CN117374967A - Offshore wind power output prediction method, device and storage medium - Google Patents
Offshore wind power output prediction method, device and storage medium Download PDFInfo
- Publication number
- CN117374967A CN117374967A CN202311517299.3A CN202311517299A CN117374967A CN 117374967 A CN117374967 A CN 117374967A CN 202311517299 A CN202311517299 A CN 202311517299A CN 117374967 A CN117374967 A CN 117374967A
- Authority
- CN
- China
- Prior art keywords
- data
- wind power
- offshore wind
- output prediction
- value
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 54
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 37
- 238000012545 processing Methods 0.000 claims abstract description 35
- 230000009471 action Effects 0.000 claims abstract description 28
- 238000000605 extraction Methods 0.000 claims abstract description 26
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 22
- 238000010248 power generation Methods 0.000 claims abstract description 19
- 230000006870 function Effects 0.000 claims description 39
- 238000004364 calculation method Methods 0.000 claims description 19
- 238000010606 normalization Methods 0.000 claims description 17
- 238000011176 pooling Methods 0.000 claims description 16
- 230000008569 process Effects 0.000 claims description 15
- 238000004590 computer program Methods 0.000 claims description 6
- 230000004913 activation Effects 0.000 claims description 4
- 238000013528 artificial neural network Methods 0.000 claims description 4
- 238000012544 monitoring process Methods 0.000 abstract description 8
- 230000007613 environmental effect Effects 0.000 abstract description 6
- 239000003795 chemical substances by application Substances 0.000 description 12
- 230000002787 reinforcement Effects 0.000 description 6
- 230000008901 benefit Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000009825 accumulation Methods 0.000 description 2
- 230000006399 behavior Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/004—Generation forecast, e.g. methods or systems for forecasting future energy generation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/10—Pre-processing; Data cleansing
- G06F18/15—Statistical pre-processing, e.g. techniques for normalisation or restoring missing data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- 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
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- 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
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/38—Arrangements for parallely feeding a single network by two or more generators, converters or transformers
- H02J3/381—Dispersed generators
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J2300/00—Systems for supplying or distributing electric power characterised by decentralized, dispersed, or local generation
- H02J2300/20—The dispersed energy generation being of renewable origin
- H02J2300/28—The renewable source being wind energy
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Business, Economics & Management (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Economics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Power Engineering (AREA)
- Bioinformatics & Computational Biology (AREA)
- Software Systems (AREA)
- Water Supply & Treatment (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Probability & Statistics with Applications (AREA)
- Public Health (AREA)
- Computational Linguistics (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Wind Motors (AREA)
Abstract
The invention discloses a method, a device and a storage medium for predicting offshore wind power output, wherein the method comprises the following steps: acquiring original data of an offshore wind farm according to an SCADA system; performing feature extraction on the original data by adopting a convolutional neural network to obtain feature data; and constructing an output prediction model by using an Actor-Critic framework, inputting characteristic data into the output prediction model, and calculating to obtain an offshore wind power output prediction value based on a double-delay depth deterministic strategy gradient algorithm, wherein the output prediction model comprises an Actor network and two Critic networks. According to the method, the influence of offshore environmental factors and wind turbine monitoring data on offshore wind power generation is comprehensively considered, the wind turbine parameters can be controlled according to the error output action of wind power processing, so that the power output of the wind turbine is controlled, and the accuracy of offshore wind power output prediction can be effectively improved.
Description
Technical Field
The invention relates to the technical field of wind power, in particular to a method and a device for predicting the output of offshore wind power and a storage medium.
Background
With the development of global economy and the growth of population, energy demand is increasing. However, fossil energy which can be developed and utilized is gradually exhausted, and the problem of energy shortage is increasingly serious. Under the background of energy crisis, wind energy is used as a new energy source which is pollution-free and can be continuously utilized, and has great development potential. The Chinese operators are wide, the coastline is 18000 km long, the offshore wind power resources are rich, and the method is particularly suitable for eastern coastal cities with obvious resource advantages. However, the conditions of offshore weather and hydrology are complex and changeable, and wind power output is influenced by factors such as offshore weather and wake flow. Meanwhile, the eastern coastal city belongs to a typical receiving end power grid, and has the characteristics of high power electronic degree of the power grid and small capacity of a local adjustable power supply. With large-scale randomness and volatility of offshore wind power access, the problems of unbalanced flexible supply and demand of a power grid, difficult real-time power balance and the like are increasingly prominent. Therefore, the method has practical significance in the aspects of accurately predicting wind power, dispatching the power grid, improving wind power access capability, reducing system operation cost and the like.
In the research of wind power prediction, a statistical prediction model is based on historical wind power output, numerical weather forecast and real-time meteorological data, and the wind power with short future time sequence is mapped through an artificial intelligence technology and a data driving method. Meanwhile, wind power output is obviously affected by the environment, compared with the onshore and offshore weather, the method is complex, so that the correlation between the environment data and the wind power output is extracted, and the accuracy of offshore wind power prediction can be effectively improved.
