of the first preset historical time period; according to the actual power value of the second preset historical time period, each Predict the power of the sub-region to obtain the cluster power prediction value under various division and combination forms of the first preset historical time period; obtain the real power value of the first preset historical time period and various division and combination forms For the error between the cluster power prediction values, the sub-area division and combination corresponding to the smallest error are taken as the optimal sub-area division and combination, so as to obtain the ultra-short-term output predicted value of the wind power cluster to be predicted according to the optimal sub-area division and combination.

In addition, the above-mentioned logic instructions in the memory 503 can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention can be embodied in the form of a software product in essence, or the part that contributes to the prior art or the part of the technical solution. The computer software product is stored in a storage medium, including Several instructions are used to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, Read-Only Memory (ROM, Read-Only Memory), Random Access Memory (RAM, Random Access Memory), magnetic disk or optical disk and other media that can store program codes .

On the other hand, an embodiment of the present invention further provides a non-transitory computer-readable storage medium on which a computer program is stored, and the computer program is implemented when executed by a processor to execute the wind power cluster provided by the above embodiments. The ultra-short-term output forecasting method, for example, includes: dividing the wind farms of the wind power cluster to be predicted into sub-regions to obtain all sub-region division combinations of the to-be-predicted wind power cluster; The real power value of the historical time period, and the real power value of the second preset historical time period, the first preset historical time period is the most recent historical time period at the current moment, and the second preset historical time period is The most recent historical time period of the first preset historical time period; according to the real value of the power of the second preset historical time period, predict the power of each sub-region of the first preset historical time period, and obtain the obtaining the cluster power prediction value under various division and combination forms of the first preset historical time period; obtaining the error between the power real value of the first preset historical time period and the cluster power prediction value under various division and combination forms , taking the sub-area division and combination corresponding to the minimum error as the optimal sub-area division and combination, so as to obtain the ultra-short-term output predicted value of the wind power cluster to be predicted according to the optimal sub-area division and combination.

The device embodiments described above are only illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in One place, or it can be distributed over multiple network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution in this embodiment. Those of ordinary skill in the art can understand and implement it without creative effort.

From the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and certainly can also be implemented by hardware. Based on this understanding, the above-mentioned technical solutions can be embodied in the form of software products in essence or the parts that make contributions to the prior art, and the computer software products can be stored in computer-readable storage media, such as ROM/RAM, magnetic A disc, an optical disc, etc., includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform the methods described in various embodiments or some parts of the embodiments.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, but not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that it can still be The technical solutions described in the foregoing embodiments are modified, or some technical features thereof are equivalently replaced; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims (8)

1. An ultra-short-term output forecasting method for wind power clusters, characterized in that, comprising: Sub-regional division is performed on the wind farm of the wind power cluster to be predicted, and all sub-region division combinations of the to-be-predicted wind power cluster are obtained; Obtain the real power value of each sub-region in the first preset historical time period and the power real value of the second preset historical time period in all sub-region division combinations, and the first preset historical time period is the latest at the current moment. The historical period of time, the second preset historical period is the most recent historical period of the first preset historical period; According to the actual power value of the second preset historical time period, the power of each sub-region of the first preset historical time period is predicted to obtain the different division and combination forms of the first preset historical time period. The predicted value of cluster power; Obtain the error between the actual power value of the first preset historical time period and the predicted value of the cluster power under various division and combination forms, and use the sub-region division and combination corresponding to the minimum error as the optimal sub-region division and combination, so that according to the The optimal sub-regional division and combination are used to obtain the ultra-short-term output forecast value of the wind power cluster to be predicted; After obtaining the error between the actual power value of the first preset historical time period and the predicted value of cluster power under various division and combination forms, the sub-region division and combination corresponding to the minimum error are taken as the optimal sub-region division and combination , the method also includes: Obtain the optimal sub-region division combination of the prediction time scale corresponding to each historical time point in the first preset historical time period; According to the actual power value of each sub-region in the first preset historical time period and the fitted model function, the power prediction value of each sub-region in the future corresponding to each historical time point is obtained; According to the power prediction value of each sub-region at the future time, the optimal sub-region division and combination corresponding to the prediction time scale at each historical time point is used as the division and combination method at the current prediction time, so as to obtain the ultra-short-term output of the wind power cluster to be predicted. Predictive value; The wind farm of the to-be-predicted wind power cluster is divided into sub-regions, and all sub-region division combinations of the to-be-predicted wind power cluster are obtained, including: Based on the Bell number formula, sub-regions are divided for the wind farms of the wind power cluster to be predicted, and all sub-region division combinations are obtained. 2 . The ultra-short-term output forecasting method for wind power clusters according to claim 1 , wherein, in the actual power value according to the second preset history time period, the first preset history The power of each sub-region in the time period is predicted, and before the cluster power prediction value under various division and combination forms of the first preset historical time period is obtained, the method further includes: obtaining a training data set, the training data set including sample power data of each sub-region of the wind power cluster; Through the training data set, the support vector machine model is trained to obtain a fitting model function for output prediction of each sub-region, so as to obtain the fitting model function according to the actual power value of the second preset historical time period, through the fitting The model function predicts the power of each sub-region in the first preset historical time period, and obtains the cluster power prediction value under various division and combination forms of the first preset historical time period. 3. The ultra-short-term output forecasting method for wind power clusters according to claim 2, wherein the acquiring a training data set specifically comprises: Divide the sample power data through the sliding time window, and obtain the sample power output characteristic data of the sample power data based on the sliding time window; According to the sample power data and the sample power output characteristic data corresponding to the sample power data, a training data set is constructed. 4. The ultra-short-term output prediction method for wind power clusters according to claim 1, wherein the number of historical time points in the first preset historical time period is not more than 16, and the number of historical time points in each historical time point is not more than 16. The time interval is 15 minutes. 5. An ultra-short-term output forecasting system for wind power clusters, comprising: The wind power cluster sub-region division module is used for sub-region division of the wind farms of the wind power cluster to be predicted, to obtain all the sub-region division combinations of the to-be-predicted wind power cluster; The acquisition module is used to acquire the real power value of each sub-region in the first preset historical time period and the real power value of the second preset historical time period of each sub-region in all sub-region division combinations. The first preset historical time period The segment is the most recent historical time segment at the current moment, and the second preset historical time segment is the most recent historical time segment of the first preset historical time segment; A power prediction module, configured to predict the power of each sub-region of the first preset historical time period according to the actual power value of the second preset historical time period, and obtain the Cluster power prediction value under various division and combination forms; The ultra-short-term output prediction module is used to obtain the error between the real power value of the first preset historical time period and the cluster power prediction value under various division and combination forms, and the sub-region division and combination corresponding to the minimum error is regarded as the optimal Sub-region division and combination, to obtain the ultra-short-term output predicted value of the wind power cluster to be predicted according to the optimal sub-region division and combination; The system is also used to: Obtain the optimal sub-region division combination of the prediction time scale corresponding to each historical time point in the first preset historical time period; According to the actual power value of each sub-region in the first preset historical time period and the fitted model function, the power prediction value of each sub-region in the future corresponding to each historical time point is obtained; According to the power prediction value of each sub-region at the future time, the optimal sub-region division and combination corresponding to the prediction time scale at each historical time point is used as the division and combination method at the current prediction time, so as to obtain the ultra-short-term output of the wind power cluster to be predicted. Predictive value; The wind power cluster sub-area division module is specifically used for: Based on the Bell number formula, sub-regions are divided for the wind farms of the wind power cluster to be predicted, and all sub-region division combinations are obtained. 6. The ultra-short-term output forecasting system for wind power clusters according to claim 5, wherein the system further comprises: a training data set building module for obtaining a training data set, the training data set including sample power data of each sub-region of the wind power cluster; The fitting model training module is used to train the support vector machine model through the training data set, and obtain the fitting model function used for output prediction of each sub-region, so as to obtain the fitting model function according to the power of the second preset historical time period. true value, predict the power of each sub-region in the first preset historical time period through the fitting model function, and obtain the cluster power predicted value under various division and combination forms of the first preset historical time period . 7. An electronic device comprising a memory, a processor and a computer program stored on the memory and running on the processor, wherein the processor implements any one of claims 1 to 4 when the processor executes the program The steps of the ultra-short-term output forecasting method for wind power clusters described in item 1. 8. A non-transitory computer-readable storage medium on which a computer program is stored, characterized in that, when the computer program is executed by a processor, the supercomputer for a wind power cluster according to any one of claims 1 to 4 is implemented. The steps of the short-term output forecasting method.

Images