CN115034485A - A data space-based wind power interval prediction method and device - Google Patents

Abstract

The invention relates to the technical field of wind power prediction, and specifically provides a data space-based wind power power interval prediction method and device, comprising: acquiring meteorological feature data related to wind power at the forecast time; The meteorological feature data is input into a pre-built deep learning wind power prediction model to obtain a wind power prediction value at the prediction time; the wind power prediction interval at the prediction time is determined based on the wind power prediction value at the prediction time. The invention provides a scientific, reasonable, practical, efficient, accurate and reliable point forecasting and interval forecasting method for short-term wind power of wind farms, which can realize short-term power point forecasting of wind farms for the next 24 hours and interval forecasting under different confidence degrees. The dispatching operation of the power system and the participation of wind farms in the power market provide decision support.

Description

A data space-based wind power interval prediction method and device

technical field

The invention relates to the technical field of wind power prediction, in particular to a data space-based wind power power interval prediction method and device.

Background technique

Because wind power itself is random, fluctuating and intermittent, and has significant anti-peak characteristics, its rapid development has brought severe challenges to the planning, dispatching and safe operation of power systems, as well as the economical and reasonable consumption of wind power. After a large-scale wind farm is connected to the power grid, the intermittent and fluctuating wind power will bring greater pressure on the power balance and frequency stability regulation of the power grid. To cope with the uncertainty of large-scale intermittent wind power integration into the grid, accurate wind power forecasting is increasingly important. The short-term wind power point and interval forecast results are the main reference data for the power grid dispatch management department to formulate power generation plans and adjust operation strategies, and are also an important basis for wind power companies to quote and participate in the electricity spot market. In addition, accurate and reliable short-term wind power point and interval forecasts are also conducive to reducing the technical and economic risks faced by power market participants caused by wind power uncertainty, increasing the benefits of wind power companies participating in power market transactions, and further promoting new Energy development and clean and low-carbon transition of the power system.

With the advent of the 5G era, new-generation information technologies such as big data, cloud services, and artificial intelligence are widely used, which can provide reliable data quality and powerful computing power support for short-term wind power forecasting. Wind farms collect and store massive multi-source, multi-dimensional, Multimodal data resources provide the basis for the construction of wind farm data space. Dataspace adheres to a new data management concept that is different from traditional big data management, shifting from business-oriented to object-oriented. Data space is a low-level framework of distributed multi-label data storage with full object-oriented full life cycle, and a technical system that allows data to be connected securely and efficiently. In line with the trend of digital wind farm transformation and construction, building a wind farm data space is an important means to improve the digital operation and maintenance management level of wind farms. Through the data storage technology and related services provided by DataSpace, wind farms can systematically manage multiple data sources in a comprehensive, safe and efficient manner, and can selectively extract useful information, thereby effectively improving the degree of data utilization, providing wind power Power forecasting and wind farm operation and maintenance management provide decision support.

Based on different modeling ideas, wind power prediction methods can be roughly divided into physical methods, statistical methods, artificial intelligence and machine learning methods, and combinatorial optimization methods. However, the physical method has high requirements on the accuracy and completeness of NWP data, and requires a clear mathematical equation description of atmospheric physical characteristics and fan characteristics, which is generally not suitable for short-term power prediction. Statistical models, artificial intelligence and machine learning models, especially hybrid multi-stage models, are widely used in the field of short-term forecasting of wind power. However, with the construction of information platforms and the improvement of big data technology, the data scale of wind farms has been continuously expanded and the data types have become more diversified. However, traditional statistical and machine learning models are difficult to describe the complex mapping relationship between wind power fluctuations and multi-dimensional meteorological data, which limits the improvement of point prediction accuracy. Moreover, in practical engineering applications, due to the risks brought about by strong wind power fluctuations, wind power stakeholders have put forward new demands for power probability interval prediction.

To sum up, there is still a certain gap in the comprehensive application of multi-dimensional data spatial data cleaning and feature mining technology of wind farms, intelligent decomposition and noise reduction methods of wind power time series data, and point forecasting methods and interval forecasting methods. The lack of application of advanced deep learning algorithms in engineering practice limits the further improvement of wind power prediction accuracy to a certain extent.

SUMMARY OF THE INVENTION

In order to overcome the above defects, the present invention proposes a data space-based wind power interval prediction method and device.

In a first aspect, a data space-based wind power interval prediction method is provided, and the data space-based wind power interval prediction method includes:

Obtain meteorological feature data related to wind power at the forecast time;

Inputting the meteorological feature data related to wind power at the forecast time into a pre-built deep learning wind power forecast model to obtain a wind power forecast value at the forecast time;

The wind power prediction interval at the prediction time is determined based on the wind power prediction value at the prediction time.

Preferably, before obtaining the meteorological feature data related to the wind power at the predicted time, the method includes:

Collect historical power data and meteorological characteristic data of wind farms;

Preprocessing the historical power data and meteorological characteristic data of the wind farm;

Calculate the correlation between the historical power data and various meteorological feature data, and screen the meteorological features related to wind power based on the correlation between the historical power data and various meteorological feature data.

Further, the preprocessing includes at least one of the following: outlier identification, data cleaning, and wind speed data preprocessing.

Further, calculating the correlation between historical power data and various meteorological feature data, and screening meteorological features related to wind power based on the correlation between historical power data and various meteorological feature data includes:

Calculate the Pearson correlation coefficient between various meteorological feature data;

If the Pearson correlation coefficient between various meteorological feature data exceeds the first threshold, then remove one of the meteorological feature data;

Calculate the Pearson correlation coefficient and gray correlation between historical power data and various meteorological feature data;

If the Pearson correlation coefficient between the meteorological feature data and the historical power data is significant at the preset significance level, and the gray correlation between the meteorological feature data and the historical power data exceeds the second threshold, then the corresponding meteorological feature data Meteorological features are meteorological features related to wind power.