The existing offshore wind power output prediction method generally establishes a prediction model through machine learning by a computer, and the model establishes a prediction rule according to the relation between the historical characteristic data and the output data to realize offshore wind power processing prediction, but the existing offshore wind power output prediction method only can extract local information, is difficult to obtain the correlation of global characteristics, causes the loss of characteristic correlation information, and causes lower accuracy of offshore wind power output prediction.
Disclosure of Invention
The invention provides a method, a device and a storage medium for predicting the output of offshore wind power, which are used for solving the technical problems that the existing method for predicting the output of offshore wind power only can extract local information, is difficult to acquire the correlation of global features, causes the loss of feature correlation information, and causes lower accuracy of the output prediction of offshore wind power.
The embodiment of the invention provides a method for predicting the output of offshore wind power, which comprises the following steps:
acquiring original data of an offshore wind farm according to an SCADA system;
performing feature extraction on the original data by adopting a convolutional neural network to obtain feature data;
and constructing an output prediction model by using an Actor-Critic framework, inputting the characteristic data into the output prediction model, and calculating to obtain an offshore wind power output prediction value based on a double-delay depth deterministic strategy gradient algorithm, wherein the output prediction model comprises an Actor network and two Critic networks.
Further, before the characteristic extraction is performed on the original data by using the convolutional neural network to obtain the characteristic data, the method further includes:
and carrying out missing data filling processing, data normalization processing and data association processing on the original data.
Further, the performing missing data filling processing, data normalization processing and data association processing on the original data includes:
filling the missing values in the original data by adopting the spatial correlation between adjacent wind turbines;
normalizing the original data by adopting the following normalization formula;
wherein X is * For normalized data, X is the original data, X max X is the maximum value summarized by the original data min Is the minimum value in the original data;
analyzing the influence of each influence factor on wind power generation power by a pearson correlation coefficient method, wherein the calculation formula of the pearson correlation coefficient method is as follows:
wherein x and y are two data sequences, M is the length of the data sequence, x i And y i The data on day i in both sequences,and->Average in two sequences, cov (x, y) is the covariance of two data sequences, ver [ x ]]And Ver [ y ]]The variances of the data sequences x and y respectively, and the absolute value of the r (x, y) value represents the correlation degree of the two data sequences; the closer the correlation coefficient r (x, y) is to 1, the stronger the correlation between the current influencing factors and the wind power generation power is shown; when the correlation coefficient r (x, y) =0, the current influencing factor and the wind power generation are representedThe power has no linear dependence.
Further, the feature extraction of the original data by using the convolutional neural network to obtain feature data includes:
and carrying out feature extraction on the original data by utilizing a convolution layer and a pooling layer of the convolution neural network to obtain feature data.
Further, the calculation formula of the convolution layer is as follows:
A i,j ＝σ(∑∑W m,n *x t +b)
wherein A is i,j Outputting a convolution layer; * Is a convolution operator; sigma is an activation function; w (W) m,n And W is p The weight thresholds of the convolution process and the pooling process are respectively; b is weight bias;
the calculation formula of the pooling layer is as follows:
y out ＝W p *A i,j
wherein y is out Outputting for a pooling layer; x is x t Is an input vector;
x t ＝(P t ,D t ,H t ,…,S t )
wherein P is t Historical power at the moment t of the wind turbine generator; d (D) t The wind speed is the time t; h t The temperature is t time; s is S t And the state of the unit at the time t.
Further, the calculation of the offshore wind power output predicted value based on the dual-delay depth deterministic strategy gradient algorithm comprises the following steps:
updating a strategy network in the output prediction model through a deterministic strategy gradient;
determining an initial objective function of the output prediction model according to the dual-delay depth deterministic strategy gradient algorithm;
obtaining an updated objective function according to the network values of the two Critic networks and the initial objective function;
setting model parameters of the output prediction model;
and calculating the predicted value of the offshore wind power output corresponding to the characteristic data based on the model parameters and the updated objective function.
Further, the setting the model parameters of the output prediction model includes:
setting the error of the offshore wind power output predicted value and the offshore wind power output rated value as a state value;
setting control parameters of the fan set as action values;
and setting a negative value of the square of the difference value between the predicted power and the actual power of the wind turbine as a reward value, wherein the calculation formula of the reward value is as follows:
r＝-(P pred -P real ) 2 。
further, the calculating the offshore wind power output predicted value corresponding to the characteristic data includes:
and carrying out online learning and offline learning according to the characteristic data based on the output prediction model to obtain the offshore wind power output predicted value.
One embodiment of the present invention provides an offshore wind power output prediction apparatus, comprising:
the original data acquisition module is used for acquiring original data of the offshore wind farm according to the SCADA system;
the feature extraction module is used for carrying out feature extraction on the original data by adopting a convolutional neural network to obtain feature data;
the output prediction module is used for constructing an output prediction model by using an Actor-Critic framework, inputting the characteristic data into the output prediction model, and calculating to obtain an offshore wind power output prediction value based on a double-delay depth deterministic strategy gradient algorithm, wherein the output prediction model comprises an Actor network and two Critic networks.
An embodiment of the present invention provides a storage medium, where the storage medium includes a stored computer program, where the computer program when executed controls a device where the storage medium is located to perform an offshore wind power output prediction method as described above.
