Parallel multi-step wind power prediction method and device

Abstract

The present invention proposes a parallel multi-step wind power forecasting method and device, wherein the method includes: obtaining unit operating data and meteorological data for a historical period, determining an input matrix based on the unit operating data and meteorological data for the historical period, and then inputting the input matrix into a prediction model for at least one prediction time segment to obtain a power forecast result for each prediction time segment, wherein the prediction time segment is determined based on the prediction period, and then aggregating the power forecast results for each prediction time segment to obtain a target power forecast result for the prediction period. In this way, the prediction period can be segmented to obtain at least one prediction time segment, so that each prediction time segment corresponds to a prediction model, and the prediction models for different prediction time segments extract different data features, effectively improving the accuracy of the prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power generation, and in particular to a parallel multi-step wind power prediction method and device. Background Art

[0002] Wind power forecasting for wind farms is divided into ultra-short-term, short-term, medium-term, and long-term forecasts. However, no matter which type of forecast is used, it is hoped to obtain wind speed or power forecast values ​​at multiple time points in the future. It can be seen that the wind power forecasting problem in the wind power field is a typical multi-step forecasting problem.

[0003] In related technologies, commonly used methods for multi-step forecasting include cumulative forecasting and direct forecasting. However, cumulative forecasting suffers from the problem of cumulative forecast errors. In direct forecasting, the further the forecast time from the current moment, the less correlation there is with historical data, resulting in lower forecast accuracy. Summary of the Invention

[0004] The present invention aims to solve one of the technical problems in the related art at least to a certain extent.

[0005] To this end, the first purpose of the present invention is to propose a parallel multi-step wind power prediction to segment the prediction period and obtain at least one prediction time segment, so that each prediction time segment corresponds to a prediction model. The prediction models of different prediction time segments extract different data features, thereby effectively improving the accuracy of the prediction.

[0006] The second object of the present invention is to provide a parallel multi-step wind power prediction device.

[0007] A third object of the present invention is to provide an electronic device.

[0008] A fourth object of the present invention is to provide a computer-readable storage medium.

[0009] A fifth object of the present invention is to provide a computer program product.

[0010] To achieve the above objectives, a first embodiment of the present invention proposes a parallel multi-step wind power prediction method, comprising:

[0011] Obtain unit operation data and meteorological data for historical periods;

[0012] Determining an input matrix based on the unit operating data and meteorological data for the historical period;

[0013] Inputting the input matrix into a prediction model of at least one prediction time segment respectively to obtain a power prediction result of each prediction time segment, wherein the prediction time segment is determined based on a prediction period;

[0014] The power prediction results of each of the prediction time segments are aggregated to obtain the target power prediction result of the prediction period.

[0015] To achieve the above-mentioned object, a second embodiment of the present invention provides a parallel multi-step wind power prediction device, comprising:

[0016] Acquisition module, used to obtain unit operation data and meteorological data of historical periods;

[0017] A first determination module is configured to determine an input matrix based on the unit operation data and meteorological data of the historical period;

[0018] a first processing module, configured to input the input matrix into a prediction model of at least one prediction time segment, respectively, to obtain a power prediction result for each prediction time segment, wherein the prediction time segment is determined based on a prediction period;

[0019] The summarizing module is used to summarize the power prediction results of each of the prediction time segments to obtain the target power prediction result of the prediction period.

[0020] To achieve the above-mentioned object, a third embodiment of the present invention provides an electronic device, comprising:

[0021] at least one processor; and

[0022] a memory communicatively connected to the at least one processor; wherein,

[0023] The memory stores instructions that can be executed by the at least one processor. The instructions are executed by the at least one processor to enable the at least one processor to execute the parallel multi-step wind power prediction method proposed in the first aspect.

[0024] In order to achieve the above objectives, a fourth embodiment of the present invention proposes a computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable the computer to execute the parallel multi-step wind power prediction method proposed in the first aspect.

[0025] In order to achieve the above objectives, a fifth embodiment of the present invention provides a computer program product, including a computer program, which, when executed by a processor, implements the parallel multi-step wind power prediction method proposed in the first aspect.

[0026] The technical solution provided by the embodiment of the present invention has the following beneficial effects:

[0027] By acquiring unit operating data and meteorological data for historical periods, an input matrix is ​​determined based on the unit operating data and meteorological data for the historical periods, and the input matrix is ​​then input into a prediction model for at least one prediction time segment to obtain a power prediction result for each prediction time segment. The prediction time segment is determined based on the prediction period, and the power prediction results for each prediction time segment are then aggregated to obtain a target power prediction result for the prediction period. Thus, the prediction period can be segmented to obtain at least one prediction time segment, where each prediction time segment corresponds to a prediction model. Prediction models for different prediction time segments extract different data features, effectively improving prediction accuracy.

[0028] Additional aspects and advantages of the present invention will be set forth in part in the description which follows and, in part, will be obvious from the description which follows, or may be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0030] Figure 1 A schematic flow chart of a parallel multi-step wind power prediction method provided by an embodiment of the present invention;

[0031] Figure 2 A schematic diagram of an input matrix provided by an embodiment of the present invention;

[0032] Figure 3 A schematic diagram of a network structure of a prediction model provided by an embodiment of the present invention;

[0033] Figure 4 A schematic diagram of a process for summarizing power prediction results provided by an embodiment of the present invention;

[0034] Figure 5 A schematic flow chart of another parallel multi-step wind power prediction method provided by an embodiment of the present invention;

[0035] Figure 6 A schematic diagram of cross-layer feature fusion provided by an embodiment of the present invention;

[0036] Figure 7 A schematic flow chart of another parallel multi-step wind power prediction method provided by an embodiment of the present invention;

[0037] Figure 8 A schematic flow chart of another parallel multi-step wind power prediction method provided by an embodiment of the present invention;

[0038] Figure 9A schematic flow chart of a parallel multi-step wind power prediction method in a scenario provided by an embodiment of the present invention;

[0039] Figure 10 A schematic structural diagram of a parallel multi-step wind power prediction device provided by an embodiment of the present invention;

[0040] Figure 11 A schematic structural diagram of an electronic device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0041] The following describes embodiments of the present invention in detail, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and are not to be construed as limiting the present invention.

[0042] In related technologies, commonly used methods for multi-step forecasting include cumulative forecasting and direct forecasting. However, cumulative forecasting suffers from the problem of cumulative forecast errors. In direct forecasting, the further the forecast time from the current moment, the less correlation there is with historical data, resulting in lower forecast accuracy.

[0043] To address the above problems, an embodiment of the present invention provides a parallel multi-step prediction method to segment the prediction period and obtain at least one prediction time segment, so that each prediction time segment corresponds to a prediction model. The prediction models of different prediction time segments extract different data features, effectively improving the accuracy of the prediction.

[0044] The following describes a parallel multi-step wind power prediction method and apparatus according to an embodiment of the present invention with reference to the accompanying drawings.

