CN114021803A - Wind power prediction method, system and equipment based on convolution transform architecture - Google Patents

Wind power prediction method, system and equipment based on convolution transform architecture Download PDF

Info

Publication number
CN114021803A
CN114021803A CN202111274987.2A CN202111274987A CN114021803A CN 114021803 A CN114021803 A CN 114021803A CN 202111274987 A CN202111274987 A CN 202111274987A CN 114021803 A CN114021803 A CN 114021803A
Authority
CN
China
Prior art keywords
data
power prediction
power
wind
vector
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202111274987.2A
Other languages
Chinese (zh)
Inventor
卢泽华
杨正军
郝小会
杨立平
强威威
张小龙
张奔
刘雪峰
任鑫
李小翔
王�华
童彤
吕亮
李邦兴
武青
杨永前
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Huaneng Clean Energy Research Institute
Huaneng Jiuquan Wind Power Co Ltd
Original Assignee
Huaneng Clean Energy Research Institute
Huaneng Jiuquan Wind Power Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Huaneng Clean Energy Research Institute, Huaneng Jiuquan Wind Power Co Ltd filed Critical Huaneng Clean Energy Research Institute
Priority to CN202111274987.2A priority Critical patent/CN114021803A/en
Priority to PCT/CN2022/072596 priority patent/WO2023070960A1/en
Publication of CN114021803A publication Critical patent/CN114021803A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/047—Probabilistic or stochastic networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Health & Medical Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Physics (AREA)
  • Biophysics (AREA)
  • Evolutionary Computation (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Computational Linguistics (AREA)
  • Software Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Economics (AREA)
  • Strategic Management (AREA)
  • Human Resources & Organizations (AREA)
  • General Business, Economics & Management (AREA)
  • Tourism & Hospitality (AREA)
  • Marketing (AREA)
  • Water Supply & Treatment (AREA)
  • Primary Health Care (AREA)
  • Public Health (AREA)
  • Probability & Statistics with Applications (AREA)
  • Development Economics (AREA)
  • Game Theory and Decision Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Supply And Distribution Of Alternating Current (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The application discloses a method, a device and a storage medium for wind power prediction based on a convolution transform architecture, relates to the technical field of new energy power, and particularly relates to a method, a device and a storage medium for wind power prediction based on a convolution transform architecture. The specific implementation scheme is as follows: acquiring meteorological data and operating data, and acquiring an embedded vector; inputting the embedded vector into a power prediction network, the power prediction network comprising an encoder and a decoder; acquiring a feature map corresponding to the embedded vector according to the encoder; inputting the feature map into a decoder to generate a predicted power. According to the embodiment of the application, the power of wind power generation can be predicted according to meteorological data and operation data, the influence of abnormal data on a prediction result is reduced by paying attention to a plurality of time point data, and the accuracy of power prediction is improved.