According to the embodiment of the invention, the influence of offshore environmental factors and wind turbine monitoring data on offshore wind power is comprehensively considered, the characteristic extraction is carried out on the acquired data by adopting a convolutional neural network, the offshore wind power output predicted value is calculated based on a double-delay depth deterministic strategy gradient algorithm, and the wind turbine parameters can be controlled according to the error output action of wind power processing, so that the power output of the wind turbine is controlled, and the accuracy of offshore wind power output prediction can be effectively improved.
Drawings
FIG. 1 is a schematic flow chart of a method for predicting the output of offshore wind power provided by an embodiment of the invention;
FIG. 2 is a schematic structural diagram of an offshore wind power output prediction system according to an embodiment of the present invention.
FIG. 3 is a schematic structural diagram of an offshore wind power output prediction device provided by an embodiment of the invention.
Detailed Description
The following description of the embodiments of the present application will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are only some, but not all, of the embodiments of the present application. All other embodiments, which can be made by one of ordinary skill in the art without undue burden from the present disclosure, are within the scope of the present disclosure.
In the description of the present application, it should be understood that the terms "first," "second," and the like are used for descriptive purposes only and are not to be construed as indicating or implying a relative importance or an implicit indication of the number of technical features being indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include one or more such feature. In the description of the present application, unless otherwise indicated, the meaning of "a plurality" is two or more.
In the description of the present application, it should be noted that, unless explicitly specified and limited otherwise, the terms "mounted," "connected," and "connected" are to be construed broadly, and may be either fixedly connected, detachably connected, or integrally connected, for example; can be mechanically or electrically connected; can be directly connected or indirectly connected through an intermediate medium, and can be communication between two elements. The specific meaning of the terms in this application will be understood by those of ordinary skill in the art in a specific context.
Referring to fig. 1, an embodiment of the present invention provides a method for predicting offshore wind power output, including:
s1, acquiring original data of an offshore wind farm according to an SCADA system;
in the embodiment of the invention, the SCADA (Supervisory Control And Data Acquisition) system is a data acquisition and monitoring control system. And massive raw data acquired by different sensors of each unit are recorded in the SCADA system.
According to the embodiment of the invention, the SCADA system can acquire the original data of the offshore wind farm, including wind speed data, wind direction data, wind turbine generator system power data and the like.
S2, performing feature extraction on the original data by adopting a convolutional neural network to obtain feature data;
s3, constructing an output prediction model by using an Actor-Critic framework, inputting characteristic data into the output prediction model, and calculating to obtain an offshore wind power output prediction value based on a double-delay depth deterministic strategy gradient algorithm, wherein the output prediction model comprises an Actor network and two Critic networks.
According to the embodiment of the invention, the influence of offshore environmental factors and wind turbine monitoring data on offshore wind power is comprehensively considered, the characteristic extraction is carried out on the acquired data by adopting a convolutional neural network, the offshore wind power output predicted value is calculated based on a double-delay depth deterministic strategy gradient algorithm, and the wind turbine parameters can be controlled according to the error output action of wind power processing, so that the power output of the wind turbine is controlled, and the accuracy of offshore wind power output prediction can be effectively improved.
In one embodiment, step S2, before performing feature extraction on the original data by using the convolutional neural network, further includes:
s201, performing missing data filling processing, data normalization processing and data association processing on the original data.
In the embodiment of the invention, the acquired SCADA data cannot be directly applied to the output prediction of the offshore wind farm, and the acquired data needs to be preprocessed, so that the error of the original data is reduced.
In one embodiment, S201, performing missing data padding processing, data normalization processing, and data association processing on the original data, includes:
s2011, filling missing values in the original data by adopting spatial correlation among adjacent wind turbines;
in the embodiment of the invention, the problem of data loss of the original data acquired by the SCADA system is mainly caused by sensor faults and network congestion, and the space correlation among adjacent wind turbines is adopted to fill the target wind turbine missing value, so that the data integrity can be effectively ensured.
S2012, carrying out normalization processing on the original data by adopting the following normalization formula;
wherein X is * For normalized data, X is the original data, X max For the maximum value of the summary of the original data, X min Is the minimum value in the original data;
in the embodiment of the invention, in all the characteristics, the wind direction reflects the projection coefficient of the wind speed in a specific direction, and the wind direction is normalized by adopting a sine function according to the physical meaning of the projection coefficient. Meanwhile, the pitch angle is used as an angle characteristic, and a sine function is also adopted for normalization calculation.
S2013, analyzing the influence of each influence factor on wind power generation power by a Pearson correlation coefficient method, wherein a calculation formula of the Pearson correlation coefficient method is as follows:
wherein x and y are two data sequences, M is the length of the data sequence, x i And y i The data on day i in both sequences,and->Average in two sequences, cov (x, y) is the covariance of two data sequences, ver [ x ]]And Ver [ y ]]The variances of the data sequences x and y respectively, and the absolute value of the r (x, y) value represents the correlation degree of the two data sequences; the closer the correlation coefficient r (x, y) is to 1, the stronger the correlation between the current influencing factors and the wind power generation power is shown; when the correlation coefficient r (x, y) =0, it indicates that the current influencing factor has no linear correlation with the wind power generation power.