[0045] It should be noted that all actions of acquiring signals, information or data in the present invention are performed in compliance with the corresponding data protection laws and policies of the country where they are located and with the authorization given by the owner of the corresponding device.

[0046] Figure 1 A schematic flow chart of a parallel multi-step wind power prediction method provided by an embodiment of the present invention.

[0047] It should be noted that the parallel multi-step wind power prediction method of the embodiment of the present invention can be performed by the parallel multi-step wind power prediction device provided by the embodiment of the present invention. The parallel multi-step wind power prediction device can be an electronic device or can be configured in an electronic device to implement the parallel multi-step wind power prediction function.

[0048] Among them, the electronic device can be any stationary or mobile computing device capable of data processing, such as mobile computing devices such as laptops, smart phones, wearable devices, or stationary computing devices such as desktop computers, or servers, or other types of computing devices, etc., and this is not limited in this embodiment.

[0049] like Figure 1 As shown, the parallel multi-step wind power prediction method includes the following steps:

[0050] Step 101: Acquire unit operation data and meteorological data for a historical period.

[0051] In this embodiment, the historical period can be any time period before the current time. The embodiment of the present invention does not limit the setting of the length of the historical period. Optionally, the length of the historical period can be set based on manual experience, for example, the length of the historical period can be set to 1 day, 6 hours, etc. Alternatively, the length of the historical period can be dynamically adjusted according to actual application requirements, which is not limited by the embodiment of the present invention.

[0052] In this embodiment, the unit operation data refers to data representing the unit operation information, which may include wind speed, wind direction, power, rotation speed, pitch angle, ambient temperature, ambient humidity, etc. Optionally, the time resolution of the unit operation data may be 15 minutes.

[0053] In this embodiment, meteorological data refers to data representing unit environmental information, and may include wind speed and direction data at different altitudes, such as wind speed and direction data at 10 meters, 30 meters, 50 meters, 70 meters, and 90 meters, as well as ambient temperature, ambient humidity, and atmospheric pressure data. Optionally, the time resolution of the meteorological data may be one hour.

[0054] In this embodiment, the parallel multi-step wind power prediction device can obtain the unit operation data and meteorological data of the historical period in various open, legal and compliant ways. For example, the parallel multi-step wind power prediction device can collect the unit operation data and meteorological data of the historical period in real time, or it can also obtain the unit operation data and meteorological data of the historical period from other devices through network transmission or physical copying, or it can also obtain the unit operation data and meteorological data of the historical period through other open, legal and compliant ways. This is not limited in this embodiment.

[0055] Step 102: Determine an input matrix based on the unit operation data and meteorological data of the historical period.

[0056] In this embodiment, after the unit operation data and meteorological data of the historical period are acquired, the input matrix may be determined according to the unit operation data and meteorological data of the historical period.

[0057] In one possible implementation of this embodiment, when the temporal resolution of the unit operation data and the temporal resolution of the meteorological data are consistent, the unit operation data and the meteorological data are determined as at least one channel of data, thereby constructing an input matrix based on the at least one channel of data. Specifically, the unit operation data and meteorological data can be treated as different channel data and constructed as a Y×N×C matrix, where Y represents the number of sample points, N represents the number of data points, and C represents the number of channels. For example, assuming that the wind power data for the next hour is predicted using historical unit operation data and meteorological data for one day with a temporal resolution of 15 minutes, since there are 96 15-minute periods in a day, there are 96 sample points, thus a 96×N×C input matrix can be constructed. Similarly, since there are 4 15-minute periods in an hour, there are 4 sample points, thus the predicted wind power data for the next hour is a 4×1 vector. Optionally, due to the high dimensionality of the unit data, the unit operation data can occupy two channels and the meteorological data can occupy one channel, forming a 96×N×3 input matrix.

[0058] In order to clearly illustrate the form of the input matrix in this embodiment, this embodiment provides the following Figure 2 The schematic diagram of the input matrix is ​​shown in . Figure 2 As shown, the input matrix includes three channels, namely unit characteristics, unit characteristics and meteorological characteristics. Based on the input matrix, the four powers in the first hour in the future can be predicted, the four powers in the second hour in the future can be predicted, the four powers in the third hour in the future can be predicted, and the four powers in the fourth hour in the future can be predicted.

[0059] Step 103: Input the input matrix to the prediction model of at least one prediction time segment respectively to obtain the power prediction result of each prediction time segment.

[0060] In this embodiment, the prediction time segment is determined based on the prediction period. For example, if the wind power result for the next 4 hours is predicted, the prediction period of the next 4 hours can be divided into 4 segments, that is, each hour is a prediction time segment.

[0061] In this embodiment, each prediction time segment corresponds to a prediction model, so the input matrix can be input into the prediction model of at least one prediction time segment respectively to obtain the power prediction result of each prediction time segment.

[0062] In a possible implementation of this embodiment, the prediction model may include a CNN (Convolutional Neural Network) module and a DNN (Deep Neural Networks) module. In order to clearly illustrate the network structure of the prediction model in this embodiment, this embodiment provides the following example: Figure 3 The network structure diagram of the prediction model is shown in Figure 2. Figure 3 As shown in Figure 2, the prediction model includes two CNN modules and one DNN module. By inputting the input matrix into the prediction model, the power prediction result of a certain prediction time segment can be obtained. Figure 3 As shown in the figure, the data passes through the CNN layer, the leaky linear rectifier Leakyrelu layer, the CNN layer, the Leakyrelu layer, and the Batch Normalization (BN) layer in sequence. That is, the CNN module includes two CNN layers, two Leakyrelu layers, and one BN layer. The DNN module consists of a multi-layer fully connected network. Multiple DNN layers can increase the prediction ability of the model and improve the accuracy of power prediction.

[0063] Step 104 : Summarize the power prediction results of each prediction time segment to obtain the target power prediction result of the prediction period.

[0064] In this embodiment, after obtaining the power prediction results of each prediction time segment, the power prediction results of each prediction time segment may be aggregated to obtain the target power prediction result of the prediction period.

[0065] In a possible implementation of this embodiment, the power prediction result can be vector data, so that the vector data of each prediction time segment can be vector-connected to obtain the target power prediction result of the prediction period, or the vector data of each prediction time segment can be vector-connected to obtain a first vector, and the first vector is input into the Attention layer for weighting to obtain the target power prediction result of the prediction period. In order to clearly illustrate the above process, this embodiment provides the following Figure 4 The flowchart of power prediction result summary is shown in FIG. Figure 4 As shown in Figure 1, the power prediction results of each prediction time segment are summarized in two ways: direct summary and summary with the addition of the Attention mechanism.

[0066] like Figure 4 As shown in the figure, assuming that there are 4 prediction time segments in total, direct aggregation is to directly connect the power prediction results of these 4 prediction time segments by vectors, and aggregate the original 4×1 vectors of each prediction time segment to obtain a 16×1 vector of the prediction period. The 16×1 vector is the target power prediction result of the prediction period.