Description

Wind power prediction method, system and equipment based on convolution transform architecture
Technical Field
The present application relates to new energy power technologies, and in particular, to a wind power prediction method, system, and device based on a convolution transform architecture.
Background
Wind power generation technology is becoming a major source for meeting future power demands. Higher share of renewable energy technology is crucial for carbon neutralization to meet the demands of future new power system grids, but also presents new grid operating challenges. The power company needs to predict the wind power generation power in order to perform the power generation scheduling operation. Prediction is a major driving factor to ensure safe and economical wind power grid connection, while establishing a link between many flexible innovations at different levels of the power system to achieve synergistic effects. Accurate wind power prediction is an important, cost-effective energy management element that also helps wind power plants and aggregate systems to efficiently and directly participate in the electricity market and to increase the efficiency of the plant by optimizing supply plans.
In the related technology, the wind energy power generation power is predicted according to a model of a recurrent neural network class, but the recurrent neural network has the problems of gradient disappearance and gradient explosion when the network deepens, and the accuracy of power prediction is low.
Disclosure of Invention
The application provides a wind power prediction method, a system and equipment based on a convolution transformer architecture. The technical scheme of the application is as follows:
according to a first aspect of embodiments of the present application, there is provided a wind power prediction method based on a convolutional transformer architecture, including:
and acquiring meteorological data and operating data, and acquiring an embedded vector.
Inputting the embedded vector into a power prediction network, the power prediction network comprising an encoder and a decoder.
And acquiring a characteristic diagram corresponding to the embedded vector according to the encoder.
Inputting the feature map into a decoder to generate a predicted power.
Optionally, the time step of the meteorological data is t, and the meteorological data includes:
station rated capacity, power generation unit model, power generation unit quantity and capacity expansion information.
And actual power of the station output meter.
Wind height, wind speed and wind direction.
The wind speed at the height of the fan hub and the wind direction at the height of the fan hub.
Air temperature, air pressure, relative humidity.
Optionally, the time step of the operation data is t, and the operation data includes:
station name, start time and forecast time.
Temperature, momentum flux, wind direction, wind speed and relative humidity at each altitude.
Sea level air pressure, cloud cover, latent heat flux, sensible heat flux, short wave radiation flux, long wave radiation flux, surface water pressure, total precipitation, large-scale precipitation and convection precipitation.
Optionally, the acquiring meteorological data and operational data includes:
and normalizing the collected meteorological data and the operation data, and cleaning invalid data.
Optionally, the obtaining an embedded vector includes:
and sliding the sliding window on the data, selecting meteorological data and operation data in the sliding window, and generating an embedded vector.
Optionally, the encoder includes a self-attention layer and a feedforward neural network, and the obtaining a feature map corresponding to the embedded vector according to the encoder includes:
the embedded vector is input from the attention layer to generate a query vector q, a key vector k, and a value vector v.
Generating a vector score according to the q and the k.
Generating a final score based on the score and the normalization parameters.
Normalizing the final score to generate a normalized score.
Calculating a weighted score vector from the v and normalized scores and calculating a sum of the weighted score vectors.
And inputting the sum of the weighted score vectors into the feed-forward neural network, and generating the feature map.
Optionally, the decoder comprises a self-attention layer, an encoding-decoding attention layer and a feedforward neural network.
According to a second aspect of the embodiments of the present application, there is provided a power prediction network training method, including:
a data set is generated from the meteorological data and the operational data.
The data set is labeled to generate a training data set.
And inputting the training data set into the power prediction network, and training by taking the minimization of the loss function as a target.
Optionally, the labeling the data set to generate a training data set includes:
and marking the actual power corresponding to the meteorological data and the operation data at each time point.
According to a third aspect of the embodiments of the present application, there is provided a wind power prediction apparatus based on a convolutional transformer architecture, including:
and the acquisition module is used for acquiring meteorological data and operation data and acquiring the embedded vector.
An input module to input the embedded vector into a power prediction network, the power prediction network comprising an encoder and a decoder.
And the characteristic extraction module is used for acquiring a characteristic diagram corresponding to the embedded vector according to the encoder.
And the prediction module is used for inputting the characteristic diagram into a decoder to generate predicted power.
Optionally, the time step of the meteorological data is t, and the meteorological data includes:
station rated capacity, power generation unit model, power generation unit quantity and capacity expansion information.
And actual power of the station output meter.
Wind height, wind speed and wind direction.
The wind speed at the height of the fan hub and the wind direction at the height of the fan hub.
Air temperature, air pressure, relative humidity.
Optionally, the time step of the operation data is t, and the operation data includes:
station name, start time and forecast time.
Temperature, momentum flux, wind direction, wind speed and relative humidity at each altitude.
Sea level air pressure, cloud cover, latent heat flux, sensible heat flux, short wave radiation flux, long wave radiation flux, surface water pressure, total precipitation, large-scale precipitation and convection precipitation.
Optionally, the acquisition module includes:
and the data cleaning submodule is used for normalizing the collected meteorological data and the operation data and cleaning invalid data.
Optionally, the acquisition module includes:
and the first vector generation submodule is used for enabling the sliding window to slide on the data, selecting meteorological data and operating data in the sliding window and generating an embedded vector.
Optionally, the encoder includes a self-attention layer and a feedforward neural network, and the feature extraction module includes:
a second vector generation submodule for inputting the embedded vector from the attention layer to generate a query vector q, a key vector k, and a value vector v.