In the embodiment of the invention, the wind power generation is commonly influenced by a plurality of factors such as wind speed, wind direction, temperature, air pressure, air density and the like, and meanwhile, the wind power generation is also related to the states of wind power generators such as a gear box system, a generator system, a yaw system and the like, so that the influence of each factor on the wind power generation power is analyzed by a Pearson correlation coefficient method.
In one embodiment, step S2, performing feature extraction on the original data by using a convolutional neural network to obtain feature data, including:
and S21, performing feature extraction on the original data by using a convolution layer and a pooling layer of the convolution neural network to obtain feature data.
In one embodiment, the calculation formula for the convolution layer is as follows:
A i,j ＝σ(∑∑W m,n *x t +b)
wherein A is i,j Outputting a convolution layer; * Is a convolution operator; sigma is an activation function; w (W) m,n And W is p The weight thresholds of the convolution process and the pooling process are respectively; b is weight bias;
the calculation formula of the pooling layer is as follows:
y out ＝W p *A i,j
wherein y is out Outputting for a pooling layer; x is x t Is an input vector;
x t ＝(P t ,D t ,H t ,…,S t )
wherein P is t Historical power at the moment t of the wind turbine generator; d (D) t The wind speed is the time t; h t The temperature is t time; s is S t And the state of the unit at the time t.
In the embodiment of the invention, the output power of the wind turbine is influenced by external environment factors such as wind speed, wind direction and the like, and is also influenced by the running state of the wind turbine. The convolutional neural network CNN is introduced to extract the characteristics of the input data, so that the influence of the input data on the prediction accuracy can be effectively reduced, and the prediction accuracy of the offshore wind power output can be effectively improved.
In one embodiment, step S3, calculating an offshore wind power output predicted value based on a dual-delay depth deterministic strategy gradient algorithm includes:
s31, updating a strategy network in the output prediction model through a deterministic strategy gradient;
in an embodiment of the present invention, the objective of reinforcement learning is to seek an optimal policy to maximize the expected return value, and under the Actor-Critic framework, the policy network (Actor) updates the network through deterministic policy gradients (Deterministic Policy Gradient, DPG), as follows:
wherein,is gradient operation; pi is a policy function; phi is a network parameter of the policy function; />Is the expected value; s is the state; a is action; q (Q) π (s, a) is a state-cost function representing the expected return value after action a is taken in state s, if the policy is followed.
The state-cost function uses the advantage of the algorithm's actions on continuous space to change the randomness strategy to a deterministic strategy, as shown in the following equation:
a t :π θ (s t |θ π )
wherein a is t The action value is the t moment; pi θ (s t |θ π ) Is a strategy of an intelligent agent.
S32, determining an initial objective function of the output prediction model according to a dual-delay depth deterministic strategy gradient algorithm;
in the embodiment of the invention, the dual-delay depth deterministic strategy gradient algorithm TD3 is a depth reinforcement learning algorithm of an Actor-Critic framework, and is developed on the basis of a depth deterministic strategy gradient algorithm (deep deterministic policy gradient, DDPG).
The TD3 algorithm selects the optimal action using the Actor network and evaluates the policy through the target Critic network as shown in the following equation:
y t ＝r(s t ,a t )+γQ θ (s t+1 ,π φ (s t+1 ))
wherein y is t As a function of the target value; r(s) t ,a t ) Is a prize value; gamma is the discount rate; q (Q) θ (s t+1 ,π φ (s t+1 ) Is a state s) t The objective cost function below.
S33, obtaining an updated objective function according to the network values of the two Critic networks and the initial objective function;
in the embodiment of the invention, the TD3 algorithm learns by adopting two Q value functions, and a smaller Critic network value is selected as an updating target during updating:
wherein,to reduce training costs, the invention is implemented as a smaller target value functionThe TD3 algorithm of the example uses a separate Actor network and two Critic networks.
In the embodiment of the invention, the TD3 algorithm adopts a strategy delay updating mechanism, the updating iteration frequency of the Actor network is lower than that of the Critic network, and the Actor network is ensured to update under the condition of lower Q value error, so that the updating efficiency of the Actor network can be effectively improved.
In the deep reinforcement learning process, under the condition that the error of the Critic network is large, the strategy is enabled to have discrete behaviors by updating the Actor network, and the TD3 algorithm updates the Actor network once after updating the Critic network for a certain number of times. Meanwhile, in order to ensure that the agent explores the global optimal strategy and avoids the local optimal strategy, noise epsilon needs to be added into the action selection strategy to improve the exploration degree.
S34, setting model parameters of an output prediction model;
in an embodiment of the invention, the model parameters include a status value, an action value, and a reward value.
And S35, calculating the predicted value of the offshore wind power output corresponding to the characteristic data based on the model parameters and the updated objective function.