[0067] Adding the Attention mechanism to aggregate the data involves first performing direct aggregation, aggregating the 4×1 vectors of each prediction time segment into a 16×1 vector for the prediction period. The direct aggregation result is then input into the Attention layer for weighting. Specifically, a weight matrix of 16×1 is designed first. The direct aggregation result is then multiplied by the weight matrix to obtain a new 16×1 matrix. The resulting new 16×1 matrix is ​​the target power prediction result for the prediction period.

[0068] The parallel wind power forecasting method provided by an embodiment of the present invention obtains unit operating data and meteorological data for a historical period, thereby determining an input matrix based on the unit operating data and meteorological data for the historical period, and then inputting the input matrix into a forecasting model for at least one forecast time segment to obtain a power forecast result for each forecast time segment. The forecast time segment is determined based on the forecast period, and the power forecast results for each forecast time segment are then aggregated to obtain a target power forecast result for the forecast period. Thus, the forecast period can be segmented to obtain at least one forecast time segment, so that each forecast time segment corresponds to a forecast model. The forecast models for different forecast time segments extract different data features, effectively improving the accuracy of the forecast.

[0069] In order to clearly illustrate how the prediction model in the present invention obtains the power prediction result based on the input matrix, this embodiment provides another parallel wind power prediction method. Figure 5 A schematic flow chart of another parallel multi-step wind power prediction method provided by an embodiment of the present invention.

[0070] like Figure 5 As shown, the parallel wind power prediction method may include the following steps:

[0071] Step 501: Acquire unit operation data and meteorological data for a historical period.

[0072] Step 502: Determine an input matrix based on the unit operation data and meteorological data of the historical period.

[0073] It should be noted that the execution process of steps 501-502 may specifically refer to steps 101-102 in the above embodiment, and the principles are the same, which will not be repeated here.

[0074] Step 503: Input the input matrix to the first CNN module of the prediction model of at least one prediction time segment for feature extraction to obtain the first feature of each prediction time segment.

[0075] In this embodiment, the input matrix can be input into the first CNN module of the prediction model of at least one prediction time segment for feature extraction to obtain the first feature of each prediction time segment. Figure 3 The first CNN module in the network structure of the prediction model shown in Figure 3 As shown in the figure, when the input matrix is ​​input to the first CNN module of the prediction model for feature extraction, it passes through the CNN layer, Leakyrelu layer, CNN layer, Leakyrelu layer and BN layer in sequence.

[0076] Step 504 : Cross-layer propagation is performed on the first features of each prediction time segment to determine fusion features of each prediction time segment except the first prediction time segment.

[0077] In this embodiment, after obtaining the first features of each predicted time segment, the first features of each predicted time segment can be propagated across layers to determine the fused features of each predicted time segment other than the first predicted time segment. The first predicted time segment is the most forward-looking predicted time segment obtained after segmenting the predicted time period. For example, assuming the predicted time period is the next four hours, the next four hours are segmented to obtain four predicted time segments, namely, the first hour in the future, the second hour in the future, the third hour in the future, and the fourth hour in the future. The first hour in the future is the first predicted time segment.

[0078] Since the first feature of each prediction time segment can be propagated across layers, cross-layer transmission of features between different prediction time segments can be achieved, thereby increasing the accuracy of prediction.

[0079] In a possible implementation of this embodiment, the first feature of the first prediction time segment can be propagated to the prediction model of the second prediction time segment to fuse the first feature of the first prediction time segment and the first feature of the second prediction time segment to obtain the fused feature of the second prediction time segment, and thus, for any prediction time segment other than the first prediction time segment and the second prediction time segment, the first feature of the first prediction time segment and the first feature of the previous prediction time segment of the prediction time segment can be propagated to the prediction model of the prediction time segment to fuse the first feature of the first prediction time segment, the first feature of the previous prediction time segment of the prediction time segment and the first feature of the prediction time segment to obtain the fused feature of the prediction time segment.

[0080] In order to clearly explain how to obtain the fusion features of each prediction time segment in this embodiment, this embodiment provides the following Figure 6 The schematic diagram of cross-layer feature fusion is shown in . Figure 6As shown, for the second prediction time segment, the first feature of the first prediction time segment and the first feature of the second prediction time segment can be matrix-connected, thereby obtaining the fusion feature of the second prediction time segment. Optionally, the matrix connection method may include direct connection, weighted addition, multiplication connection, etc. Among them, direct connection is to directly connect the first feature of the first prediction time segment and the first feature of the second prediction time segment by a matrix, and the number of fusion features obtained after the connection is doubled; the weighted addition method is to assign different weights to the first feature of the first prediction time segment and the first feature of the second prediction time segment, and then perform matrix addition, that is, add the corresponding elements of the matrix; the multiplication connection method is to perform matrix multiplication on the first feature of the first prediction time segment and the first feature of the second prediction time segment, and obtain the final connection result by multiplying the corresponding elements of the matrix.

[0081] In step 505, the first feature of the first prediction time segment is input into the second CNN module of the prediction model of the first prediction time segment for feature extraction to obtain the second feature of the first prediction time segment, and the fusion features of each prediction time segment other than the first prediction time segment are input into the second CNN module of the prediction model of each prediction time segment other than the first prediction time segment for feature extraction to obtain the second feature of each prediction time segment other than the first prediction time segment.

[0082] In this embodiment, there is no fusion feature for the first prediction time segment, and the first feature of the first prediction time segment can be directly input into the second CNN module of the prediction model of the first prediction time segment for feature extraction, thereby obtaining the second feature of the first prediction time segment. However, each prediction time segment except the first prediction time segment has a fusion feature, so the fusion features of each prediction time segment except the first prediction time segment can be input into the second CNN module of the prediction model of each prediction time segment except the first prediction time segment for feature extraction, thereby obtaining the second feature of each prediction time segment except the first prediction time segment. Among them, the first CNN module is Figure 3 The second CNN module in the network structure of the prediction model shown.

[0083] Step 506: Input the second feature of each prediction time segment into the DNN module of the prediction model of the prediction time segment to obtain the power prediction result of each prediction time segment.

[0084] In this embodiment, after obtaining the second feature of each predicted time segment, the second feature of each predicted time segment can be input into the DNN module of the prediction model of the predicted time segment to obtain the power prediction result of each predicted time segment.

[0085] Step 507: Summarize the power prediction results of each prediction time segment to obtain the target power prediction result of the prediction period.

[0086] It should be noted that the execution process of this step may specifically refer to step 104 in the above embodiment, and the principles are the same, so it will not be repeated here.