A first scoring submodule to generate a vector score from the q and the k.
A second scoring submodule to generate a final score based on the score and the normalization parameter.
A third scoring submodule to normalize the final score to generate a normalized score.
And the fourth grading submodule calculates weighted grading vectors according to the v and the normalized grading and calculates the sum of the weighted grading vectors.
And the feature extraction submodule is used for inputting the sum of the weighted scoring vectors into the feedforward neural network and generating the feature map.
Optionally, the decoder comprises a self-attention layer, an encoding-decoding attention layer and a feedforward neural network.
According to a fourth aspect of the embodiments of the present application, there is provided a power prediction network training apparatus, including:
and the data acquisition module is used for generating a data set according to the meteorological data and the operation data.
And the marking module is used for marking the data set to generate a training data set.
And the training module is used for inputting the training data set into the power prediction network and training by taking the minimization of the loss function as a target.
Optionally, the labeling module includes:
and the marking submodule is used for marking the actual power corresponding to the meteorological data and the operation data at each time point.
According to a fifth aspect of the embodiments of the present application, there is provided a wind power prediction apparatus based on a convolutional transformer architecture, including:
a processor.
A memory for storing the processor-executable instructions;
wherein the processor is configured to execute the instructions to implement the wind power prediction method based on the convolutional transform architecture according to any of the first aspect.
According to a sixth aspect of embodiments of the present application, there is provided a non-transitory computer-readable storage medium, wherein instructions of the storage medium, when executed by a processor of a wind power prediction apparatus based on a convolutional transformer architecture, enable the wind power prediction apparatus based on the convolutional transformer architecture to perform the wind power prediction method based on the convolutional transformer architecture as described in any one of the above first aspects.
According to a seventh aspect of the embodiments of the present application, there is provided a power prediction network training apparatus, including:
a processor.
A memory for storing the processor-executable instructions;
wherein the processor is configured to execute the instructions to implement the power prediction network training method according to the second aspect.
According to an eighth aspect of embodiments of the present application, there is provided a non-transitory computer-readable storage medium, wherein instructions of the storage medium, when executed by a processor of a power prediction network training apparatus, enable the power prediction network training apparatus to perform the power prediction network training method according to the second aspect.
The technical scheme provided by the embodiment of the application at least has the following beneficial effects:
by paying attention to a plurality of time point data, the attention to local context information is enhanced, the influence of abnormal data on a prediction result is reduced, and the accuracy of power prediction is improved.
And when q and k are calculated, a convolution kernel is adopted to carry out convolution operation, so that attention is paid to local context, and more relevant features can be matched.
The improved power prediction network can be fitted more quickly, the prediction accuracy of the model can be improved in a complex data set, and lower training loss is achieved.
It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the application.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the application and are not to be construed as limiting the application.
FIG. 1 is a flow diagram illustrating a method of wind power prediction based on a convolutional transformer architecture, according to an exemplary embodiment.
FIG. 2 is a flow diagram illustrating a method of wind power prediction based on a convolutional transformer architecture, according to an exemplary embodiment.
FIG. 3 is a flow diagram illustrating a method of power prediction network training in accordance with an exemplary embodiment.
FIG. 4 is a block diagram illustrating a wind power prediction apparatus based on a convolutional transformer architecture, according to an exemplary embodiment.
FIG. 5 is a block diagram illustrating a wind power prediction apparatus based on a convolutional transformer architecture, according to an exemplary embodiment.
FIG. 6 is a block diagram illustrating a power prediction network training apparatus in accordance with an example embodiment.
Fig. 7 is a schematic diagram illustrating a power prediction network prediction flow according to an example embodiment.
Fig. 8 is a schematic diagram of an encoder structure shown in accordance with an exemplary embodiment.
Fig. 9 is a schematic diagram of a decoder structure shown in accordance with an example embodiment.
FIG. 10 is a block diagram illustrating a wind power prediction apparatus based on a convolutional transformer architecture, according to an exemplary embodiment.
Detailed Description
In order to make the technical solutions of the present application better understood by those of ordinary skill in the art, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.
It should be noted that the terms "first," "second," and the like in the description and claims of this application and in the drawings described above are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the application described herein are capable of operation in sequences other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present application, as detailed in the appended claims.
Wind power generation technology is becoming a major source for meeting future power demands. Higher share of renewable energy technology is crucial for carbon neutralization to meet the demands of future new power system grids, but also presents new grid operating challenges. The power company needs to predict the wind power generation power in order to perform the power generation scheduling operation. Prediction is a major driving factor to ensure safe and economical wind power grid connection, while establishing a link between many flexible innovations at different levels of the power system to achieve synergistic effects. Accurate wind power prediction is an important, cost-effective energy management element that also helps wind power plants and aggregate systems to efficiently and directly participate in the electricity market and to increase the efficiency of the plant by optimizing supply plans.