In one embodiment, step S34, setting model parameters of the output prediction model includes:
s341, setting an error between the offshore wind power output predicted value and the offshore wind power output rated value as a state value;
s342, setting control parameters of the fan set as action values;
s343, setting a negative value of the square of the difference value between the predicted power and the actual power of the wind turbine as a reward value, wherein the calculation formula of the reward value is as follows:
r＝-(P pred -P real ) 2 。
in one embodiment, step S35, calculating a predicted value of the offshore wind power output corresponding to the characteristic data includes:
and carrying out online learning and offline learning according to the characteristic data based on the output prediction model to obtain the offshore wind power output predicted value.
In the embodiment of the invention, an Agent (Agent) interacts with an offshore wind farm Environment (Environment) in offline learning, and learns how to control parameter settings of wind turbines and how to control power generated by the wind turbines in different wind turbine states and marine weather environments. In offline learning, firstly, random action accumulation experience is carried out, after learning a part of experience, the Agent observes the environment state output action, and meanwhile, certain exploration noise is added for further optimization.
In the online learning process, an Agent observes the field environment and guides the wind turbine generator to act. In the offline learning and online learning, learned experiences are stored in an experience playback pool, and what Critic network of an Actor is updated to realize closed-loop control.
Referring to fig. 3, a schematic structural diagram of an offshore wind power output prediction system according to an embodiment of the invention is shown.
The embodiment of the invention has the following beneficial effects:
according to the embodiment of the invention, the influence of offshore environmental factors and wind turbine monitoring data on offshore wind power is comprehensively considered, the characteristic extraction is carried out on the acquired data by adopting a convolutional neural network, the offshore wind power output predicted value is calculated based on a double-delay depth deterministic strategy gradient algorithm, and the wind turbine parameters can be controlled according to the error output action of wind power processing, so that the power output of the wind turbine is controlled, and the accuracy of offshore wind power output prediction can be effectively improved.
Furthermore, the embodiment of the invention can effectively ensure the comprehensiveness of the data and accurately determine the factors related to the output of the offshore wind farm by carrying out missing data filling processing, data normalization processing and data association processing on the original data, thereby further improving the accuracy of the output of the offshore wind farm.
Referring to fig. 3, based on the same inventive concept as the above embodiment, an embodiment of the present invention provides an offshore wind power output prediction apparatus, including:
the original data acquisition module 10 is used for acquiring original data of the offshore wind farm according to the SCADA system;
in the embodiment of the invention, the SCADA (Supervisory Control And Data Acquisition) system is a data acquisition and monitoring control system. And massive raw data acquired by different sensors of each unit are recorded in the SCADA system.
According to the embodiment of the invention, the SCADA system can acquire the original data of the offshore wind farm, including wind speed data, wind direction data, wind turbine generator system power data and the like.
The feature extraction module 20 is configured to perform feature extraction on the original data by using a convolutional neural network to obtain feature data;
the output prediction module 30 is configured to construct an output prediction model with an Actor-Critic framework, input feature data into the output prediction model, calculate and obtain an offshore wind power output prediction value based on a dual-delay depth deterministic strategy gradient algorithm, and the output prediction model comprises an Actor network and two Critic networks.
According to the embodiment of the invention, the influence of offshore environmental factors and wind turbine monitoring data on offshore wind power is comprehensively considered, the characteristic extraction is carried out on the acquired data by adopting a convolutional neural network, the offshore wind power output predicted value is calculated based on a double-delay depth deterministic strategy gradient algorithm, and the wind turbine parameters can be controlled according to the error output action of wind power processing, so that the power output of the wind turbine is controlled, and the accuracy of offshore wind power output prediction can be effectively improved.
In one embodiment, the apparatus further comprises a preprocessing module for:
and carrying out missing data filling processing, data normalization processing and data association processing on the original data.
In the embodiment of the invention, the acquired SCADA data cannot be directly applied to the output prediction of the offshore wind farm, and the acquired data needs to be preprocessed, so that the error of the original data is reduced.
In one embodiment, the missing data padding process, the data normalization process and the data association process are performed on the original data, including:
filling the missing values in the original data by adopting the spatial correlation between adjacent wind turbines;
in the embodiment of the invention, the problem of data loss of the original data acquired by the SCADA system is mainly caused by sensor faults and network congestion, and the space correlation among adjacent wind turbines is adopted to fill the target wind turbine missing value, so that the data integrity can be effectively ensured.
Normalizing the original data by adopting the following normalization formula;
wherein X is * For normalized data, X is the original data, X max For the maximum value of the summary of the original data, X min Is the minimum value in the original data;
in the embodiment of the invention, in all the characteristics, the wind direction reflects the projection coefficient of the wind speed in a specific direction, and the wind direction is normalized by adopting a sine function according to the physical meaning of the projection coefficient. Meanwhile, the pitch angle is used as an angle characteristic, and a sine function is also adopted for normalization calculation.
The influence of each influence factor on wind power generation power is analyzed through a pearson correlation coefficient method, and the calculation formula of the pearson correlation coefficient method is as follows:
wherein x and y are two data sequences, M is the length of the data sequence, x i And y i The data on day i in both sequences,and->Average in two sequences, cov (x, y) is the covariance of two data sequences, ver [ x ]]And Ver [ y ]]Respectively data sequences x andthe variance of y, the absolute value of the r (x, y) value represents the degree of correlation of two data sequences; the closer the correlation coefficient r (x, y) is to 1, the stronger the correlation between the current influencing factors and the wind power generation power is shown; when the correlation coefficient r (x, y) =0, it indicates that the current influencing factor has no linear correlation with the wind power generation power.