Preferably, the training process of the pre-built deep learning wind power prediction model includes:

The training set and the validation set are constructed by using the intrinsic modal components and residual components obtained after decomposing the historical power data of the wind farm and the meteorological characteristic data related to the wind power;

The initial deep learning wind power prediction model is trained by using the training set and the verification set to obtain the pre-built deep learning wind power prediction model.

Further, determining the wind power prediction interval at the prediction time based on the wind power prediction value at the prediction time includes:

Inputting the meteorological feature data related to wind power in the verification set into a pre-built deep learning wind power prediction model, to obtain the wind power prediction value of each sample point in the verification set;

Taking the difference between the wind power prediction value of each sample point in the verification set and the historical wind power power of each sample point in the verification set as the wind power prediction error of each sample point in the verification set;

Determine the relative error sample sequence of the verification set based on the wind power prediction value of each sample point of the verification set and the wind power prediction error of each sample point of the verification set;

The upper and lower confidence limits corresponding to the relative error sample sequence of the validation set obtained based on the kernel density estimation method;

The wind power prediction interval at the prediction time is determined based on the upper and lower confidence limits corresponding to the relative error sample sequence of the verification set.

Further, the calculation formula of the relative error sample sequence of the verification set is as follows:

e _p = _er /p _v

In the above formula, _ep is the relative error sample sequence of the verification set, _er is the wind power prediction error sequence of each sample point in the verification set, and _pv is the wind power prediction value of each sample point in the verification set.

Further, the calculation formula of the wind power prediction interval at the prediction time is as follows:

为预测时刻的风电功率预测区间，

为所述预测时刻的风电功率预测值在置信度水平μ下的上限，

为所述预测时刻的风电功率预测值在置信度水平μ下的下限。In the above formula,

is the forecast interval of wind power at the forecast time,

is the upper limit of the predicted value of wind power at the predicted time under the confidence level μ,

is the lower limit of the predicted value of wind power at the predicted time under the confidence level μ.

Further, the calculation formula of the upper limit of the predicted value of wind power at the predicted time under the confidence level μ is as follows:

The calculation formula of the lower limit of the predicted value of wind power at the predicted time under the confidence level μ is as follows:

为所述验证集的相对误差样本序列对应的置信度水平μ的上限，

为所述验证集的相对误差样本序列对应的置信度水平μ的下限。In the above formula, _pi is the predicted value of wind power at the predicted time,

is the upper limit of the confidence level μ corresponding to the relative error sample sequence of the validation set,

is the lower limit of the confidence level μ corresponding to the relative error sample sequence of the validation set.

In a second aspect, a data space-based wind power interval prediction device is provided, and the data space-based wind power interval prediction device includes:

The acquisition module is used to acquire the meteorological feature data related to the wind power at the forecast time from the wind farm data space;

a first determination module, configured to input the meteorological feature data related to the wind power at the forecast time into a pre-built deep learning wind power forecast model to obtain a wind power forecast value at the forecast time;

The second determination module is configured to determine the wind power prediction interval at the prediction time based on the wind power prediction value at the prediction time.

In a third aspect, a computer device is provided, comprising: one or more processors;

the processor for storing one or more programs;

When the one or more programs are executed by the one or more processors, the data space-based wind power interval prediction method is implemented.

In a fourth aspect, a computer-readable storage medium is provided on which a computer program is stored, and when the computer program is executed, the data space-based wind power interval prediction method is implemented.

The above-mentioned one or more technical solutions of the present invention have at least one or more of the following beneficial effects:

The invention provides a data space-based wind power interval prediction method and device, including: acquiring meteorological feature data related to wind power at the forecast time; inputting the meteorological feature data related to the wind power at the forecast time into a pre-built The wind power prediction model is deeply learned to obtain the wind power prediction value at the prediction time; the wind power prediction interval at the prediction time is determined based on the wind power prediction value at the prediction time. The invention provides a scientific, reasonable, practical and high-efficiency method for accurate point prediction and interval prediction of short-term wind power of wind farms, which can realize short-term power point prediction of wind farms for the next 24 hours and interval predictions under different confidence degrees. It provides decision support for dispatching operation and participation of wind farms in the electricity market.

Description of drawings

FIG. 1 is a schematic flowchart of main steps of a data space-based wind power interval prediction method according to an embodiment of the present invention;

2 is a schematic structural diagram of an LSTM neural network model according to an embodiment of the present invention;

3 is a schematic flowchart of detailed steps of a data space-based wind power interval prediction method according to an embodiment of the present invention;

4 is a result diagram of a training set wind power data sequence after variational modal decomposition according to an embodiment of the present invention;

FIG. 5 is a prediction curve and a relative error diagram of the wind power point prediction method based on BiLSTM-Attention for predicting the set wind power according to an embodiment of the present invention;

6 is a forecast interval diagram of the forecast set wind power under different confidence levels calculated by the data space-based wind power interval prediction method based on kernel density estimation according to an embodiment of the present invention;

FIG. 7 is a main structural block diagram of a data space-based wind power interval prediction apparatus according to an embodiment of the present invention.

Detailed ways

The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

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 view of the shortcomings of the existing wind power prediction technology, the purpose of the present invention is to combine the data storage management technology of the data space and the powerful nonlinear data fitting ability of the deep learning algorithm to provide a scientific, reasonable, practical, efficient, accurate and reliable wind power The short-term wind power interval prediction method of the wind farm can realize the short-term power point forecast of the wind farm for the next 24 hours (the time resolution is 15 minutes or less), and the interval forecast under different confidence levels. The electricity market provides decision support.