Most wind power predictions are based on wind turbine related data measured at regular time intervals by a time series analysis method. In the related technology, a recurrent neural network type model is adopted to analyze and predict a time sequence, but the recurrent neural network has the problems of gradient disappearance and gradient explosion when the network deepens. Even long-term and short-term memory networks still have great efforts to capture long-term dependencies. The subsequent development of a Transformer architecture has stronger long-term dependence modeling capability, and the effect on processing a longer-term sequence is obviously improved. The method based on the recurrent neural network can not completely eliminate the problems of gradient disappearance and gradient explosion when facing a long sequence, but the Transformer architecture can solve the problems, the effect is better on the long sequence, but the self-attention calculation method of the original Transformer architecture has the problem of insensitivity to local information, so that the model is easily influenced by abnormal points or abnormal data to cause prediction deviation.
In order to solve the above problems, the present application provides a method, an apparatus, and a storage medium for wind power prediction based on a convolutional transformer architecture.
Fig. 1 is a flowchart illustrating a wind power prediction method based on a convolutional transformer architecture according to an exemplary embodiment, where the method includes the following steps:
step 101, collecting meteorological data and operation data, and obtaining an embedded vector.
In the embodiment of the application, data needs to be collected to input the power prediction network. The power of wind power generation has two major influence factors: the running state of the wind generating set and the meteorological conditions around the wind generating set. The embodiment of the application collects the meteorological data and the operation data so as to predict the power of the wind generating set.
The meteorological data includes: the station name, the rated capacity, the type of the generating unit, the number of the generating units and the capacity expansion information of the wind power plant. And the station outbound meter in the station comprises time and actual power. The meteorological data comprises wind power data, and the wind power data comprises: wind speed, wind direction, air temperature, air pressure and relative humidity at a specified altitude. The designated height can be adjusted by an implementer according to actual conditions, and the designated height is not limited in the application. In one possible embodiment, the specified heights are 10 meters, 30 meters, 50 meters, 70 meters and fan hub height. The operation record comprises the starting time, the ending time and the corresponding maximum output upper limit value.
The operational data includes: the station name, the time of departure, the time of forecast, the wind speed, the wind direction, the temperature, the relative humidity at a given height, in a possible embodiment 10 meters, 30 meters, 70 meters, 100 meters. And simultaneously, sea level air pressure, cloud cover, latent heat flux, sensible heat flux, momentum flux, short wave radiation flux, long wave radiation flux, surface water pressure, total precipitation, large-scale precipitation and convection precipitation are required to be measured.
It should be noted that the meteorological data and the operation data are periodically collected, and the meteorological data and the operation data are collected once every time step t passes, and the specific value of t may be adjusted by an implementer according to an actual situation, and the application does not limit t. In a possible embodiment, the time step t is 15 minutes.
Will meteorological data and operational data constitute the time series data, and the implementation of this application predicts the wind energy power generation power of next time point according to the data that a plurality of time point were gathered, utilizes the sliding window to be in slide on the time series data and select the data on a plurality of continuous time point, for power prediction network discerns smoothly the time series sequence, the data generation that chooses according to the sliding window corresponds the embedding vector.
Step 102, inputting the embedded vector into a power prediction network, wherein the power prediction network comprises an encoder and a decoder.
In an embodiment of the present application, the power prediction network is a neural network of a convolutional migration transform architecture, and the power prediction network includes an encoder and a decoder.
And 103, acquiring a feature map corresponding to the embedded vector according to the encoder.
In this embodiment, the encoder includes a self-attention layer and a feedforward neural network, the embedded vector is input into the self-attention layer and converted into a query vector q, a key vector k and a value vector v, and then q, k and v are input into the feedforward neural network to extract features, so as to generate the feature map.
And 104, inputting the characteristic diagram into a decoder to generate predicted power.
In an embodiment of the present application, the decoder includes a self-attention layer, an encoding-decoding attention layer, and a feedforward neural network, and is configured to perform dimension reduction on the feature map to generate the predicted power.
Optionally, the time step of the meteorological data is t, and the meteorological data includes:
station rated capacity, power generation unit model, power generation unit quantity and capacity expansion information.
And actual power of the station output meter.
Wind height, wind speed and wind direction.
The wind speed at the height of the fan hub and the wind direction at the height of the fan hub.
Air temperature, air pressure, relative humidity.
In an embodiment of the present application, the meteorological data includes: the station name, the rated capacity, the type of the generating unit, the number of the generating units and the capacity expansion information of the wind power plant. And the station outbound meter in the station comprises time and actual power. The meteorological data comprises wind power data, and the wind power data comprises: wind speed, wind direction, air temperature, air pressure and relative humidity at a specified altitude. The designated height can be adjusted by an implementer according to actual conditions, and the designated height is not limited in the application. In one possible embodiment, the specified heights are 10 meters, 30 meters, 50 meters, 70 meters and fan hub height. The operation record comprises the starting time, the ending time and the corresponding maximum output upper limit value.
Optionally, the time step of the operation data is t, and the operation data includes:
station name, start time and forecast time.
Temperature, momentum flux, wind direction, wind speed and relative humidity at each altitude.
Sea level air pressure, cloud cover, latent heat flux, sensible heat flux, short wave radiation flux, long wave radiation flux, surface water pressure, total precipitation, large-scale precipitation and convection precipitation.
In an embodiment of the present application, the operation data includes: the station name, the time of departure, the time of forecast, the wind speed, the wind direction, the temperature, the relative humidity at a given height, in a possible embodiment 10 meters, 30 meters, 70 meters, 100 meters. And simultaneously, sea level air pressure, cloud cover, latent heat flux, sensible heat flux, momentum flux, short wave radiation flux, long wave radiation flux, surface water pressure, total precipitation, large-scale precipitation and convection precipitation are required to be measured.