In the embodiment of the invention, the wind power generation is commonly influenced by a plurality of factors such as wind speed, wind direction, temperature, air pressure, air density and the like, and meanwhile, the wind power generation is also related to the states of wind power generators such as a gear box system, a generator system, a yaw system and the like, so that the influence of each factor on the wind power generation power is analyzed by a Pearson correlation coefficient method.
In one embodiment, the feature extraction module 20 is further configured to:
and carrying out feature extraction on the original data by utilizing a convolution layer and a pooling layer of the convolution neural network to obtain feature data.
In one embodiment, the calculation formula for the convolution layer is as follows:
A i,j ＝σ(∑∑W m,n *x t +b)
wherein A is i,j Outputting a convolution layer; * Is a convolution operator; sigma is an activation function; w (W) m,n And W is p The weight thresholds of the convolution process and the pooling process are respectively; b is weight bias;
the calculation formula of the pooling layer is as follows:
y out ＝W p *A i,j
wherein y is out Outputting for a pooling layer; x is x t Is an input vector;
x t ＝(P t ,D t ,H t ,…,S t )
wherein P is t Historical power at the moment t of the wind turbine generator; d (D) t The wind speed is the time t; h t The temperature is t time; s is S t And the state of the unit at the time t.
In the embodiment of the invention, the output power of the wind turbine is influenced by external environment factors such as wind speed, wind direction and the like, and is also influenced by the running state of the wind turbine. The convolutional neural network CNN is introduced to extract the characteristics of the input data, so that the influence of the input data on the prediction accuracy can be effectively reduced, and the prediction accuracy of the offshore wind power output can be effectively improved.
In one embodiment, the output prediction module 30 is further configured to:
updating a strategy network in the output prediction model through a deterministic strategy gradient;
in an embodiment of the present invention, the objective of reinforcement learning is to seek an optimal policy to maximize the expected return value, and under the Actor-Critic framework, the policy network (Actor) updates the network through deterministic policy gradients (Deterministic Policy Gradient, DPG), as follows:
wherein,is gradient operation; pi is a policy function; phi is a network parameter of the policy function; />Is the expected value; s is the state; a is action; q (Q) π (s, a) is a state-cost function representing the expected return value after action a is taken in state s, if the policy is followed.
The state-cost function uses the advantage of the algorithm's actions on continuous space to change the randomness strategy to a deterministic strategy, as shown in the following equation:
a t :π θ (s t |θ π )
wherein a is t The action value is the t moment; pi θ (s t |θ π ) Is a strategy of an intelligent agent.
Determining an initial objective function of the output prediction model according to a dual-delay depth deterministic strategy gradient algorithm;
in the embodiment of the invention, the dual-delay depth deterministic strategy gradient algorithm TD3 is a depth reinforcement learning algorithm of an Actor-Critic framework, and is developed on the basis of a depth deterministic strategy gradient algorithm (deep deterministic policy gradient, DDPG).
The TD3 algorithm selects the optimal action using the Actor network and evaluates the policy through the target Critic network as shown in the following equation:
y t ＝r(s t ,a t )+γQ θ (s t+1 ,π φ (s t+1 ))
wherein y is t As a function of the target value; r(s) t ,a t ) Is a prize value; gamma is the discount rate; q (Q) θ (s t+1 ,π φ (s t+1 ) Is a state s) t The objective cost function below.
Obtaining an updated objective function according to the network values of the two Critic networks and the initial objective function;
in the embodiment of the invention, the TD3 algorithm learns by adopting two Q value functions, and a smaller Critic network value is selected as an updating target during updating:
wherein,for smaller target value functions, the TD3 algorithm of the embodiment of the invention adopts an independent Actor network and two Critic networks in order to reduce training cost.
In the embodiment of the invention, the TD3 algorithm adopts a strategy delay updating mechanism, the updating iteration frequency of the Actor network is lower than that of the Critic network, and the Actor network is ensured to update under the condition of lower Q value error, so that the updating efficiency of the Actor network can be effectively improved.
In the deep reinforcement learning process, under the condition that the error of the Critic network is large, the strategy is enabled to have discrete behaviors by updating the Actor network, and the TD3 algorithm updates the Actor network once after updating the Critic network for a certain number of times. Meanwhile, in order to ensure that the agent explores the global optimal strategy and avoids the local optimal strategy, noise epsilon needs to be added into the action selection strategy to improve the exploration degree.
Setting model parameters of an output prediction model;
in an embodiment of the invention, the model parameters include a status value, an action value, and a reward value.
And calculating the predicted value of the offshore wind power output corresponding to the characteristic data based on the model parameters and the updated objective function.