[0087] The parallel wind power prediction method provided by the embodiment of the present invention inputs the input matrix into the first CNN module of the prediction model of at least one prediction time segment for feature extraction to obtain the first feature of each prediction time segment, realizes cross-layer propagation of the first feature of each prediction time segment to determine the fusion feature of each prediction time segment other than the first prediction time segment, thereby inputting the first feature of the first prediction time segment into the second CNN module of the prediction model of the first prediction time segment for feature extraction to obtain the second feature of the first prediction time segment, and inputting the fusion feature of each prediction time segment other than the first prediction time segment into the second CNN module of the prediction model of each prediction time segment other than the first prediction time segment for feature extraction to obtain the second feature of each prediction time segment other than the first prediction time segment, and then inputting the second feature of each prediction time segment into the DNN module of the prediction model of the prediction time segment to obtain the power prediction result of each prediction time segment. In this way, parallel prediction of multiple prediction time segments can be achieved, and different network structures and weight information can be conveniently designed for each prediction time segment. At the same time, in order to ensure the accuracy of the predicted time segments that are far away from the current moment, the features of the model of the previous time period can be transferred to form a cross-layer transfer effect of features between different predicted time segments, thereby increasing the accuracy of the prediction.

[0088] From the above analysis, it can be seen that in the present invention, the time resolutions of the unit operation data and meteorological data obtained in the historical period can be different. However, when determining the input matrix, the time resolutions of the unit operation data and meteorological data need to be consistent. Therefore, it is necessary to process the unit operation data and meteorological data with inconsistent time resolutions. In order to clearly illustrate this process, this embodiment provides another parallel wind power prediction method. Figure 7 A schematic flow chart of another parallel multi-step wind power prediction method provided by an embodiment of the present invention.

[0089] like Figure 7 As shown, the parallel wind power prediction method may include the following steps:

[0090] Step 701: Acquire unit operation data and meteorological data for a historical period.

[0091] It should be noted that the execution process of this step can specifically refer to step 101 in the above embodiment, and the principles are the same, so it will not be repeated here.

[0092] Step 702 : When the time resolutions of the unit operation data and the meteorological data in the historical period are inconsistent, determine the target time resolution.

[0093] In this embodiment, a target time resolution can be determined when the time resolutions of the unit operation data and meteorological data during a historical period are inconsistent. Optionally, either the time resolution of the unit operation data or the time resolution of the meteorological data can be determined as the target time resolution, or a time resolution other than the time resolution of the unit operation data or the time resolution of the meteorological data can be determined as the target time resolution, which is not limited in this embodiment.

[0094] As an example, assuming that the time resolution of the unit operation data is 15 minutes and the time resolution of the meteorological data is 1 hour, 15 minutes can be determined as the target time resolution, 1 hour can be determined as the target time resolution, 30 minutes can be determined as the target time resolution, and so on. There is no restriction on this in this embodiment.

[0095] Step 703 : In response to the time resolution of the unit operation data being inconsistent with the target time resolution, interpolation processing is performed on the unit operation data to obtain unit operation data having a time resolution consistent with the target time resolution.

[0096] In this embodiment, when the time resolution of the unit operation data is inconsistent with the target time resolution, in order to ensure the consistency between the time resolution of the unit operation data and the target time resolution, it is necessary to interpolate the unit operation data to obtain unit operation data with a time resolution consistent with the target time resolution.

[0097] Optionally, the unit operation data can be directly repeated, such as by repeated filling forward or repeated filling backward, to obtain unit operation data with a time resolution consistent with the target time resolution. For example, assuming that the target time resolution is 15 minutes, and the data between 1 hour and 2 hours needs to be filled, then the forward repeated filling is to replace the data that needs to be filled in the time period with 1 hour of unit operation data, thereby obtaining 15 minutes of unit operation data. Similarly, the backward repeated filling is to replace the data that needs to be filled in the time period with 2 hours of unit operation data, thereby obtaining 15 minutes of unit operation data; or a linear interpolation method can be used to obtain unit operation data with a time resolution consistent with the target time resolution. Specifically, the part that needs to be filled can be regarded as a linear change process, and the missing unit operation data can be obtained by linear interpolation; or the average of the two moments can be used as the result for filling, etc. This is not limited in this embodiment.

[0098] Step 704 : In response to the time resolution of the meteorological data being inconsistent with the target time resolution, interpolation processing is performed on the meteorological data to obtain meteorological data having a time resolution consistent with the target time resolution.

[0099] In this embodiment, when the time resolution of the meteorological data is inconsistent with the target time resolution, in order to ensure the consistency of the time resolution of the meteorological data with the target time resolution, it is necessary to interpolate the meteorological data to obtain meteorological data with a time resolution consistent with the target time resolution.

[0100] Optionally, meteorological data can be directly repeated, such as repeated filling forward or repeated filling backward, to obtain meteorological data with a time resolution consistent with the target time resolution. For example, assuming that the target time resolution is 15 minutes, and data between 1 hour and 2 hours needs to be filled, then forward repeated filling is to replace the data that needs to be filled in this time period with 1 hour of meteorological data, thereby obtaining 15 minutes of meteorological data. Similarly, backward repeated filling is to replace the data that needs to be filled in this time period with 2 hours of meteorological data, thereby obtaining 15 minutes of meteorological data; or a linear interpolation method can be used to obtain meteorological data with a time resolution consistent with the target time resolution. Specifically, the part that needs to be filled can be regarded as a linear change process, and the missing meteorological data can be obtained by linear interpolation; or the average of the two moments can be used as the result for filling, and so on. This is not limited in this embodiment.

[0101] Step 705: Determine the input matrix based on the unit operation data and meteorological data of the historical period.

[0102] Step 706: Input the input matrix to the prediction model of at least one prediction time segment respectively to obtain the power prediction result of each prediction time segment.

[0103] Step 707: Summarize the power prediction results of each prediction time segment to obtain the target power prediction result of the prediction period.

[0104] It should be noted that the execution process of steps 705-707 may specifically refer to steps 102-104 in the above embodiment, and the principles are the same, so they will not be repeated here.

[0105] The parallel wind power prediction method provided by an embodiment of the present invention determines a target time resolution when the time resolutions of the unit operation data and meteorological data in a historical period are inconsistent. Thus, in response to the inconsistency between the time resolution of the unit operation data and the target time resolution, the unit operation data is interpolated to obtain unit operation data with a time resolution consistent with the target time resolution, and / or, in response to the inconsistency between the time resolution of the meteorological data and the target time resolution, the meteorological data is interpolated to obtain meteorological data with a time resolution consistent with the target time resolution. Thus, when the time resolutions of the unit operation data and meteorological data obtained in the historical period are inconsistent, the time resolutions of the unit operation data and meteorological data in the historical period can be made consistent through interpolation.

[0106] It should be noted that, in the present invention, after the input matrix is ​​determined, the data in the input matrix needs to be normalized. In order to clearly illustrate this process, this embodiment provides another parallel wind power prediction method. Figure 8 A schematic flow chart of another parallel multi-step wind power prediction method provided by an embodiment of the present invention.

[0107] like Figure 8 As shown, the parallel wind power prediction method may include the following steps:

[0108] Step 801: Acquire unit operation data and meteorological data for a historical period.

[0109] Step 802: Determine an input matrix based on the unit operation data and meteorological data of the historical period.