Optionally, the acquiring meteorological data and operational data includes:
and normalizing the collected meteorological data and the operation data, and cleaning invalid data.
In the embodiment of the application, in order to reduce the error of the input power prediction network data, invalid operation data and meteorological data need to be cleared. And performing data cleaning on the operation data and the meteorological data, and deleting abnormal data. In one possible embodiment, data that is significantly different from the normal case is detected by setting a threshold range, or missing data and duplicate measurements are detected by searching for nulls. All detected errors and missing data are discarded from the initial data set. Meanwhile, in order to prevent gradient explosion, the cleaned data needs to be normalized. In one possible embodiment, the normalized formula is:
Figure BDA0003329755230000081
wherein x isnormIs normalized value, x is original value, xminIs the minimum value of the original values, xmaxIs the maximum value of the original numerical values.
Optionally, the obtaining an embedded vector includes:
and sliding the sliding window on the data, selecting meteorological data and operation data in the sliding window, and generating an embedded vector.
In the embodiment of the application, the meteorological data and the operation data form time sequence data, the wind power generation power of the next time point is predicted according to the data collected by a plurality of time points, a sliding window is utilized to slide on the time sequence data to select data on a plurality of continuous time points, the time sequence is smoothly identified for a power prediction network, and corresponding embedded vectors are generated according to the data selected by the sliding window.
Fig. 2 is a flowchart illustrating a wind power prediction method based on a convolutional transformer architecture, according to an exemplary embodiment, the encoder includes a self-attention layer and a feedforward neural network, as shown in fig. 2, the method includes the following steps:
step 201, the embedded vector is input from the attention layer to generate a query vector q, a key vector k and a value vector v.
In the embodiment of the application, the query vector q, the key vector k and the value vector v corresponding to the embedded vector are obtained from the attention layer to perform subsequent score calculation, so as to obtain the attention score.
Step 202, generating a vector score according to the q and the k.
In the embodiment of the present application, q and k are used to calculate a score of the embedding vector, where the calculation formula of score is: score is | qxk |, and is obtained by multiplying q and k.
Step 203, generate a final score based on the score and the normalization parameters.
In the present embodiment, in order to stabilize the gradient, it is necessary to normalize the score, i.e. divide the score by the normalization parameter
Figure BDA0003329755230000091
In a possible embodiment, dkIs the number of dimensions of the key vector k. In another possible embodiment, the score is 112, the number of k dimensions is 64, and the final score is 64
Figure BDA0003329755230000092
Step 204, normalizing the final score to generate a normalized score.
In the embodiment of the present application, the final score is normalized by using a normalization function. In one possible embodiment, the normalization function is a softmax function, and the final score is input to the softmax function to generate the normalized score. The normalization score represents the contribution of the embedded vector corresponding to the current time point to the prediction power, and the higher the normalization score is, the tighter the relationship between the data corresponding to the embedded vector and the prediction power is, the greater the contribution to the prediction power is. In one possible embodiment, the final score is 12, and a normalized score of 0.88 is output after normalization by the softmax function, which is used to subsequently weight z.
Step 205, calculating a weighted score vector based on the v and normalized scores and calculating the sum of the weighted score vectors.
In the embodiment of the application, the normalized scores and the v are multiplied to obtain weighted score vectors, and the weighted score vectors are added and collected to obtain the sum of the weighted score vectors.
Step 206, inputting the sum of the weighted score vectors into the feedforward neural network, and generating the feature map.
And inputting the sum of the weighted score vectors into the feed-forward neural network, and extracting features to generate the feature map.
Optionally, the decoder comprises a self-attention layer, an encoding-decoding attention layer and a feedforward neural network.
Fig. 8 is a schematic diagram of an encoder structure shown in accordance with an exemplary embodiment. As shown in fig. 8, the encoder includes a self-attention layer and a feed-forward neural network.
In the embodiment of the present application, fig. 9 is a schematic diagram illustrating a decoder structure according to an exemplary embodiment. As shown in fig. 9, the decoder also has a self-attention layer and a feed-forward neural network of the encoder. In addition, there is a coding-decoding layer (i.e., coding-decoding attention layer) between the two layers to focus on the relevant part of the input embedded vector. The coding-decoding attention layer is a fully-connected network, wherein two layers of networks exist, the activation function of the first layer is ReLU, and the formula expression of the ReLU activation function is
Figure BDA0003329755230000101
The model after sparse realization through the ReLU can better mine relevant characteristics and fit training data; the second layer is a linear activation function. The entire encoding-decoding attention layer may be summarized as an ffn (z) function: ffn (z) ═ max (0, ZW)1+b1)W2+b2。
Fig. 7 is a schematic diagram illustrating a power prediction network prediction flow according to an example embodiment. As shown in the figure, meteorological data and operation data at 4 time points are selected through a sliding window, corresponding query vectors q, key vectors k and value vectors v are generated according to a convolution kernel, the query vectors q, the key vectors k and the value vectors v are input into a self-attention layer in an encoder, attention correlation scores are calculated, and the sum of weighted score vectors is output.
FIG. 3 is a flow chart illustrating a method of power prediction network training, as shown in FIG. 3, according to an exemplary embodiment, the method comprising the steps of:
step 301, a data set is generated from the meteorological data and the operational data.
In the embodiment of the application, after the meteorological data and the operation data are collected by using various sensors, a data set can be constructed to train the power prediction network. The data set is a time series data set, and the time step of the meteorological data and the operational data is t, which in one possible embodiment is 15 minutes. In one possible embodiment, the data set is partitioned using different data partitioning methods, with the data set recorded within 2 years being partitioned into a training set and a test set. 10 different training sets are extracted from the original time series, and the evaluation dataset of the first year is divided into 10% training set, 30% training set, 50% training set and 70% training set sequentially or randomly.