In one embodiment, setting model parameters of the output prediction model includes:
setting the error of the offshore wind power output predicted value and the offshore wind power output rated value as a state value;
setting control parameters of the fan set as action values;
and setting a negative value of the square of the difference value between the predicted power and the actual power of the wind turbine as a reward value, wherein the calculation formula of the reward value is as follows:
r＝-(P pred -P real ) 2 。
in one embodiment, calculating the offshore wind power output predicted value corresponding to the characteristic data includes:
and carrying out online learning and offline learning according to the characteristic data based on the output prediction model to obtain the offshore wind power output predicted value.
In the embodiment of the invention, an Agent (Agent) interacts with an offshore wind farm Environment (Environment) in offline learning, and learns how to control parameter settings of wind turbines and how to control power generated by the wind turbines in different wind turbine states and marine weather environments. In offline learning, firstly, random action accumulation experience is carried out, after learning a part of experience, the Agent observes the environment state output action, and meanwhile, certain exploration noise is added for further optimization.
In the online learning process, an Agent observes the field environment and guides the wind turbine generator to act. In the offline learning and online learning, learned experiences are stored in an experience playback pool, and what Critic network of an Actor is updated to realize closed-loop control.
The embodiment of the invention has the following beneficial effects:
according to the embodiment of the invention, the influence of offshore environmental factors and wind turbine monitoring data on offshore wind power is comprehensively considered, the characteristic extraction is carried out on the acquired data by adopting a convolutional neural network, the offshore wind power output predicted value is calculated based on a double-delay depth deterministic strategy gradient algorithm, and the wind turbine parameters can be controlled according to the error output action of wind power processing, so that the power output of the wind turbine is controlled, and the accuracy of offshore wind power output prediction can be effectively improved.
Furthermore, the embodiment of the invention can effectively ensure the comprehensiveness of the data and accurately determine the factors related to the output of the offshore wind farm by carrying out missing data filling processing, data normalization processing and data association processing on the original data, thereby further improving the accuracy of the output of the offshore wind farm.
An embodiment of the present invention provides a storage medium, which includes a stored computer program, where the computer program when executed controls a device in which the storage medium is located to perform an offshore wind power output prediction method as described above.
The foregoing is a preferred embodiment of the present invention and it should be noted that modifications and adaptations to those skilled in the art may be made without departing from the principles of the present invention and are intended to be comprehended within the scope of the present invention.
Claims (10)
1. The marine wind power output prediction method is characterized by comprising the following steps of:
acquiring original data of an offshore wind farm according to an SCADA system;
performing feature extraction on the original data by adopting a convolutional neural network to obtain feature data;
and constructing an output prediction model by using an Actor-Critic framework, inputting the characteristic data into the output prediction model, and calculating to obtain an offshore wind power output prediction value based on a double-delay depth deterministic strategy gradient algorithm, wherein the output prediction model comprises an Actor network and two Critic networks.
2. The offshore wind power output prediction method of claim 1, further comprising, before the characteristic extraction is performed on the raw data by using a convolutional neural network to obtain characteristic data:
and carrying out missing data filling processing, data normalization processing and data association processing on the original data.
3. The offshore wind power output prediction method of claim 2, wherein the performing missing data filling processing, data normalization processing and data association processing on the raw data comprises:
filling the missing values in the original data by adopting the spatial correlation between adjacent wind turbines;
normalizing the original data by adopting the following normalization formula;
wherein X is * For normalized data, X is the original data, X max X is the maximum value summarized by the original data min Is the minimum value in the original data;
analyzing the influence of each influence factor on wind power generation power by a pearson correlation coefficient method, wherein the calculation formula of the pearson correlation coefficient method is as follows:
wherein x and y are two data sequences, M is the length of the data sequence, x i And y i The data on day i in both sequences,and->Average in two sequences, cov (x, y) is the covariance of two data sequences, ver [ x ]]And Ver [ y ]]The variances of the data sequences x and y respectively, and the absolute value of the r (x, y) value represents the correlation degree of the two data sequences; the closer the correlation coefficient r (x, y) is to 1, the stronger the correlation between the current influencing factors and the wind power generation power is shown; when the correlation coefficient r (x, y) =0, it indicates that the current influencing factor has no linear correlation with the wind power generation power.
4. The method for predicting the output of offshore wind power according to claim 1, wherein the step of performing feature extraction on the raw data by using a convolutional neural network to obtain feature data comprises the steps of:
and carrying out feature extraction on the original data by utilizing a convolution layer and a pooling layer of the convolution neural network to obtain feature data.
5. The offshore wind turbine treatment prediction method of claim 4, wherein the convolution layer has a calculation formula as follows:
A i,j ＝σ(∑∑W m,n *x t +b)
wherein A is i,j Outputting a convolution layer; * Is a convolution operator; sigma is an activation function; w (W) m,n And W is p The weight thresholds of the convolution process and the pooling process are respectively; b is weight bias;
the calculation formula of the pooling layer is as follows:
y out ＝W p *A i,j
wherein y is out Outputting for a pooling layer; x is x t Is an input vector;
x t ＝(P t ,D t ,H t ,···,S t )
wherein P is t Historical power at the moment t of the wind turbine generator; d (D) t The wind speed is the time t; h t The temperature is t time; s is S t And the state of the unit at the time t.