Due to redundant feature dimensions, the prediction model will be too complex, which limits the improvement of model prediction efficiency and prediction accuracy. Based on the screening of key features by Pearson correlation analysis and grey correlation analysis, the causal relationship between input features and wind power can be preliminarily mined from the multi-source and multi-dimensional wind power influencing factor data. By screening out the key features that affect wind power, the effective information resources in the data can be fully mined, and prior knowledge can be added to model training, thereby improving model prediction accuracy and goodness of fit.

Due to the strong randomness and volatility of the original wind power sequence, it will adversely affect the accuracy and stability of the prediction model. When using variational modal decomposition to decompose the wind power sequence signal, the total number of modal decomposition and the value of the quadratic penalty coefficient will directly affect the final effect of the original signal decomposition. The variational modal decomposition method optimized by the gray wolf algorithm can realize the optimal values of the total number of modal decomposition and the quadratic penalty coefficient, and then decompose the wind power signal with strong randomness and volatility into stationary signals of different frequencies. components, thereby improving the stability and accuracy of the model's predictions.

Since the accuracy of the current wind power point prediction method still needs to be further improved, the present invention constructs a deep learning model based on BiLSTM-Attention to achieve short-term wind power point prediction. With the help of the sensitivity of the long short-term memory (LSTM) model to time series data, the bidirectional long short-term memory (BiLSTM) network can take into account the information of the time series data before and after, and can Capturing information that is ignored by unidirectional LSTMs, more fully mines the underlying information in the data sequence, and enhances the network's ability to handle complex situations. The BiLSTM model is used to predict each component of wind power, strengthen the time series features of the data, and dig deeper into the causal relationship between input features and wind power, which can further improve the prediction accuracy of the model. The Attention mechanism can use limited computing resources to process more important information. The invention introduces the soft attention mechanism into BiLSTM, improves the defect that the BiLSTM network loses important information due to too long time series data by assigning weights by probability, and highlights the influence of important information in key features, thereby improving the prediction accuracy and prediction of the model. efficiency.

Most of the traditional wind power forecasting research focuses on improving the point forecasting accuracy, while ignoring the additional information in the forecasting error. By describing the upper and lower boundaries of the wind power prediction curve, interval prediction can provide more information for power system decision makers in practical engineering applications, which can help improve grid stability and reduce operating costs. In order to improve the decision-making reference value of wind power prediction results, it is necessary to extract valuable information from massive multi-source high-dimensional data, reduce the non-stationarity of wind power sequences, improve the quality of model training sets, and then improve the accuracy of wind power point forecasting and interval forecasting. precision.

Example 1

Referring to FIG. 1 , FIG. 1 is a schematic flowchart of main steps of a data space-based wind power interval prediction method according to an embodiment of the present invention. As shown in FIG. 1 , the data space-based wind power interval prediction method in the embodiment of the present invention mainly includes the following steps:

Step S101: obtaining meteorological feature data related to wind power at the predicted time;

Step S102: Input the meteorological feature data related to the wind power at the predicted time into a pre-built deep learning wind power prediction model to obtain a predicted value of the wind power at the predicted time;

Step S103: Determine a wind power prediction interval at the prediction time based on the wind power prediction value at the prediction time.

In this embodiment, in the step S101, information such as historical power data and meteorological characteristics (including wind speed, wind direction, temperature, humidity, air pressure, etc.) of the wind farm can be collected through information systems such as SCADA and NWP, and information such as wind farm data can be obtained through the NWP system for the next day. Meteorological characteristic data values, and then the data collected from different platforms (the data time resolution is consistent) are aggregated, integrated and stored in the wind farm data space according to unified data standards, specifications and protocols for unified management, which is used for wind power forecasting. Provide safe, stable and efficient data support services.

In this embodiment, before the step S101, it includes:

Collect historical power data and meteorological characteristic data of wind farms;

Preprocessing the historical power data and meteorological characteristic data of the wind farm;

Calculate the correlation between the historical power data and various meteorological feature data, and screen the meteorological features related to wind power based on the correlation between the historical power data and various meteorological feature data.

Wherein, the preprocessing includes at least one of the following: outlier identification, data cleaning, and wind speed data preprocessing.

In one embodiment, in order to avoid data outliers and missing values in the historical data set from affecting the accuracy of the data-driven model, the data set of wind power and its influencing factors (features) in the wind farm data space needs to be preprocessed. of:

Step 1: Outlier identification. There are abnormal values in the original wind farm data set, which can be mainly divided into the following categories: the wind speed is greater than the cut-in wind speed, but the power value is 0; the wind power is greater than 0, but the wind speed record is 0; data sensor sampling failure, abnormal data collection or Missing data records, etc. For the missing of the entire data record, it is removed from the sample, and further processing is performed for missing values and outliers.

Step 2: Data cleaning. Considering that the changes of meteorological and power data have a certain continuity, the data mean of the adjacent sampling time intervals before and after the missing data value is used as the supplementary value of the missing value or as a substitute value for the abnormal value.

Step 3: Preprocessing of wind speed data. The wind direction ranges from 0° to 360°. Physically speaking, 0°, 360° and 1° and 359° are basically equivalent to the fan output, but for data-driven deep learning methods, the input The numerical difference is very large. Therefore, the trigonometric function is used for the wind direction, that is, the sin value of the wind direction is used to replace the original wind direction data.

Step 4: After outlier identification, data cleaning and wind speed data preprocessing, data samples are obtained and divided into training set, validation set and validation set.