Step 302, labeling the data set to generate a training data set.
In the embodiment of the application, the data in the data set are labeled, and the actual power of the wind power generation corresponding to the meteorological data and the operating data collected at each time point is labeled so as to train the power prediction network.
Step 303, inputting the training data set into the power prediction network, and training with a loss function minimization as a target.
In the embodiment of the application, the training data set is input into the power prediction network for iterative training, a sliding window is used for sliding on the training data set to select data at a plurality of continuous time points and input the data into the power prediction network, the predicted power is output, the predicted power is compared with the actual power, and a loss function is calculated. Optimizing parameters in the power prediction network with the minimization of the loss function as a target. And obtaining the recommended power prediction network after training.
Optionally, the labeling the data set to generate a training data set includes:
and marking the actual power corresponding to the meteorological data and the operation data at each time point.
FIG. 4 is a block diagram illustrating a wind power prediction apparatus based on a convolutional transformer architecture, according to an exemplary embodiment. Referring to fig. 4, the apparatus 400 includes an acquisition module 410, an input module 420, a feature extraction module 430, and a prediction module 440.
And the acquisition module 410 is used for acquiring meteorological data and operation data and acquiring the embedded vector.
An input module 420 for inputting the embedded vector into a power prediction network, the power prediction network comprising an encoder and a decoder.
And a feature extraction module 430, configured to obtain a feature map corresponding to the embedded vector according to the encoder.
A prediction module 440 for inputting the feature map into a decoder to generate a predicted power.
Optionally, the time step of the meteorological data is t, and the meteorological data includes:
station rated capacity, power generation unit model, power generation unit quantity and capacity expansion information.
And actual power of the station output meter.
Wind height, wind speed and wind direction.
The wind speed at the height of the fan hub and the wind direction at the height of the fan hub.
Air temperature, air pressure, relative humidity.
Optionally, the time step of the operation data is t, and the operation data includes:
station name, start time and forecast time.
Temperature, momentum flux, wind direction, wind speed and relative humidity at each altitude.
Sea level air pressure, cloud cover, latent heat flux, sensible heat flux, short wave radiation flux, long wave radiation flux, surface water pressure, total precipitation, large-scale precipitation and convection precipitation.
Optionally, the acquiring module 410 includes:
and the data cleaning submodule 411 is used for normalizing the collected meteorological data and the operation data and cleaning invalid data.
Optionally, the acquiring module 410 includes:
a first vector generation submodule 412 configured to slide the sliding window on the data, select the meteorological data and the operational data in the sliding window, and generate an embedded vector.
FIG. 5 is a block diagram illustrating a wind power prediction apparatus based on a convolutional transformer architecture, according to an exemplary embodiment. Referring to fig. 5, the apparatus 500 includes a second vector generation sub-module 510, a first scoring sub-module 520, a second scoring sub-module 530, a third scoring sub-module 540, a fourth scoring sub-module 550, and a feature extraction sub-module 560.
A second vector generation sub-module 510 for inputting the embedded vector from the attention layer to generate a query vector q, a key vector k, and a value vector v.
A first scoring submodule 520 for generating a vector score from said q and said k.
A second scoring submodule 530 generates a final score based on the score and the normalization parameters.
A third scoring submodule 540 normalizes the final score to generate a normalized score.
A fourth scoring submodule 550 calculates a weighted score vector from the v and normalized scores and calculates a sum of the weighted score vectors.
And the feature extraction sub-module 560 is used for inputting the sum of the weighted score vectors into the feed-forward neural network and generating the feature map.
Optionally, the decoder comprises a self-attention layer, an encoding-decoding attention layer and a feedforward neural network.
FIG. 6 is a block diagram illustrating a power prediction network training apparatus in accordance with an example embodiment. Referring to fig. 6, the apparatus 600 includes a data acquisition module 610, a labeling module 620, and a training module 630.
And a data acquisition module 610 for generating a data set from the meteorological data and the operational data.
A labeling module 620, configured to label the data set to generate a training data set.
A training module 630, configured to input the training data set into the power prediction network, and train with a loss function minimization as a target.
Optionally, the labeling module 620 includes:
and the labeling submodule 621 is used for labeling the actual power corresponding to the meteorological data and the operating data at each time point.
With regard to the apparatus in the above-described embodiment, the specific manner in which each module performs the operation has been described in detail in the embodiment related to the method, and will not be elaborated here.
Fig. 10 is a block diagram illustrating an apparatus 1000 for implementing the wind power prediction method based on the convolutional transformer architecture according to an exemplary embodiment.
In an exemplary embodiment, there is also provided a storage medium comprising instructions, such as the memory 1010 comprising instructions, the interface 1030, the instructions executable by the processor 1020 of the apparatus 1000 to perform the method described above. Alternatively, the storage medium may be a non-transitory computer readable storage medium, which may be, for example, a ROM, a Random Access Memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, and the like.
Other embodiments of the present application will be apparent to those skilled in the art from consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the invention following, in general, the principles of the application and including such departures from the present disclosure as come within known or customary practice within the art to which the invention pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the application being indicated by the following claims.
It will be understood that the present application is not limited to the precise arrangements described above and shown in the drawings and that various modifications and changes may be made without departing from the scope thereof. The scope of the application is limited only by the appended claims.