6. The method for predicting the output of the offshore wind power according to claim 1, wherein the step of calculating the predicted output value of the offshore wind power based on the dual-delay depth deterministic strategy gradient algorithm comprises the following steps:
updating a strategy network in the output prediction model through a deterministic strategy gradient;
determining an initial objective function of the output prediction model according to the dual-delay depth deterministic strategy gradient algorithm;
obtaining an updated objective function according to the network values of the two Critic networks and the initial objective function;
setting model parameters of the output prediction model;
and calculating the predicted value of the offshore wind power output corresponding to the characteristic data based on the model parameters and the updated objective function.
7. The offshore wind power output prediction method of claim 6, wherein the setting model parameters of the output prediction model comprises:
setting the error of the offshore wind power output predicted value and the offshore wind power output rated value as a state value;
setting control parameters of the fan set as action values;
and setting a negative value of the square of the difference value between the predicted power and the actual power of the wind turbine as a reward value, wherein the calculation formula of the reward value is as follows:
r＝-(P pred -P real ) 2 。
8. the method for predicting the output of offshore wind power according to claim 7, wherein the calculating the predicted value of the output of offshore wind power corresponding to the characteristic data comprises:
and carrying out online learning and offline learning according to the characteristic data based on the output prediction model to obtain the offshore wind power output predicted value.
9. An offshore wind power output prediction device, comprising:
the original data acquisition module is used for acquiring original data of the offshore wind farm according to the SCADA system;
the feature extraction module is used for carrying out feature extraction on the original data by adopting a convolutional neural network to obtain feature data;
the output prediction module is used for constructing an output prediction model by using an Actor-Critic framework, inputting the characteristic data into the output prediction model, and calculating to obtain an offshore wind power output prediction value based on a double-delay depth deterministic strategy gradient algorithm, wherein the output prediction model comprises an Actor network and two Critic networks.
10. A storage medium comprising a stored computer program, wherein the computer program, when run, controls a device in which the storage medium is located to perform an offshore wind power output prediction method according to any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311517299.3A CN117374967A (en) | 2023-11-14 | 2023-11-14 | Offshore wind power output prediction method, device and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311517299.3A CN117374967A (en) | 2023-11-14 | 2023-11-14 | Offshore wind power output prediction method, device and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117374967A true CN117374967A (en) | 2024-01-09 |
Family
ID=89391073
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311517299.3A Pending CN117374967A (en) | 2023-11-14 | 2023-11-14 | Offshore wind power output prediction method, device and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117374967A (en) |
-
2023
- 2023-11-14 CN CN202311517299.3A patent/CN117374967A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2019214785A1 (en) | Wind turbine control method | |
| CN111144663B (en) | Ultra-short-term wind power prediction method for offshore wind farm considering output fluctuation process | |
| CN114362175B (en) | Wind power prediction method and system based on depth certainty strategy gradient algorithm | |
| CN114696351A (en) | Dynamic optimization method and device for battery energy storage system, electronic equipment and storage medium | |
| CN116933619A (en) | Digital twin distribution network fault scene generation method and system based on reinforcement learning | |
| CN105631550A (en) | Short-term wind power combined prediction method | |
| CN114784823A (en) | Micro-grid frequency control method and system based on depth certainty strategy gradient | |
| CN117039981A (en) | Large-scale power grid optimal scheduling method, device and storage medium for new energy | |
| CN113988481A (en) | Wind power prediction method based on dynamic matrix prediction control | |
| CN115330040A (en) | Deep learning-based comprehensive energy distributed wind power generation prediction method and system | |
| Li et al. | A closed-loop maintenance strategy for offshore wind farms: Incorporating dynamic wind farm states and uncertainty-awareness in decision-making | |
| CN117374967A (en) | Offshore wind power output prediction method, device and storage medium | |
| Du et al. | Measurement-driven source tracing of torsional subsynchronous oscillations caused by open-loop modal resonance | |
| CN115965134A (en) | Regional power grid wind power generation power prediction optimization method | |
| CN109872248A (en) | A kind of wind power plant cluster output calculation method and system | |
| CN115395502A (en) | Photovoltaic power station power prediction method and system | |
| CN114049014A (en) | Method, device and system for evaluating operation state of offshore wind turbine generator | |
| Yang et al. | Offshore Wind Power Prediction Based on Variational Mode Decomposition and Long Short Term Memory Networks | |
| Duan et al. | Study on performance evaluation and prediction of mixed-flow hydraulic turbine units under variable operating conditions | |
| Liu et al. | Ultra-short-term wind power forecasting based on stacking model | |
| CN117175595B (en) | Power grid regulation and control method and system based on multi-level data | |
| Zhao et al. | Cooperative Wind Farm Control with Hybrid-Model-Based Deep Deterministic Policy Gradient and Model Selection | |
| CN117117989A (en) | Deep reinforcement learning solving method for unit combination | |
| CN117974395A (en) | Environment-friendly community management method and system based on carbon emission monitoring | |
| CN116681147A (en) | Electric energy demand prediction method and related equipment |
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 |