Further, calculating the correlation between historical power data and various meteorological feature data, and screening meteorological features related to wind power based on the correlation between historical power data and various meteorological feature data includes:

Calculate the Pearson correlation coefficient between various meteorological feature data;

If the Pearson correlation coefficient between various meteorological feature data exceeds the first threshold, then remove one of the meteorological feature data;

Calculate the Pearson correlation coefficient and gray correlation between historical power data and various meteorological feature data;

If the Pearson correlation coefficient between the meteorological feature data and the historical power data is significant at the preset significance level, and the gray correlation between the meteorological feature data and the historical power data exceeds the second threshold, then the corresponding meteorological feature data Meteorological features are meteorological features related to wind power.

In one embodiment, the Pearson correlation coefficient and the grey correlation degree are used to analyze the influencing factors of wind power, to screen key features and reduce data redundancy; in the embodiment provided by the present invention, the first threshold is set to 0.95, and the preset The significance level was 0.01 and the second threshold was set at 0.80.

In this embodiment, the training process of the pre-built deep learning wind power prediction model includes:

The training set and the validation set are constructed by using the intrinsic modal components and residual components obtained after decomposing the historical power data of the wind farm and the meteorological characteristic data related to the wind power;

The initial deep learning wind power prediction model is trained by using the training set and the verification set to obtain the pre-built deep learning wind power prediction model.

Specifically, the specific process of decomposing the wind power sequence uses the Variational Modal Decomposition (VMD) method optimized by the Grey Wolf Optimization (GWO) algorithm to decompose the historical wind power sequence to obtain multiple different frequencies. The intrinsic mode components (Intrinsic ModeFunctions, IMFs) and a residual component (Residual, RES).

Using the gray wolf optimization algorithm, the local envelope entropy minimum value of each modal component is used as the fitness function to optimize the total number of modal decomposition K and the quadratic penalty coefficient α ₀ of the wind power sequence. The formula for calculating the envelope entropy of the modal components x(j), j=1,2,...,n is as follows.

In the formula, a(j) is the envelope signal obtained by Hilbert demodulation of the α ₀ modal component x(j) obtained after variational modal decomposition, and p _j is the normalized form of a(j).

The total number of optimal modal decompositions K and the quadratic penalty coefficient α ₀ obtained by the gray wolf optimization algorithm are used as the parameters of the variational modal decomposition, and finally the multiple intrinsic modal components (IMFs) of the decomposed wind power sequence are calculated. and a residual component (RES).

In one embodiment, during the training process of the pre-built deep learning wind power prediction model, the wind farm historical data set is divided into a training set and a verification set, and the training set is used to train a BiLSTM-Attention (attention mechanism optimized model) Bidirectional long short-term memory neural network) deep learning prediction model, using the validation set to determine the hyperparameters of the deep learning model, and recording the prediction error of the validation set. In the prediction model, the min-max (maximum and minimum value) normalization method is used to normalize all input data, and the model prediction results are de-normalized and output. The key feature data of the wind farm to be predicted in the next day is input into the BiLSTM-Attention model trained by the historical data set, and the point prediction result of wind power is obtained.

Specifically, an LSTM neural network model is established, as shown in FIG. 2 .

i _t =σ(W _i x _t +U _i h _t-1 +b _i )

f _t =σ(W _f x _t +U _f h _t-1 +b _f )

o _t =σ(W _o x _t +U _o h _t-1 +b _o )

h _t =o _t ⊙tanh(c _t )

In the formula, σ represents the sigmoid activation function, W _i , W _f , and W _o respectively refer to the weight vector from the input layer to the input gate, forget gate, output gate and cell state; U _i , U _f , and U _o respectively refer to The weight vector from the hidden layer to the input gate, forgetting gate, output gate and cell state; b _i , b _f , respectively refer to the bias of the input gate, forgetting gate, output gate and cell state.

On the basis of the above, the BiLSTM neural network model is established. The specific calculation formula of the total output value h' of the BiLSTM computing unit at time t is as follows:

和

分别表示前向LSTM单元和后向LSTM单元的输出值，

为向量拼接操作。In the formula,

and

represent the output values of the forward LSTM unit and the backward LSTM unit, respectively,

It is a vector concatenation operation.

The soft attention mechanism is introduced into BiLSTM to improve the BiLSTM network's defect of losing important information due to too long time series data by replacing the random weight assignment with probability assignment weight, thereby improving the prediction accuracy and prediction efficiency of the BiLSTM-Attention model.

Further, input the screened key feature sequence and decomposed wind power sequence data into the deep learning prediction model based on BiLSTM-Attention; use the training set to train the prediction model, and determine the hyperparameters of the model based on the validation set, and Record the error of the verification set; then input the key feature data of the prediction set into the BiLSTM-Attention model trained on the historical data set, obtain the predicted value of each component of the power sequence, and add up the components to obtain the point prediction result of wind power power .

In one embodiment, the present invention also performs wind power point prediction result evaluation, specifically:

Select Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE) and R2 (R- ^squared ) as the model point prediction results. evaluation indicators.

Further, the smaller the mean absolute error, the mean absolute percentage error and the root mean square error, the higher the model prediction accuracy and the better the point prediction effect; the closer R ² is to 1, the better the fitting degree between the predicted power curve and the actual power curve. The better, the better the point prediction.

In this embodiment, the kernel density estimation method is used to calculate the upper and lower confidence limits of the relative error sample sequence of the validation set under different confidence levels, and the upper and lower confidence limits of the predicted value of wind power under different confidence levels are calculated in combination with the point prediction results at the time to be predicted. The lower limit of the wind power is obtained, and the wind power prediction interval is obtained, and the wind power prediction interval at the prediction moment is determined based on the wind power prediction value at the prediction moment, including:

Inputting the meteorological feature data related to wind power in the verification set into a pre-built deep learning wind power prediction model, to obtain the wind power prediction value of each sample point in the verification set;

Taking the difference between the wind power prediction value of each sample point in the verification set and the historical wind power power of each sample point in the verification set as the wind power prediction error of each sample point in the verification set;

Determine the relative error sample sequence of the verification set based on the wind power prediction value of each sample point of the verification set and the wind power prediction error of each sample point of the verification set;

The upper and lower confidence limits corresponding to the relative error sample sequence of the validation set obtained based on the kernel density estimation method;

The wind power prediction interval at the prediction time is determined based on the upper and lower confidence limits corresponding to the relative error sample sequence of the verification set.

In one embodiment, the calculation formula of the relative error sample sequence of the validation set is as follows:

e _p = _er /p _v

In the above formula, _ep is the relative error sample sequence of the verification set, _er is the wind power prediction error sequence of each sample point in the verification set, and _pv is the wind power prediction value of each sample point in the verification set.

In one embodiment, the calculation formula of the wind power prediction interval at the prediction time is as follows:

为预测时刻的风电功率预测区间，

为所述预测时刻的风电功率预测值在置信度水平μ下的上限，

为所述预测时刻的风电功率预测值在置信度水平μ下的下限。In the above formula,

is the forecast interval of wind power at the forecast time,

is the upper limit of the predicted value of wind power at the predicted time under the confidence level μ,

is the lower limit of the predicted value of wind power at the predicted time under the confidence level μ.

In one embodiment, the calculation formula of the upper limit of the predicted value of wind power at the predicted moment under the confidence level μ is as follows:

The calculation formula of the lower limit of the predicted value of wind power at the predicted time under the confidence level μ is as follows:

为所述验证集的相对误差样本序列对应的置信度水平μ的上限，

为所述验证集的相对误差样本序列对应的置信度水平μ的下限。In the above formula, _pi is the predicted value of wind power at the predicted time,

is the upper limit of the confidence level μ corresponding to the relative error sample sequence of the validation set,

is the lower limit of the confidence level μ corresponding to the relative error sample sequence of the validation set.

In one embodiment, for the data space-based wind power interval prediction effect evaluation, the prediction interval coverage ratio (Prediction Interval Coverage Probability, PICP), the prediction interval normalized average width (PINAW) and the comprehensive reliability index are selected. (Coverage width-based criterion, CWC) for comparative analysis.

The forecast interval coverage index reflects the probability that the actual wind power falls within the forecast interval under the specified confidence level. If PICP is less than the confidence level μ, the forecast is invalid; otherwise, the forecast is valid. The larger the PICP, the greater the probability that the actual power falls between the upper and lower limits of the prediction, and its expression is as follows.

In the formula, N is the number of prediction samples, and c _i is a Boolean variable whose values are as follows.

The average bandwidth index of the prediction interval can reflect the average value of the width between the upper and lower limits of the prediction interval. When the PICP of the prediction result is the same, a smaller PINAW corresponds to a better prediction effect. Specifically, it can be expressed as:

In the formula, Z is the change interval of the wind power value.

Since PICP and PINAW only reflect unilateral evaluation criteria, they cannot reflect the comprehensive performance of the predicted results. Therefore, a comprehensive reliability evaluation index (Coverage width-based criterion, CWC) is introduced, and its calculation formula is as follows:

CWC=PINAW[1+γ _PIPC e ^-η(PIPC-μ) ], η>0

In the formula, η is a parameter greater than 0, which is taken as 1 in the present invention; μ is a given confidence level.

Further, the present invention provides a specific embodiment, as shown in Figure 3, which specifically includes the following implementation steps:

Step 1: Wind farm data collection and storage. Taking a wind farm with an installed capacity of 150MW in northern China as an example, the wind power data and meteorological feature data in the wind farm SCADA platform are stored in the wind farm data space. The wind farm data set includes wind power and meteorological features such as wind speed at 10m, wind direction at 10m, wind speed at 30m, wind direction at 30m, wind speed at 50m, wind direction at 50m, wind speed at 70m, wind direction at the hub, wind speed at the hub, pressure, temperature, humidity, etc. The data sampling time interval for 5min. The measured historical data and power data of the wind farm from May 15, 2013 to May 31, 2018 of the wind farm are selected to train and test the method proposed by the present invention.

Step 2: Data preprocessing. Perform outlier identification, data cleaning, and wind speed data preprocessing on the collected data set.

Step 2.1: For the absence of the entire data record, remove it from the sample.

Step 2.2: For individual missing data values, the data mean value at the adjacent moments before and after the missing data value is used as the supplementary value for the missing value; for individual abnormal data values, the data mean value at the adjacent moments before and after the abnormal data value is used as the anomaly. Alternate value for the value.

Step 2.3: Use trigonometric function processing on the wind direction, that is, replace the original wind direction data with the sin value of the wind direction.

Step 2.4: After data preprocessing, a total of 4420 data samples were obtained, as shown in Table 1 below. Use the first 3844 data as training samples to train the prediction model; then use the next 288 points as the validation set to debug the model hyperparameters and record the prediction error; use the last 288 points as validation The set is used to calculate the evaluation index of the prediction method, and to verify the effect of the point prediction and interval prediction methods proposed by the present invention.

Table 1

Step 3: Screening of key features of wind power. Pearson correlation analysis and grey correlation analysis are used to analyze the influencing factors of wind power, screen key features, and reduce data redundancy. Based on the calculation results of Pearson correlation coefficient and grey correlation degree, 10m wind speed, 10m wind direction, 30m wind speed, 30m wind direction, 50m wind speed, 50m wind direction, wind speed at the hub, wind direction at the hub, temperature and humidity are selected as the input features of the prediction model. .

Step 3.1: Calculate the Pearson correlation coefficient between different features and the Pearson correlation coefficient between different features, the preset coefficient threshold is 0.95; and the Pearson correlation coefficient between different features and wind power, the preset significance level is 0.01. After calculation, the correlation between the wind speed at 70m and the wind speed at the hub is 0.993, and the correlation between the wind direction at 70m and the wind direction at the hub is 0.998, both of which are greater than the preset 0.95 threshold. Therefore, the wind speed and wind direction at 70m are deleted; and the correlation between the other features All are lower than 0.95. Moreover, after deleting the wind speed and wind direction at 70m, the Pearson correlation coefficients between all remaining features and wind power are significant at the 0.01 level of significance.

Step 3.2: Calculate the grey correlation degree between different features and wind power. The preset correlation threshold is 0.80. The results are shown in Table 2 below. The gray correlation between air pressure and wind power is lower than the preset 0.80 threshold, so it is removed.

Table 2

Step 4: Decomposition of wind power sequence. First, the original wind power sequence is input into the variational modal decomposition model optimized by the gray wolf algorithm. The maximum number of iterations of the gray wolf optimization algorithm is set to 30, the initial population number is set to 20, and the parameters to be optimized include the total number of modal decompositions and the quadratic penalty coefficient, so the variable dimension is set to 2 dimensions; The value range of the total number of state decompositions is set to [3, 10], and the value range of the quadratic penalty coefficient is set to [200, 2000]. The results of the variational modal decomposition parameters optimized by the gray wolf optimization algorithm are as follows: the total number of modal decompositions is 6, and the quadratic penalty coefficient is 178.02. Figure 4 shows the result of the variational modal decomposition of the wind power data sequence in the training set.

Step 5: Short-term wind power point prediction. The structure of the LSTM neural network model is shown in Figure 2. Using the data of the wind farm training set and validation set, the hyperparameters of the BiLSTM-Attention prediction model are set experimentally. Among them, the BiLSTM network contains 2 hidden layers, the number of hidden units is 50 and 50, BatchSize=64, Timestep=36, the learning rate is 0.001, Optimiser='adam', the number of iterations is Epochs=200; the Attention size is 64 . In order to prove the effectiveness of the technical solution proposed by the present invention, the BiLSTM model, the LSTM model and the BP neural network (Back Propagation Neural Network, BPNN) model are selected to predict the wind power of the forecast set respectively, and the forecast results are compared with those based on the present invention. The wind power prediction results of the BiLSTM-Attention model are compared and analyzed. Among them, the parameter settings of LSTM are the same as those of BiLSTM network. BiLSTM-Attention, BiLSTM, LSTM models are implemented using Python 3.8.8 and Tensorflow 2.4.1. The number of neurons in the input layer of BPNN is the number of influencing factors 10, the number of neurons in the hidden layer is 20, the number of neurons in the output layer is 1, the number of training times is 5000, and the accuracy target is set to 0.001. According to the above parameter settings, different models are trained respectively, and finally, the key feature data of the prediction set is input into the trained prediction model, and the prediction results of wind power points of different models are obtained.

The prediction curve and relative error of the wind power point prediction method based on BiLSTM-Attention proposed by the present invention are shown in FIG.

Step 6: Evaluation of wind power point prediction results. Calculate the mean absolute error (MAE), mean absolute percentage error (MAPE), root mean square error (RMSE) and R-squared of the prediction results of different models according to the actual power value of the prediction set and the predicted value of different models. The results are as follows: 3 shown. According to the calculation result of the error evaluation index, it can be seen that the short-term wind power point prediction method based on the BiLSTM-Attention model proposed by the present invention has a smaller prediction error, a higher goodness of fit, and a better point prediction effect.

table 3

predictive model MAE(MW) MAPE MSE(MW) RMSE(MW) R² BiLSTM-Attention 0.4788 3.06% 0.5665 0.7527 0.9995 BiLSTM 0.9556 4.32% 2.1214 1.4565 0.9983 LSTM 1.2017 5.23% 5.6712 2.3814 0.9955 BPNN 2.2655 12.61% 12.2017 3.4931 0.9902

和置信下限

最后，根据预测模型输出的预测集的风电功率值p_i，计算得到该点预测功率p_i不同置信水平μ下的上限

和下限

进而得到预测曲线的上下区间。Step 7: Wind power interval prediction based on data space. Using the actual value of wind power in the verification set and the predicted value of the wind power in the verification set calculated based on the BiLSTM-Attention model, the relative error sample sequence of the verification set can be calculated e _p = _er /p _v =[e _p1 ,e _p2 ,..., e _pn ]. Then, using the kernel density estimation method, the upper confidence limit of the relative error sample sequence of the validation set under different confidence levels is calculated based on the Gaussian kernel function

and lower confidence limit

Finally, according to the wind power value p _i of the forecast set output by the forecast model, the upper limit of the predicted power p _i at this point under different confidence levels μ is calculated.

and lower bound

Then, the upper and lower intervals of the prediction curve are obtained.

The data space-based wind power interval prediction method based on the kernel density estimation proposed by the present invention is shown in FIG.

Step 8: Evaluation of the forecast effect of wind power in the interval based on kernel density estimation. In order to verify the validity of the wind power interval prediction method based on kernel density estimation proposed in the present invention, a parameter estimation method assuming that the error distribution obeys a normal distribution is selected to calculate the prediction interval of wind power under different confidence levels of the forecast set. According to the results of different prediction interval estimation methods, the interval prediction effect evaluation indicators are calculated respectively, and the results are shown in Table 4 below. It can be seen from Table 4 that under different confidence levels, the kernel density estimation method based on the Gaussian kernel function can obtain the interval coverage ratio PICP greater than the preset confidence level, and it is relatively stable. Under the same confidence level, the comprehensive reliability index CWC of the kernel density estimation method based on Gaussian kernel function is better than the interval estimation method based on normal distribution. This means that the wind power interval prediction method based on kernel density estimation proposed in the present invention can closely follow the change trend of the wind power sequence, and at the same confidence level, it can cover more actual wind power with a narrower prediction interval width. value.

Table 4

Example 2

Based on the same inventive concept, the present invention also provides a data space-based wind power interval prediction device, as shown in FIG. 7 , the data space-based wind power interval prediction device includes:

The acquisition module is used to acquire the meteorological feature data related to the wind power at the forecast time from the wind farm data space;

a first determination module, configured to input the meteorological feature data related to the wind power at the forecast time into a pre-built deep learning wind power forecast model to obtain a wind power forecast value at the forecast time;

The second determination module is configured to determine the wind power prediction interval at the prediction time based on the wind power prediction value at the prediction time.

Preferably, before obtaining the meteorological feature data related to the wind power at the predicted time, the method includes:

Collect historical power data and meteorological characteristic data of wind farms;

Preprocessing the historical power data and meteorological characteristic data of the wind farm;

Calculate the correlation between the historical power data and various meteorological feature data, and screen the meteorological features related to wind power based on the correlation between the historical power data and various meteorological feature data.

Further, the preprocessing includes at least one of the following: outlier identification, data cleaning, and wind speed data preprocessing.

Further, calculating the correlation between historical power data and various meteorological feature data, and screening meteorological features related to wind power based on the correlation between historical power data and various meteorological feature data includes:

Calculate the Pearson correlation coefficient between various meteorological feature data;

If the Pearson correlation coefficient between various meteorological feature data exceeds the first threshold, then remove one of the meteorological feature data;

Calculate the Pearson correlation coefficient and gray correlation between historical power data and various meteorological feature data;

If the Pearson correlation coefficient between the meteorological feature data and the historical power data is significant at the preset significance level, and the gray correlation between the meteorological feature data and the historical power data exceeds the second threshold, then the corresponding meteorological feature data Meteorological features are meteorological features related to wind power.

Preferably, the training process of the pre-built deep learning wind power prediction model includes:

The training set and the validation set are constructed by using the intrinsic modal components and residual components obtained after decomposing the historical power data of the wind farm and the meteorological characteristic data related to the wind power;

The initial deep learning wind power prediction model is trained by using the training set and the verification set to obtain the pre-built deep learning wind power prediction model.

Further, determining the wind power prediction interval at the prediction time based on the wind power prediction value at the prediction time includes:

Inputting the meteorological feature data related to wind power in the verification set into a pre-built deep learning wind power prediction model, to obtain the wind power prediction value of each sample point in the verification set;

Taking the difference between the wind power prediction value of each sample point in the verification set and the historical wind power power of each sample point in the verification set as the wind power prediction error of each sample point in the verification set;

Determine the relative error sample sequence of the verification set based on the wind power prediction value of each sample point of the verification set and the wind power prediction error of each sample point of the verification set;

The upper and lower confidence limits corresponding to the relative error sample sequence of the validation set obtained based on the kernel density estimation method;

The wind power prediction interval at the prediction time is determined based on the upper and lower confidence limits corresponding to the relative error sample sequence of the verification set.

Further, the calculation formula of the relative error sample sequence of the verification set is as follows:

e _p = _er /p _v

In the above formula, _ep is the relative error sample sequence of the verification set, _er is the wind power prediction error sequence of each sample point in the verification set, and _pv is the wind power prediction value of each sample point in the verification set.

Further, the calculation formula of the wind power prediction interval at the prediction time is as follows:

为预测时刻的风电功率预测区间，

为所述预测时刻的风电功率预测值在置信度水平μ下的上限，

为所述预测时刻的风电功率预测值在置信度水平μ下的下限。In the above formula,

is the forecast interval of wind power at the forecast time,

is the upper limit of the predicted value of wind power at the predicted time under the confidence level μ,

is the lower limit of the predicted value of wind power at the predicted time under the confidence level μ.

Further, the calculation formula of the upper limit of the predicted value of wind power at the predicted time under the confidence level μ is as follows:

The calculation formula of the lower limit of the predicted value of wind power at the predicted time under the confidence level μ is as follows:

为所述验证集的相对误差样本序列对应的置信度水平μ的上限，

为所述验证集的相对误差样本序列对应的置信度水平μ的下限。In the above formula, _pi is the predicted value of wind power at the predicted time,

is the upper limit of the confidence level μ corresponding to the relative error sample sequence of the validation set,

is the lower limit of the confidence level μ corresponding to the relative error sample sequence of the validation set.

Example 3

Based on the same inventive concept, the present invention also provides a computer device, the computer device includes a processor and a memory, the memory is used for storing a computer program, the computer program includes program instructions, and the processor is used for executing the Program instructions stored by the computer storage medium. The processor may be a central processing unit (Central Processing Unit, CPU), or other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), application specific integrated circuits (Application Specific Integrated Circuit, ASIC), off-the-shelf programmable gate array ( Field-Programmable GateArray, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., which are the computing core and control core of the terminal, which are suitable for implementing one or more instructions, specifically suitable for loading And execute one or more instructions in the computer storage medium to realize the corresponding method flow or corresponding function, so as to realize the steps of a data space-based wind power interval prediction method in the above embodiment.

Example 4

Based on the same inventive concept, the present invention also provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device for storing programs and data. It can be understood that, the computer-readable storage medium here may include both a built-in storage medium in a computer device, and certainly also an extended storage medium supported by the computer device. The computer-readable storage medium provides storage space in which the operating system of the terminal is stored. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space, and these instructions may be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory, or a non-volatile memory (non-volatile memory), such as at least one disk memory. One or more instructions stored in the computer-readable storage medium can be loaded and executed by the processor to implement the steps of the data space-based wind power interval prediction method in the above embodiment.