Claims (14)

1. A wind power prediction method based on a convolution transform architecture is characterized by comprising the following steps:
acquiring meteorological data and operating data, and acquiring an embedded vector;
inputting the embedded vector into a power prediction network, the power prediction network comprising an encoder and a decoder;
acquiring a feature map corresponding to the embedded vector according to the encoder;
inputting the feature map into a decoder to generate a predicted power.
2. The method of claim 1, wherein the meteorological data has a time step t, and comprises:
station rated capacity, power generation unit model, power generation unit number and capacity expansion information;
actual power of a station output meter;
wind height, wind speed and wind direction;
the wind speed at the height of the fan hub and the wind direction at the height of the fan hub are measured;
air temperature, air pressure, relative humidity.
3. The method of claim 1, wherein the operational data has a time step t, and wherein the operational data comprises:
station name, start time and forecast time;
temperature, momentum flux, wind direction, wind speed and relative humidity at each altitude;
sea level air pressure, cloud cover, latent heat flux, sensible heat flux, short wave radiation flux, long wave radiation flux, surface water pressure, total precipitation, large-scale precipitation and convection precipitation.
4. The method of claim 2 or 3, wherein said collecting meteorological data and operational data comprises:
and normalizing the collected meteorological data and the operation data, and cleaning invalid data.
5. The method of claim 1, wherein the obtaining the embedded vector comprises:
and sliding the sliding window on the data, selecting meteorological data and operation data in the sliding window, and generating an embedded vector.
6. The method of claim 1, wherein the encoder comprises a self-attention layer and a feedforward neural network, and the obtaining the feature map corresponding to the embedded vector according to the encoder comprises:
inputting the embedded vector into a self-attention layer to generate a query vector q, a key vector k and a value vector v;
generating a vector score according to the q and the k;
generating a final score according to the score and the normalization parameters;
normalizing the final score to generate a normalized score;
calculating a weighted score vector according to the v and the normalized scores and calculating the sum of the weighted score vectors;
and inputting the sum of the weighted score vectors into the feed-forward neural network, and generating the feature map.
7. The method of claim 1, wherein the decoder comprises a self-attention layer, an encoding-decoding attention layer, and a feed-forward neural network.
8. A power prediction network training method for training the power prediction network of any one of claims 1-7, comprising:
generating a data set according to the meteorological data and the operating data;
labeling the data set to generate a training data set;
and inputting the training data set into the power prediction network, and training by taking the minimization of the loss function as a target.
9. The method of claim 8, wherein labeling the data set to generate a training data set comprises:
and marking the actual power corresponding to the meteorological data and the operation data at each time point.
10. A wind power prediction device based on a convolution transformer architecture is characterized by comprising:
the acquisition module is used for acquiring meteorological data and operation data and acquiring an embedded vector;
an input module to input the embedded vector into a power prediction network, the power prediction network comprising an encoder and a decoder;
the characteristic extraction module is used for acquiring a characteristic diagram corresponding to the embedded vector according to the encoder;
and the prediction module is used for inputting the characteristic diagram into a decoder to generate predicted power.
11. A wind power prediction device based on a convolution transformer architecture is characterized by comprising:
a processor;
a memory for storing the processor-executable instructions;
wherein the processor is configured to execute the instructions to implement the convolution transform architecture based wind power prediction method of any of claims 1 to 7.
12. A non-transitory computer-readable storage medium, wherein instructions in the storage medium, when executed by a processor of a wind power prediction apparatus based on a convolutional fransformer architecture, enable the wind power prediction apparatus based on a convolutional fransformer architecture to perform the wind power prediction method based on a convolutional fransformer architecture according to any one of claims 1 to 7.
13. A power prediction network training apparatus, comprising:
a processor;
a memory for storing the processor-executable instructions;
wherein the processor is configured to execute the instructions to implement the power prediction network training method of claim 8 or 9.
14. A non-transitory computer-readable storage medium, wherein instructions in the storage medium, when executed by a processor of a power prediction network training apparatus, enable the power prediction network training apparatus to perform the power prediction network training method of claim 8 or 9.
CN202111274987.2A 2021-10-29 2021-10-29 Wind power prediction method, system and equipment based on convolution transform architecture Pending CN114021803A (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN202111274987.2A CN114021803A (en) 2021-10-29 2021-10-29 Wind power prediction method, system and equipment based on convolution transform architecture
PCT/CN2022/072596 WO2023070960A1 (en) 2021-10-29 2022-01-18 Wind power prediction method based on convolutional transformer architecture, and system and device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111274987.2A CN114021803A (en) 2021-10-29 2021-10-29 Wind power prediction method, system and equipment based on convolution transform architecture