[0110] It should be noted that the execution process of steps 801-802 may specifically refer to steps 101-102 in the above embodiment, and the principles are the same, so they will not be repeated here.

[0111] Step 803: Use the cosine normalization method to normalize the wind direction data in the input matrix.

[0112] In this embodiment, since the wind direction data range is [0, 360], and for the unit control strategy, there is no obvious difference between 360 degrees and 0 degrees, in order to eliminate the numerical influence, the cosine normalization method can be used to normalize the wind direction data in the input matrix. The specific calculation formula is:

[0113]

[0114] in, Represents the normalized value of any wind direction data in the input matrix, and x represents any wind direction data in the input matrix.

[0115] Step 804 : normalize the data in the input matrix except the wind direction data using a mean normalization method.

[0116] In this embodiment, the mean normalization method can be used to normalize the data other than the wind direction data in the input matrix. The specific calculation formula is:

[0117]

[0118] in, represents the normalized value of any data in the input matrix except wind direction data, x represents any data in the input matrix except wind direction data, mean(x) represents the mean of the column where data x is located, and std(x) represents the standard deviation of the column where data x is located.

[0119] Step 804: Input the input matrix to the prediction model of at least one prediction time segment to obtain the power prediction result of each prediction time segment.

[0120] Step 805 , summarizing the power prediction results of each prediction time segment to obtain the target power prediction result of the prediction period.

[0121] It should be noted that the execution process of steps 804-805 may specifically refer to steps 103-104 in the above embodiment, and the principles are the same, so they will not be repeated here.

[0122] The parallel wind power prediction method provided by an embodiment of the present invention adopts a cosine normalization method to normalize the wind direction data in the input matrix, and adopts a mean normalization method to normalize other data in the input matrix except the wind direction data. After the input matrix is ​​determined, the data in the input matrix can be normalized.

[0123] In order to clearly illustrate the above embodiment, an example is now given for illustration.

[0124] Figure 9A schematic flow chart of a parallel wind power prediction method in a scenario provided by an embodiment of the present invention.

[0125] like Figure 9 As shown, the parallel wind power prediction method includes the following steps:

[0126] Step 901, obtain data. The present invention conducts short-term wind power forecasting research, such as predicting wind power data for the next 4 hours, with a target time resolution of 15 minutes. In this embodiment, unit operation data and meteorological data for historical periods can be obtained. The unit operation data obtained may include data such as wind speed, wind direction, power, speed, pitch angle, ambient temperature, ambient humidity, etc., with a time resolution of 15 minutes. The meteorological data obtained may include wind speed and wind direction data at different altitudes, such as wind speed and wind direction data at 10 meters, 30 meters, 50 meters, 70 meters, and 90 meters, as well as ambient temperature, ambient humidity, atmospheric pressure data, etc., with a time resolution of 1 hour.

[0127] Step 902, determine the input matrix. Since the time resolution of the unit operation data is 15 minutes, and the time resolution of the meteorological data is 1 hour, at the same time, the wind power forecast result is required to be a 15-minute point-by-point result for 4 hours, that is, 16 sample points, and the target time resolution is 15 minutes. At this time, since the time resolution of the unit operation data is consistent with the target time resolution, and the time resolution of the meteorological data is inconsistent with the target time resolution, it is necessary to interpolate the meteorological data. Optionally, the meteorological data can be directly repeated, such as repeated filling forward or repeated filling backward, to obtain meteorological data with a time resolution consistent with the target time resolution. For example, assuming that the target time resolution is 15 minutes, and the data between 1 hour and 2 hours needs to be filled, then repeated filling forward is to replace the data that needs to be filled in this time period with 1 hour of meteorological data, thereby obtaining 15 minutes of meteorological data. Similarly, backward repeated filling is to replace the data that needs to be filled in the time period with 2 hours of meteorological data, thereby obtaining 15 minutes of meteorological data; or a linear interpolation method can be used to obtain meteorological data with a time resolution consistent with the target time resolution. Specifically, the part that needs to be filled can be regarded as a linear change process, and the missing meteorological data can be obtained by linear interpolation; or the average of the two moments can be used as the result for filling, and so on. This is not limited in this embodiment.

[0128] After the meteorological data is filled, the time resolution of the unit operation data and the time resolution of the meteorological data are consistent with the target time resolution. Therefore, the input matrix can be determined based on the unit operation data and meteorological data of the historical period. Optionally, the unit operation data and meteorological data can be determined as at least one channel data, and the input matrix can be constructed based on the at least one channel data. Specifically, the unit operation data and meteorological data can be treated as different channel data and constructed into a Y×N×C matrix, where Y represents the number of sample points, N represents the number of data, and C represents the number of channels. For example, assuming that the unit operation data and meteorological data of the past day are used to predict the wind power data for the next hour, the time resolution is 15 minutes. Since there are 96 15-minute periods in a day, there are 96 sample points. Therefore, a 96×N×C input matrix can be constructed. Similarly, since there are 4 15-minute periods in an hour, there are 4 sample points. Therefore, the predicted wind power data for the next hour is a 4×1 vector. Optionally, since the unit data has more dimensions, the unit operation data can occupy two channels, and the meteorological data can occupy one channel, forming an input matrix of 96×N×3.

[0129] After determining the input matrix, the data in the input matrix needs to be normalized. In this embodiment, since the wind direction data range is [0, 360], and for the unit control strategy, there is no obvious difference between 360 degrees and 0 degrees, in order to eliminate the numerical influence, the cosine normalization method can be used to normalize the wind direction data in the input matrix. The specific calculation formula is:

[0130]

[0131] in, Represents the normalized value of any wind direction data in the input matrix, and x represents any wind direction data in the input matrix.

[0132] For other data in the input matrix except wind direction data, the mean normalization method can be used for normalization. The specific calculation formula is:

[0133]

[0134] in, represents the normalized value of any data in the input matrix except wind direction data, x represents any data in the input matrix except wind direction data, mean(x) represents the mean of the column where data x is located, and std(x) represents the standard deviation of the column where data x is located.

[0135] Step 903: Build the prediction model. Step 902 builds the input matrix for each prediction time segment. Step 903 builds a separate prediction model for each prediction time segment, learning weight information for each model independently. The model building process is described using the first prediction time segment as an example; the process for other time segments is similar.

[0136] The prediction time segment is determined based on the prediction period. For example, if the wind power results for the next 4 hours are predicted, the prediction period of the next 4 hours can be divided into 4 segments, that is, each hour is a prediction time segment.

[0137] It should be noted that the H1, H2, H3, and H4 data structures in the figure are all input matrices determined in step 902. H1 represents the first prediction time segment, H2 represents the second prediction time segment, H3 represents the third prediction time segment, and H4 represents the fourth prediction time segment. The prediction results H1, H2, H3, and H4 in the figure are the prediction results output by the prediction model for the corresponding prediction time segment.

[0138] In this embodiment, the model network structure of the prediction time segment H1 is as follows: Figure 3 As shown in Figure 1, the H1 data structure is the input matrix determined in step 902. The input matrix is ​​subjected to feature extraction by multiple CNN modules, and then subjected to result prediction by the DNN module, and finally the prediction result of the prediction time segment is obtained, namely the H1 prediction result. Figure 3 As shown in the figure, the data passes through the CNN layer, leakyrelu layer, CNN layer, Leakyrelu layer, and BN layer in sequence. That is, the CNN module includes two CNN layers, two Leakyrelu layers and one BN layer; the DNN module is composed of a multi-layer fully connected network. Multiple DNN layers can increase the prediction ability of the model and improve the accuracy of power prediction.

[0139] In this embodiment, the H1 data structure is the input matrix determined in step 902 and input to the first CNN module ( Figure 9 The first CNN module in the model network structure of the predicted time segment H1 is used to extract features, and the first feature of the first predicted time segment H1 can be obtained. Then, the first feature of the first predicted time segment H1 is input into the second CNN module ( Figure 9 The second CNN module in the model network structure of the predicted time segment H1 is used for feature extraction to obtain the second feature of the first predicted time segment H1. Finally, the second feature of the first predicted time segment H1 is input into the DNN module to obtain the power prediction result of the first predicted time segment H1, that is, the H1 prediction result.

[0140] The model structure of other prediction time segments is similar to the model network structure of the prediction time segment H1. The difference is that the features of other layers take into account the feature information of the first layer and the previous layer. The cross-layer propagation method is used to propagate the feature information of the first layer and the previous layer across layers to increase the accuracy of other prediction periods. The cross-layer feature fusion process is as follows: Figure 6 As shown, the first feature of the first prediction time segment and the first feature of the second prediction time segment are matrix-connected to obtain the fused feature of the second prediction time segment. Optionally, the matrix connection method is divided into direct connection, weighted addition, multiplication connection and the like. Among them, direct connection is to directly connect the first feature of the first prediction time segment and the first feature of the second prediction time segment by a matrix, and the number of fused features obtained after the connection is doubled; the weighted addition method is to assign different weights to the first feature of the first prediction time segment and the first feature of the second prediction time segment, and then perform matrix addition, that is, add the corresponding elements of the matrix; the multiplication connection method is to perform matrix multiplication on the first feature of the first prediction time segment and the first feature of the second prediction time segment, and obtain the final connection result by multiplying the corresponding elements of the matrix.

[0141] In this embodiment, for the second prediction time segment H2, the first feature of the first prediction time segment H1 and the first feature of the second prediction time segment H2 are matrix-connected to obtain the fused feature of the second prediction time segment H2. For the third prediction time segment H3, the first feature of the first prediction time segment H1 and the first feature of the previous prediction time segment of the third prediction time segment H3, that is, the first feature of the second prediction time segment H2 and the first feature of the third prediction time segment H3, are matrix-connected to obtain the fused feature of the third prediction time segment H3. For the fourth prediction time segment H4, the first feature of the first prediction time segment H1 and the first feature of the previous prediction time segment of the fourth prediction time segment H4, that is, the first feature of the third prediction time segment H3 and the first feature of the fourth prediction time segment H4, are matrix-connected to obtain the fused feature of the fourth prediction time segment H4.

[0142] In this embodiment, the H2 data structure is the input matrix determined in step 902 and input to the first CNN module ( Figure 9 The first CNN module in the model network structure of the predicted time segment H2 is used for feature extraction to obtain the first feature of the second predicted time segment H2. The first feature of the first predicted time segment H1 and the first feature of the second predicted time segment H2 are then matrix-connected to obtain the fused feature of the second predicted time segment H2. The fused feature of the second predicted time segment H2 is then input into the second CNN module ( Figure 9The second CNN module in the model network structure of the predicted time segment H2 is used for feature extraction to obtain the second feature of the second predicted time segment H2. Finally, the second feature of the second predicted time segment H2 is input into the DNN module to obtain the power prediction result of the second predicted time segment H2, that is, the H2 prediction result.

[0143] In this embodiment, the H3 data structure is the input matrix determined in step 902 and input to the first CNN module ( Figure 9 The first CNN module in the model network structure of the predicted time segment H3 is used for feature extraction to obtain the first feature of the third predicted time segment H3. Then, the first feature of the first predicted time segment H1, the first feature of the previous predicted time segment of the third predicted time segment H3, that is, the first feature of the second predicted time segment H2 and the first feature of the third predicted time segment H3 are matrix-connected to obtain the fused feature of the third predicted time segment H3. The fused feature of the third predicted time segment H3 is then input into the second CNN module ( Figure 9 The second CNN module in the model network structure of the predicted time segment H3 is used for feature extraction to obtain the second feature of the third predicted time segment H3. Finally, the second feature of the third predicted time segment H3 is input into the DNN module to obtain the power prediction result of the third predicted time segment H3, that is, the H3 prediction result.

[0144] In this embodiment, the H4 data structure is the input matrix determined in step 902 and input to the first CNN module ( Figure 9 The first CNN module in the model network structure of the predicted time segment H4 is used for feature extraction to obtain the first feature of the fourth predicted time segment H4. Then, the first feature of the first predicted time segment H1, the first feature of the previous predicted time segment of the fourth predicted time segment H4, that is, the first feature of the third predicted time segment H3 and the first feature of the fourth predicted time segment H4 are matrix-connected to obtain the fused feature of the fourth predicted time segment H4. The fused feature of the fourth predicted time segment H4 is then input into the second CNN module ( Figure 9 The second CNN module in the model network structure of the predicted time segment H4 is used for feature extraction to obtain the second feature of the fourth predicted time segment H4. Finally, the second feature of the fourth predicted time segment H4 is input into the DNN module to obtain the power prediction result of the fourth predicted time segment H4, that is, the H4 prediction result.

[0145] In summary, through the time period modeling method in step 903, the power prediction results of different prediction time segments can be obtained.

[0146] Step 904: Fusion of prediction results. This module aggregates the prediction results to obtain the target power prediction results for the prediction period. Figure 4 As shown in Figure 1, the power prediction results of each prediction time segment are summarized in two ways: direct summary and summary with the addition of the Attention mechanism.

[0147] Assuming that there are four prediction time segments, direct aggregation means directly connecting the power prediction results of these four prediction time segments by vector. The original 4×1 vector of each prediction time segment is aggregated to obtain a 16×1 vector of the prediction period. The 16×1 vector is the target power prediction result of the prediction period.

[0148] Adding the Attention mechanism to aggregate the data involves first performing direct aggregation, aggregating the 4×1 vectors of each prediction time segment into a 16×1 vector for the prediction period. The direct aggregation result is then input into the Attention layer for weighting. Specifically, a weight matrix of 16×1 is designed first. The direct aggregation result is then multiplied by the weight matrix to obtain a new 16×1 matrix. The resulting new 16×1 matrix is ​​the target power prediction result for the prediction period.

[0149] After the network structure is designed, model training is performed to obtain the final optimal model. For the data to be predicted, the same data construction method is used and input into the trained optimal model to obtain the final prediction result, thereby realizing parallel multi-step power prediction capabilities and effectively improving the accuracy of power prediction.

[0150] In order to implement the above embodiment, the present invention also proposes a parallel multi-step wind power prediction device.

[0151] Figure 10 A schematic structural diagram of a parallel multi-step wind power prediction device provided by an embodiment of the present invention.

[0152] like Figure 10 As shown, the parallel multi-step wind power prediction device includes: an acquisition module 11, a first determination module 12, a first processing module 13 and a summary module 14.

[0153] Acquisition module 11, used to obtain unit operation data and meteorological data of historical periods;

[0154] A first determination module 12 is used to determine an input matrix based on unit operation data and meteorological data of a historical period;

[0155] A first processing module 13 is configured to input the input matrix into a prediction model of at least one prediction time segment to obtain a power prediction result for each prediction time segment, wherein the prediction time segment is determined based on a prediction period;

[0156] The summarizing module 14 is configured to summarize the power prediction results of each prediction time segment to obtain a target power prediction result for the prediction period.

[0157] Furthermore, in a possible implementation of the embodiment of the present invention, the prediction model includes a convolutional neural network (CNN) module and a deep neural network (DNN) module; the first processing module 13 includes:

[0158] a first feature extraction unit, configured to input the input matrix into a first CNN module of a prediction model of at least one prediction time segment for feature extraction, so as to obtain a first feature of each prediction time segment;

[0159] a cross-layer propagation unit, configured to perform cross-layer propagation on the first features of each prediction time segment to determine fusion features of each prediction time segment except the first prediction time segment;

[0160] a second feature extraction unit, configured to input the first feature of the first prediction time segment into a second CNN module of the prediction model of the first prediction time segment for feature extraction to obtain a second feature of the first prediction time segment, and input the fused features of each prediction time segment other than the first prediction time segment into the second CNN module of the prediction model of each prediction time segment other than the first prediction time segment for feature extraction to obtain a second feature of each prediction time segment other than the first prediction time segment;

[0161] The input unit inputs the second feature of each prediction time segment into the DNN module of the prediction model of the prediction time segment to obtain the power prediction result of each prediction time segment.

[0162] Furthermore, in a possible implementation of the embodiment of the present invention, the cross-layer propagation unit is further configured to:

[0163] Propagating the first feature of the first prediction time segment to the prediction model of the second prediction time segment to fuse the first feature of the first prediction time segment and the first feature of the second prediction time segment to obtain a fused feature of the second prediction time segment;

[0164] For any prediction time segment except the first prediction time segment and the second prediction time segment, the first feature of the first prediction time segment and the first feature of the previous prediction time segment of the prediction time segment are propagated to the prediction model of the prediction time segment to fuse the first feature of the first prediction time segment, the first feature of the previous prediction time segment of the prediction time segment and the first feature of the prediction time segment to obtain the fused feature of the prediction time segment.

[0165] Furthermore, in a possible implementation of an embodiment of the present invention, the CNN module includes two CNN layers, two leaky linear rectification Leakyrelu layers and a batch normalization BN layer.

[0166] Furthermore, in a possible implementation of the embodiment of the present invention, the power prediction result is vector data; the summarizing module 14 is further configured to:

[0167] Perform vector connection on the vector data of each prediction time segment to obtain the target power prediction result of the prediction period;

[0168] or;

[0169] Performing vector connection on the vector data of each prediction time segment to obtain a first vector;

[0170] The first vector is input into the Attention layer for weighting to obtain the target power prediction result of the prediction period.

[0171] Furthermore, in a possible implementation of the embodiment of the present invention, the first determining module 12 is further configured to:

[0172] Determine the unit operation data and meteorological data as at least one channel of data;

[0173] Construct an input matrix based on at least one channel data.

[0174] Furthermore, in a possible implementation of the embodiment of the present invention, the apparatus further includes:

[0175] The second determination module is used to determine the target time resolution when the time resolutions of the unit operation data and the meteorological data in the historical period are inconsistent;

[0176] a second processing module, configured to, in response to a time resolution of the unit operation data being inconsistent with a target time resolution, perform interpolation processing on the unit operation data to obtain unit operation data having a time resolution consistent with the target time resolution;

[0177] And / or, in response to the time resolution of the meteorological data being inconsistent with the target time resolution, interpolation processing is performed on the meteorological data to obtain meteorological data having a time resolution consistent with the target time resolution.

[0178] Furthermore, in a possible implementation of the embodiment of the present invention, the apparatus further includes:

[0179] The third processing module is used to perform normalization processing on the data in the input matrix.

[0180] Furthermore, in a possible implementation of the embodiment of the present invention, the third processing module is further configured to:

[0181] The cosine normalization method is used to normalize the wind direction data in the input matrix;

[0182] The mean normalization method is used to normalize the data in the input matrix except the wind direction data.

[0183] It should be noted that the above explanation of the embodiment of the parallel multi-step wind power prediction method is also applicable to the parallel multi-step wind power prediction device of this embodiment, and will not be repeated here.

[0184] The parallel wind power prediction device provided by an embodiment of the present invention obtains unit operating data and meteorological data for a historical period, thereby determining an input matrix based on the unit operating data and meteorological data for the historical period, and then inputting the input matrix into a prediction model for at least one prediction time segment to obtain a power prediction result for each prediction time segment, wherein the prediction time segment is determined based on the prediction period, and then the power prediction results for each prediction time segment are aggregated to obtain a target power prediction result for the prediction period. In this way, the prediction period can be segmented to obtain at least one prediction time segment, so that each prediction time segment corresponds to a prediction model, and the prediction models for different prediction time segments extract different data features, effectively improving the accuracy of the prediction.

[0185] In order to implement the above embodiments, the present invention also proposes an electronic device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the parallel wind power prediction method proposed in any of the above embodiments of the present invention.

[0186] Figure 11 This is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. It should be noted that: Figure 11 The electronic device shown is only an example and should not limit the functions and scope of use of the embodiments of the present invention.

[0187] like Figure 11As shown, the electronic device may include: a shell 21, a processor 22, a memory 23, a circuit board 24 and a power supply circuit 25, wherein the circuit board 24 is placed inside the space enclosed by the shell 21, and the processor 22 and the memory 23 are arranged on the circuit board 24; the power supply circuit 25 is used to supply power to various circuits or devices of the above-mentioned electronic device; the memory 23 is used to store executable program code; the processor 22 runs the program corresponding to the executable program code by reading the executable program code stored in the memory 23, so as to execute the parallel wind power prediction method proposed in any of the above embodiments of the present invention.

[0188] For details on the specific execution process of the above steps by the processor 22 and the steps further executed by the processor 22 by running the executable program code, please refer to the present invention. Figure 1-9 The description of the illustrated embodiment will not be repeated here.

[0189] To implement the above embodiments, the present invention further proposes a computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable a computer to execute the parallel wind power prediction method proposed in any of the above embodiments of the present invention.

[0190] In order to implement the above embodiments, the present invention further proposes a computer program product, including a computer program, which, when executed by a processor, implements the parallel wind power prediction method proposed in any of the above embodiments of the present invention.

[0191] In the description of this specification, the reference terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0192] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of the present invention, "plurality" means at least two, such as two, three, etc., unless otherwise specifically defined.

[0193] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code comprising one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present invention includes alternative implementations in which functions may be performed out of the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present invention pertain.

[0194] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and programmable read-only memory (EPROM or flash memory), fiber optic devices, and a portable compact disc read-only memory (CDROM). Furthermore, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing it in another suitable manner if necessary, and then storing it in a computer memory.

[0195] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application-specific integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0196] Those skilled in the art will understand that all or part of the steps in the method of the above embodiment can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiment.

[0197] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing module, or each unit may exist physically separately, or two or more units may be integrated into a single module. The aforementioned integrated modules may be implemented in the form of hardware or in the form of software functional modules. If the integrated modules are implemented in the form of software functional modules and sold or used as independent products, they may also be stored in a computer-readable storage medium.

[0198] The storage medium mentioned above may be a read-only memory, a magnetic disk, or an optical disk, etc. Although the embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and are not to be construed as limiting the present invention. Persons skilled in the art may make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

Claims

1. A parallel multi-step wind power prediction method, characterized in that: The following steps are involved: Obtain unit operation data and meteorological data for historical periods; Determining an input matrix based on the unit operating data and meteorological data for the historical period; Inputting the input matrix into a prediction model of at least one prediction time segment respectively to obtain a power prediction result of each prediction time segment, wherein the prediction time segment is determined based on a prediction period; Summarizing the power prediction results of each of the prediction time segments to obtain a target power prediction result for the prediction period; The prediction model includes a convolutional neural network (CNN) module and a deep neural network (DNN) module; and the input matrix is ​​input into the prediction model of at least one prediction time segment to obtain a power prediction result for each prediction time segment, including: Inputting the input matrix into the first CNN module of the prediction model of at least one prediction time segment for feature extraction to obtain a first feature of each prediction time segment; Performing cross-layer propagation on the first features of each of the predicted time segments to determine fused features of each of the predicted time segments except the first predicted time segment; Inputting the first feature of the first prediction time segment into the second CNN module of the prediction model of the first prediction time segment for feature extraction to obtain the second feature of the first prediction time segment, and inputting the fused features of each of the prediction time segments other than the first prediction time segment into the second CNN module of the prediction model of each of the prediction time segments other than the first prediction time segment for feature extraction to obtain the second feature of each of the prediction time segments other than the first prediction time segment; The second feature of each predicted time segment is input into the DNN module of the prediction model of the predicted time segment to obtain the power prediction result of each predicted time segment.

2. The method according to claim 1, characterized in that The cross-layer propagation of the first features of each of the predicted time segments to determine the fusion features of each of the predicted time segments except the first predicted time segment includes: Propagating the first feature of the first prediction time segment to the prediction model of the second prediction time segment, so as to fuse the first feature of the first prediction time segment and the first feature of the second prediction time segment to obtain a fused feature of the second prediction time segment; For any prediction time segment other than the first prediction time segment and the second prediction time segment, the first feature of the first prediction time segment and the first feature of the previous prediction time segment of the prediction time segment are propagated to the prediction model of the prediction time segment to fuse the first feature of the first prediction time segment, the first feature of the previous prediction time segment of the prediction time segment and the first feature of the prediction time segment to obtain the fused feature of the prediction time segment.

3. The method according to claim 1, characterized in that The CNN module includes two CNN layers, two leaky linear rectification Leakyrelu layers and a batch normalization layer BN.

4. The method according to claim 1, wherein The power prediction result is vector data; the power prediction results of each prediction time segment are aggregated to obtain the target power prediction result of the prediction period, including: Performing vector connection on the vector data of each of the prediction time segments to obtain a target power prediction result for the prediction period; or; Performing vector connection on the vector data of each of the predicted time segments to obtain a first vector; The first vector is input into the Attention layer for weighting to obtain the target power prediction result of the prediction period.

5. The method according to claim 1, wherein The step of determining the input matrix based on the unit operation data and meteorological data of the historical period includes: determining the unit operation data and the meteorological data as at least one channel data; An input matrix is ​​constructed according to the at least one channel data.

6. The method according to any one of claims 1 to 5, characterized in that Before determining the input matrix based on the unit operation data and meteorological data of the historical period, the method further includes: In the event that the time resolutions of the unit operation data and the meteorological data during the historical period are inconsistent, determining a target time resolution; In response to a time resolution of the unit operation data being inconsistent with the target time resolution, interpolating the unit operation data to obtain unit operation data with a time resolution consistent with the target time resolution; And / or, in response to the time resolution of the meteorological data being inconsistent with the target time resolution, interpolation processing is performed on the meteorological data to obtain meteorological data having a time resolution consistent with the target time resolution.

7. The method according to any one of claims 1 to 5, characterized in that After determining the input matrix based on the unit operation data and meteorological data of the historical period, the method further includes: Normalize the data in the input matrix.

8. The method according to claim 7, characterized in that Normalizing the data in the input matrix includes: Using a cosine normalization method, the wind direction data in the input matrix is ​​normalized; A mean normalization method is adopted to normalize the other data in the input matrix except the wind direction data.

9. A parallel multi-step wind power prediction device, characterized in that: include: Acquisition module, used to obtain unit operation data and meteorological data of historical periods; A first determination module is configured to determine an input matrix based on the unit operation data and meteorological data of the historical period; a first processing module, configured to input the input matrix into a prediction model of at least one prediction time segment, respectively, to obtain a power prediction result for each prediction time segment, wherein the prediction time segment is determined based on a prediction period; A summarizing module, configured to summarize the power prediction results of each of the prediction time segments to obtain a target power prediction result for the prediction period; The prediction model includes a convolutional neural network (CNN) module and a deep neural network (DNN) module; the first processing module includes: a first feature extraction unit, configured to input the input matrix into a first CNN module of a prediction model of at least one prediction time segment for feature extraction, so as to obtain a first feature of each prediction time segment; a cross-layer propagation unit, configured to perform cross-layer propagation on the first features of each of the prediction time segments to determine a fusion feature of each of the prediction time segments except the first prediction time segment; a second feature extraction unit, configured to input the first feature of the first prediction time segment into a second CNN module of a prediction model of the first prediction time segment for feature extraction, so as to obtain a second feature of the first prediction time segment, and input the fused features of each of the prediction time segments other than the first prediction time segment into the second CNN module of the prediction model of each of the prediction time segments other than the first prediction time segment for feature extraction, so as to obtain a second feature of each of the prediction time segments other than the first prediction time segment; An input unit is used to input the second feature of each prediction time segment into the DNN module of the prediction model of the prediction time segment to obtain the power prediction result of each prediction time segment.

Images