As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block in the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to the processor of a general purpose computer, special purpose computer, embedded processor or other programmable data processing device to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing device produce Means for implementing the functions specified in a flow or flow of a flowchart and/or a block or blocks of a block diagram.

These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory result in an article of manufacture comprising instruction means, the instructions The apparatus implements the functions specified in the flow or flow of the flowcharts and/or the block or blocks of the block diagrams.

These computer program instructions can also be loaded on a computer or other programmable data processing device to cause a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process such that The instructions provide steps for implementing the functions specified in the flow or blocks of the flowcharts and/or the block or blocks of the block diagrams.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: the present invention can still be Modifications or equivalent replacements are made to the specific embodiments of the present invention, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention shall be included within the protection scope of the claims of the present invention.

Claims (12)

1. A wind power interval prediction method based on a data space is characterized by comprising the following steps:

acquiring meteorological feature data related to wind power at a prediction moment;

inputting the meteorological feature data related to the wind power at the prediction moment into a pre-constructed deep learning wind power prediction model to obtain a wind power prediction value at the prediction moment;

and determining the wind power prediction interval at the prediction moment based on the wind power prediction value at the prediction moment.

2. The method of claim 1, wherein said obtaining meteorological feature data relating wind power at a predicted time comprises, prior to:

acquiring historical power data and meteorological characteristic data of a wind power plant;

preprocessing the historical power data and the meteorological feature data of the wind power plant;

and calculating the correlation between the historical power data and various meteorological feature data, and screening the meteorological features related to the wind power based on the correlation between the historical power data and the various meteorological feature data.

3. The method of claim 2, wherein the pre-processing comprises at least one of: abnormal value identification, data cleaning and wind speed data preprocessing.

4. The method of claim 2, wherein calculating correlations between the historical power data and various meteorological signature data, and screening meteorological signatures related to wind power based on the correlations between the historical power data and the various meteorological signature data comprises:

calculating a Pearson correlation coefficient among various meteorological characteristic data;

if the Pearson correlation coefficient among various meteorological characteristic data exceeds a first threshold value, one meteorological characteristic data is removed;

calculating a Pearson correlation coefficient and a gray correlation degree between historical power data and various meteorological characteristic data;

and if the Pearson correlation coefficient between the meteorological feature data and the historical power data is significant under the preset significance level, and the gray correlation degree between the meteorological feature data and the historical power data exceeds a second threshold value, the meteorological feature corresponding to the meteorological feature data is the meteorological feature related to the wind power.

5. The method of claim 1, wherein the training process of the pre-built deep learning wind power prediction model comprises:

constructing a training set and a verification set by using an inherent modal component and a residual component obtained after the historical power data of the wind power plant are decomposed and meteorological feature data related to the wind power;

and training an initial deep learning wind power prediction model by utilizing the training set and the verification set to obtain the pre-constructed deep learning wind power prediction model.

6. The method of claim 5, wherein the determining the wind power prediction interval for the prediction time based on the wind power prediction value for the prediction time comprises:

inputting meteorological feature data related to the wind power in the verification set into a pre-constructed deep learning wind power prediction model to obtain a wind power prediction value of each sample point of the verification set;

taking the difference between the wind power predicted value of each sample point in the verification set and the historical wind power of each sample point in the verification set as a wind power prediction error of each sample point in the verification set;

determining a relative error sample sequence of the verification set based on the wind power prediction value of each sample point of the verification set and the wind power prediction error of each sample point of the verification set;

obtaining a confidence upper limit and a confidence lower limit corresponding to the relative error sample sequence of the verification set based on a nuclear density estimation method;

and determining the wind power prediction interval at the prediction moment based on the confidence upper limit and the confidence lower limit corresponding to the relative error sample sequence of the verification set.

7. The method of claim 6, wherein the validation set is computed as a sequence of relative error samples as follows:

e _p ＝e _r p _v

in the above formula, e _p A sequence of relative error samples for the validation set, e _r Predicting an error sequence, p, for the wind power at each sample point of the validation set _v And the wind power predicted value of each sample point of the verification set is obtained.

8. The method of claim 6, wherein the wind power prediction interval at the prediction time is calculated as follows:

in the above formula, the first and second carbon atoms are,

in order to predict the wind power prediction interval at the moment,

is the upper limit of the wind power predicted value at the predicted moment under the confidence level mu,

and the lower limit of the wind power predicted value at the predicted moment under the confidence level mu is set.

9. The method according to claim 8, characterized in that the upper limit of the wind power prediction value at the prediction moment at the confidence level μ is calculated as follows:

the lower limit of the wind power predicted value at the predicted time under the confidence level mu is calculated according to the following formula:

in the above formula, p _i The wind power predicted value at the predicted time is the wind power predicted value,

an upper limit for the confidence level mu corresponding to the sequence of relative error samples of the validation set,

the lower limit of the confidence level mu corresponding to the relative error sample sequence of the validation set.

10. A wind power interval prediction device based on data space is characterized by comprising:

the acquisition module is used for acquiring meteorological characteristic data related to wind power at a prediction moment from a wind power plant data space;

the first determination module is used for inputting the meteorological feature data related to the wind power at the prediction moment into a pre-constructed deep learning wind power prediction model to obtain a wind power prediction value at the prediction moment;

and the second determination module is used for determining the wind power prediction interval at the prediction moment based on the wind power prediction value at the prediction moment.

11. A computer device, comprising: one or more processors;

the processor to store one or more programs;

the one or more programs, when executed by the one or more processors, implement the data space-based wind power interval prediction method of any of claims 1 to 9.

12. A computer-readable storage medium, on which a computer program is stored which, when executed, implements a data space-based wind power interval prediction method according to any one of claims 1 to 9.

Images