Publications (1)

Publication Number Publication Date
CN114021803A true CN114021803A (en) 2022-02-08

Family

ID=80058831

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111274987.2A Pending CN114021803A (en) 2021-10-29 2021-10-29 Wind power prediction method, system and equipment based on convolution transform architecture

Country Status (2)

Country Link
CN (1) CN114021803A (en)
WO (1) WO2023070960A1 (en)

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117293916B (en) * 2023-08-22 2024-04-23 国网安徽省电力有限公司天长市供电公司 User-oriented power grid dispatching method and device and computing equipment
CN116930973B (en) * 2023-09-15 2023-12-08 湖南赛能环测科技有限公司 Scene-adaptive sodar wind power accurate prediction method
CN117132606B (en) * 2023-10-24 2024-01-09 四川大学 Segmentation method for lung lesion image
CN117111540B (en) * 2023-10-25 2023-12-29 南京德克威尔自动化有限公司 Environment monitoring and early warning method and system for IO remote control bus module
CN117559519A (en) * 2023-10-31 2024-02-13 北京瑞科同创能源科技有限公司 Wind power plant power data prediction method and device and electronic equipment
CN117674123B (en) * 2023-12-07 2024-06-14 上海交通大学 Wind power short-term power prediction method considering operation cost of power system
CN117933363B (en) * 2024-03-08 2024-06-11 广东工业大学 Method and system for predicting short-term wind power of newly-built wind power plant based on sample migration

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200034436A1 (en) * 2018-07-26 2020-01-30 Google Llc Machine translation using neural network models
CN112653142A (en) * 2020-12-18 2021-04-13 武汉大学 Wind power prediction method and system for optimizing depth transform network
CN112991090A (en) * 2021-02-05 2021-06-18 江南大学 Photovoltaic power prediction method based on Transformer model
CN113177633A (en) * 2021-04-20 2021-07-27 浙江大学 Deep decoupling time sequence prediction method

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2019216449A1 (en) * 2018-05-09 2019-11-14 주식회사 알고리고 Method and apparatus for time series artificial neural network electric vehicle power demand prediction, using spatio-temporal fusion of power demand data and heterogeneous data
CN113343590B (en) * 2021-07-08 2022-03-22 华北电力大学（保定） Wind speed prediction method and system based on combined model
CN113935534A (en) * 2021-10-20 2022-01-14 华能大理风力发电有限公司洱源分公司 Photovoltaic day-ahead prediction method, system and equipment based on convolution transform architecture

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200034436A1 (en) * 2018-07-26 2020-01-30 Google Llc Machine translation using neural network models
CN112653142A (en) * 2020-12-18 2021-04-13 武汉大学 Wind power prediction method and system for optimizing depth transform network
CN112991090A (en) * 2021-02-05 2021-06-18 江南大学 Photovoltaic power prediction method based on Transformer model
CN113177633A (en) * 2021-04-20 2021-07-27 浙江大学 Deep decoupling time sequence prediction method

Also Published As

Publication number Publication date
WO2023070960A1 (en) 2023-05-04

Similar Documents

Publication Publication Date Title
CN114021803A (en) Wind power prediction method, system and equipment based on convolution transform architecture
Cui et al. Wind power ramp event forecasting using a stochastic scenario generation method
Xu et al. A short-term wind power forecasting approach with adjustment of numerical weather prediction input by data mining
Zeng et al. Short-term wind power prediction using a wavelet support vector machine
CN110929963A (en) Wind speed prediction method, wind speed prediction device, and storage medium
Brusca et al. A new design methodology to predict wind farm energy production by means of a spiking neural network–based system
CN103117546B (en) A kind of Ultrashort-term slide prediction method for wind power
Zhang et al. Forecast of solar energy production-A deep learning approach
CN109165774A (en) A kind of short-term photovoltaic power prediction technique
Ayodele et al. A statistical analysis of wind distribution and wind power potential in the coastal region of South Africa
CN114202129A (en) Wind power output prediction method, electronic device, storage medium and system
Rao et al. Dropout and pruned neural networks for fault classification in photovoltaic arrays
TWI540533B (en) Method of short-term wind power generation forecasting
Li et al. Deep spatio-temporal wind power forecasting
Tümse et al. Estimation of wind turbine output power using soft computing models
CN115908051A (en) Method for determining energy storage capacity of power system
Kim et al. Wind power forecasting based on hourly wind speed data in South Korea using machine learning algorithms
CN109802634A (en) A kind of intelligent O&M method and operational system of the photovoltaic plant based on big data
CN117277304A (en) Photovoltaic power generation ultra-short-term power prediction method and system considering sunrise and sunset time
CN104834816A (en) Short-term wind speed prediction method
Kartini et al. Short term forecasting of global solar irradiance by k-nearest neighbor multilayer backpropagation learning neural network algorithm
Deshmukh et al. Application of genetic algorithm to neural network model for estimation of wind power potential
Parrado-Duque et al. A comparative analysis of machine learning methods for short-term load forecasting systems
Cheng et al. Probabilistic wind power generation model: Derivation and applications
Baruque et al. Small-wind turbine power generation prediction from atmospheric variables based on intelligent techniques

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination