CN111784535A - Wind energy direct interval prediction method and device based on feedback neural network - Google Patents

Wind energy direct interval prediction method and device based on feedback neural network Download PDF

Info

Publication number
CN111784535A
CN111784535A CN202010602413.2A CN202010602413A CN111784535A CN 111784535 A CN111784535 A CN 111784535A CN 202010602413 A CN202010602413 A CN 202010602413A CN 111784535 A CN111784535 A CN 111784535A
Authority
CN
China
Prior art keywords
wind energy
interval
energy data
preset
prediction
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
CN202010602413.2A
Other languages
Chinese (zh)
Inventor
史志超
张涛
刘亚杰
雷洪涛
王锐
黄生俊
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.)
National University of Defense Technology
Original Assignee
National University of Defense Technology
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 National University of Defense Technology filed Critical National University of Defense Technology
Priority to CN202010602413.2A priority Critical patent/CN111784535A/en
Publication of CN111784535A publication Critical patent/CN111784535A/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
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/004—Artificial life, i.e. computing arrangements simulating life
    • G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
    • 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
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • 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
    • 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
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
    • 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)
  • Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Economics (AREA)
  • Strategic Management (AREA)
  • General Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Development Economics (AREA)
  • General Business, Economics & Management (AREA)
  • Tourism & Hospitality (AREA)
  • Marketing (AREA)
  • General Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • Mathematical Physics (AREA)
  • Educational Administration (AREA)
  • Operations Research (AREA)
  • Data Mining & Analysis (AREA)
  • Quality & Reliability (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Artificial Intelligence (AREA)
  • Game Theory and Decision Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Public Health (AREA)
  • Primary Health Care (AREA)
  • Water Supply & Treatment (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The application relates to a method and apparatus based on a feedback neural network. The method comprises the following steps: and acquiring wind energy data from a preset data source, and reconstructing the wind energy data into a wind energy data sequence containing wind energy time-varying information based on a delay embedding theorem to serve as a sample set of model training. And training a preset wind energy interval prediction model based on a feedback neural network according to the sample set and a preset target function. The objective function is constructed according to comprehensive evaluation indexes of the prediction interval output by the wind energy interval prediction model. And acquiring a real-time data sequence corresponding to the real-time wind energy data, and inputting the real-time data sequence into the trained wind energy interval prediction model to obtain a wind energy direct interval prediction result of the real-time wind energy data. By adopting the method, the prediction result error does not need to be subjected to the prior distribution assumption, and the quantitative information of the wind energy uncertainty can be provided in a wind energy interval mode on the premise of obviously reducing the calculated amount, so that the accuracy of the prediction of the wind energy interval is improved.

Description

Wind energy direct interval prediction method and device based on feedback neural network
Technical Field
The application relates to the technical field of neural networks and wind energy prediction, in particular to a method and a device for predicting a direct wind energy interval based on a feedback neural network.
Background
With the development of advanced power generation technology, the renewable energy power generation has also increased greatly in recent years. Wind power has become one of the most popular renewable resources, being a clean and readily available resource. But due to the variability of wind speed, wind energy is therefore intermittent and uncertain. In order to deal with the characteristic of wind energy and adapt to the operation and scheduling requirements of the smart grid system, the wind energy needs to be effectively predicted.
There are many methods for realizing wind energy prediction, and for these methods, direct prediction and indirect prediction can be classified according to whether the methods respectively predict wind power and wind/electricity conversion characteristics; according to the scale of the prediction time, the prediction is divided into short-term prediction, medium-term prediction and long-term prediction; according to the type of the prediction model, physical model prediction and statistical model prediction are divided; the prediction method is classified into point prediction, interval prediction and the like according to the type of the prediction value. Currently, point prediction methods are widely used, and wind energy is predicted by generating a certain predicted value at a certain time. Point prediction cannot provide uncertainty information of its predicted value, so prediction errors cannot be eliminated, and prediction accuracy is highly variable. Therefore, wind energy prediction using a point prediction method may have a significant impact on the stability and reliability of the operation of the power system.
Compared with the point prediction method, the probability prediction method can provide additional quantitative information of wind energy uncertainty. At present, there are some wind energy interval prediction researches based on a feedforward neural network prediction model. However, because the information in the feedforward neural network can only be transmitted in one direction, and the network does not have a memory function, the wind energy interval prediction accuracy of the wind energy prediction model based on the feedforward neural network is not high.
Disclosure of Invention
In view of the foregoing, there is a need to provide a method and an apparatus for predicting a direct interval of wind energy based on a feedback neural network.
A wind energy direct interval prediction method based on a feedback neural network, the method comprising:
the method comprises the steps of obtaining wind energy data from a preset data source, reconstructing the wind energy data into a wind energy data sequence containing wind energy time-varying information based on a delay embedding theorem, and constructing a sample set for model training according to the wind energy data sequence.
And training a preset wind energy interval prediction model based on a feedback neural network according to the sample set and a preset target function. The objective function is constructed according to comprehensive evaluation indexes of the prediction interval output by the wind energy interval prediction model.
And acquiring a real-time data sequence corresponding to the real-time wind energy data, and inputting the real-time data sequence into the trained wind energy interval prediction model to obtain a wind energy direct interval prediction result of the real-time wind energy data.
In one embodiment, the comprehensive evaluation index includes: prediction interval mean square error, coverage and normalized mean width.
The method for constructing the target function according to the comprehensive evaluation indexes of the prediction interval output by the wind energy interval prediction model comprises the following steps:
the objective function constructed according to the mean square error of the prediction interval, the coverage rate and the normalized mean width is as follows:
NCWC＝PINAW+γ(PICP)·e-η(PICP-μ)+PIMSE
Figure BDA0002559447430000021
Figure BDA0002559447430000022
wherein NCWC represents an objective function, PIMSE represents a prediction interval mean square error, PICP represents a coverage rate, PINAW represents a normalized average width, N is the number of wind energy data, y is the number of wind energy dataiIs the wind energy data, LiIs a predicted inter-zone upper endpoint value, UiIs the predicted interval lower endpoint value, η and μ are preset constants.
In one embodiment, the method for constructing the preset wind energy interval prediction model based on the feedback neural network comprises the following steps:
and acquiring the embedding dimension of the wind energy data sequence, and setting the number of input nodes of the feedback neural network according to the embedding dimension.
And setting a connection weight value of the feedback neural network according to a preset initialization value.
And inputting the sample set into a feedback neural network to obtain a prediction interval endpoint value output by the feedback neural network, and determining the number of hidden layer nodes and the number of carrying layer nodes of the feedback neural network according to a target function to obtain a wind energy interval prediction model.
In one embodiment, the steps of acquiring wind energy data from a preset data source, reconstructing the wind energy data into a wind energy data sequence containing wind energy time-varying information based on a delayed embedding theorem, and constructing a sample set for model training according to the wind energy data sequence include:
wind energy data is obtained from a preset data source.
Determining the value of time delay tau by using a mutual information function, determining the value of an embedding dimension m by using a false nearest neighbor method, and reconstructing wind energy data into a wind energy data sequence X containing wind energy time-varying information based on a delay embedding theorem0(t)：
X0(t)＝[X(t),X(t-τ),…,X(t-(m-1)τ)]
Wherein X (t) is the wind energy data at time t.
And constructing a sample set for model training according to the wind energy data sequence.
In one embodiment, the step of training a preset wind energy interval prediction model based on a feedback neural network according to the sample set and a preset target function comprises:
and inputting the sample set into a preset wind energy interval prediction model based on a feedback neural network to obtain a prediction interval endpoint value output by the wind energy interval prediction model.
And constructing a speed vector and a position vector of an individual at the 1 st moment in the dragonfly algorithm according to the initialized value of the inter-node connection weight of the wind energy interval prediction model.
And acquiring the speed vector and the position vector of the individual at the ith moment, calculating the speed vector and the position vector of the individual at the (i + 1) th moment by using a preset dragonfly algorithm rule, and acquiring the fitness of the individual according to a target function. Wherein i is more than or equal to 1 and less than or equal to N-1.
And obtaining the individual with the highest fitness at the Nth moment, and setting a wind energy interval prediction model according to the connection weight value corresponding to the individual to obtain a final wind energy interval prediction model.
In one embodiment, the step of obtaining the speed vector and the position vector of the individual at the ith moment, calculating the speed vector and the position vector of the individual at the (i + 1) th moment by using a preset dragonfly algorithm rule, and obtaining the fitness of the individual according to a target function includes:
and acquiring the speed vector and the position vector of the individual at the ith moment, and acquiring the number of neighbors of the individual according to a preset algorithm.
And for the individuals with the neighbor numbers larger than 0, calculating the speed vector and the position vector of the individual at the (i + 1) th moment by using a preset dragonfly algorithm rule.
And for the individual with the number of the neighbors equal to 0, calculating the speed vector and the position vector of the individual at the i +1 th moment by using a Levy flight algorithm.
And obtaining the fitness of the individual according to the objective function.
In one embodiment, the steps of acquiring wind energy data from a preset data source, reconstructing the wind energy data into a wind energy data sequence containing wind energy time-varying information based on a delayed embedding theorem, and constructing a sample set for model training according to the wind energy data sequence include:
wind energy data are obtained from a preset data source, and normalization processing is carried out on the wind energy data to obtain normalized wind energy data with the numerical value interval of [ -1,1 ].
And reconstructing the normalized wind energy data into a wind energy data sequence containing wind energy time-varying information based on a delay embedding theorem, and constructing a sample set for model training according to the wind energy data sequence.
A wind energy direct interval prediction device based on a feedback neural network, the device comprising:
the training sample set generation module is used for acquiring wind energy data from a preset data source, reconstructing the wind energy data into a wind energy data sequence containing wind energy time-varying information based on a delayed embedding theorem, and constructing a sample set for model training according to the wind energy data sequence.
And the wind energy interval prediction model training module is used for training a preset wind energy interval prediction model based on a feedback neural network according to the sample set and a preset target function. And the objective function is constructed according to the comprehensive evaluation index of the prediction interval output by the wind energy interval prediction model.
And the wind energy interval prediction module is used for acquiring a real-time data sequence corresponding to the real-time wind energy data, inputting the real-time data sequence into the trained wind energy interval prediction model, and obtaining a wind energy direct interval prediction result of the real-time wind energy data.
A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor, when executing the computer program, implements the steps of the method in any of the above embodiments.
A computer-readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, is adapted to carry out the steps of the method of any of the above embodiments.
According to the wind energy direct interval prediction method and device based on the feedback neural network, the wind energy direct interval prediction model is established based on the feedback neural network, the target function is established based on historical wind energy data, the final wind energy interval prediction model is obtained by using evolutionary algorithm training, the prediction result error does not need to be subjected to the pre-distribution assumption, the quantitative information of wind energy uncertainty can be provided in a wind energy interval mode on the premise of obviously reducing the calculated amount, and the accuracy of wind energy prediction is improved.
Drawings
FIG. 1 is a schematic flow chart illustrating steps of a method for predicting a direct interval of wind energy based on a feedback neural network according to an embodiment;
FIG. 2 is a schematic structural diagram of a feedback neural network used in the method for predicting the direct interval of wind energy based on the feedback neural network according to an embodiment;
FIG. 3 is a wind energy data graph used in a direct interval wind energy prediction method based on a feedback neural network according to an embodiment;
FIG. 4 is a detailed flowchart of a wind energy direct interval prediction method based on a feedback neural network in one embodiment;
FIG. 5 is a data diagram of the interval prediction result of the direct wind energy interval prediction method based on the feedback neural network in one embodiment;
FIG. 6 is a diagram illustrating an internal structure of a computer device according to an embodiment.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.
The feedback Neural Network (RNN) refers to a Neural Network with a feedback structure, which has higher accuracy in time series-based prediction and is more suitable for dynamic system modeling than a feedforward Neural Network.
In probability prediction, uncertainty can be expressed by probability measures such as probability density function, quantile, interval, etc., wherein interval prediction is the most convenient method for visual expression. Among different interval prediction methods, an upper and lower boundary estimation method is a nonparametric method capable of directly establishing a prediction interval, and the calculation amount in the prediction process can be greatly reduced because no assumption is made on prediction error distribution.
Based on the consideration of improving the prediction performance and reducing the calculation burden, the method establishes a prediction model based on the feedback neural network, carries out direct interval prediction on the wind energy, and provides a wind energy direct interval prediction method based on the feedback neural network.
In one embodiment, as shown in fig. 1, a method for predicting a direct interval of wind energy based on a feedback neural network is provided, which is described by taking a feedback neural network model as shown in fig. 2 as an example, and includes the following steps:
step 102: the method comprises the steps of obtaining wind energy data from a preset data source, reconstructing the wind energy data into a wind energy data sequence containing wind energy time-varying information based on a delay embedding theorem, and constructing a sample set for model training according to the wind energy data sequence.
In consideration of the changeable dynamics of wind energy, the present embodiment adopts a state space reconstruction technique based on the delayed embedding theorem to process the collected historical wind energy data, and constructs a sample set for model training according to the wind energy data sequence. Besides the sample set, a corresponding verification data set and a corresponding test data set can be constructed according to historical wind energy data and are used for verifying a wind energy interval prediction model and testing the generalization capability of the model.
The state space reconstruction technology based on the delay embedding theorem considers that high-dimensional information is contained in one-dimensional chaotic data, and the one-dimensional time data can be reconstructed into an m-dimensional delay vector by setting a time delay tau and an embedding dimension m:
X0(t)＝[X(t),X(t-τ),…,X(t-(m-1)τ)](5)
and the obtained delay vector is used as input data of the wind energy interval prediction model, and corresponding upper end point value of the prediction interval and lower end point value of the prediction region can be obtained and output.
Step 104: and training a preset wind energy interval prediction model based on a feedback neural network according to the sample set and a preset target function. The objective function is constructed according to comprehensive evaluation indexes of the prediction interval output by the wind energy interval prediction model.
Specifically, the preset wind energy interval prediction model adopts a feedback neural network as shown in fig. 2. The feedback neural network comprises four layers, namely an input layer, a hidden layer, a carrying layer and an output layer. The model outputs a prediction interval corresponding to each input vector, namely the model has two output nodes which respectively output the upper end point value and the lower end point value of the prediction interval. Feedback connection is arranged between the hidden layer and the receiving layer in the model, and the connection weight W of the feedback connection2Is fixed, forward connection weight W between other nodes1、W3It is variably adjustable. The model can be expressed mathematically as follows:
x(k)＝φ[W1I(k-1)+W2xc(k)+b1]
xc(k)＝x(k-1)
z(k)＝f(W3x(k)+b2) (6)
wherein x (-) and z (-) denote the output of the hidden layer and the output layer, respectively, xc(. h) is the output of the pinned layer, phi (·) represents a transfer function, typically a hyperbolic tangent sigmoid function, and f represents a purely linear activation function.
In this embodiment, the number of input nodes, the number of nodes of the hidden layer, and the connection weight value in the feedforward neural network are preset, where the number of input nodes is equal to the dimension m of the wind energy data sequence in step 102. In order to optimize the value of the inter-node connection weight in the model and obtain the optimal wind energy interval prediction model, the embodiment trains the wind energy interval prediction model by using an evolutionary algorithm, and the evolutionary algorithm which can be used comprises: genetic Algorithms (GA), particle swarm optimization algorithms (PSO), and the like. In the training process, the sample set constructed in the step 102 is used as input, the value of the corresponding objective function is calculated according to the upper end point value and the lower end point value of the prediction interval output by the model, and the fitness of the individual corresponding to the connection weight between the nodes in the current model in the evolutionary algorithm is measured according to the value of the objective function. And updating the setting of the connection weight between the nodes in the model according to the evolutionary algorithm, and selecting the connection weight between the nodes with the highest fitness when a preset training termination condition is reached to obtain a corresponding optimal wind energy interval prediction model.
For the model training process, the objective function is a key control factor, and the definition mode of the objective function greatly influences the optimization degree of the final model. In this application, performing interval prediction is a process of constructing an estimation interval with a certain confidence level. A high-quality prediction interval should have high reliability and small interval width, and the indexes that can be used to evaluate the quality of the prediction interval include the Prediction Interval Coverage Probability (PICP), the Prediction Interval Normalized Average Width (PINAW), etc. Wherein, the PICP represents the probability that the real target value is covered by the prediction interval, the higher the PICP is, the more target values are located in the constructed prediction interval, and the expression is:
Figure BDA0002559447430000071
wherein the content of the first and second substances,
Figure BDA0002559447430000081
n is the number of test samples, yiIs the target value, LiAnd UiRespectively, the lower and upper bounds of the prediction interval.
However, as the prediction interval width increases, the PICP value inevitably increases. Therefore, it is necessary to provide a quantization index, namely PINAW, which measures the prediction interval width and is expressed as:
Figure BDA0002559447430000082
where Rg represents the range of target values, i.e. the difference between the target maximum and minimum values.
Based on PICP and PINAW, a comprehensive index can be defined, i.e. based on the Coverage Width Criterion (CWC), whose expression is:
CWC＝PINAW+γ(PICP)·e-η(PICP-μ)
Figure BDA0002559447430000083
where η and μ are preset control parameters, the former is typically a large constant to penalize invalid prediction intervals, and the latter can be determined based on a nominal confidence level.
Step 106: and acquiring a real-time data sequence corresponding to the real-time wind energy data, and inputting the real-time data sequence into the trained wind energy interval prediction model to obtain a wind energy direct interval prediction result of the real-time wind energy data.
Specifically, for wind energy direct interval prediction, real-time wind energy data are generated into corresponding real-time data sequences, the real-time data sequences are input into a trained wind energy interval prediction model, and a direct prediction result of a future wind energy interval is obtained according to the real-time wind energy data.
According to the wind energy direct interval prediction method based on the feedback neural network, a wind energy direct interval prediction model is established based on the feedback neural network, a target function is established based on historical wind energy data, a final wind energy interval prediction model is obtained by using an evolutionary algorithm for training, the prediction result error does not need to be subjected to a pre-distribution assumption, quantitative information of wind energy uncertainty can be provided in a wind energy interval mode on the premise of obviously reducing the calculated amount, and the accuracy of wind energy prediction is improved.
In one embodiment, the comprehensive evaluation index includes: prediction interval mean square error, coverage and normalized mean width. The method for constructing the target function according to the comprehensive evaluation indexes of the prediction interval output by the wind energy interval prediction model comprises the following steps: the objective function constructed according to the mean square error of the prediction interval, the coverage rate and the normalized mean width is as follows:
NCWC＝PINAW+γ(PICP)·e-η(PICP-μ)+PIMSE
Figure BDA0002559447430000091
Figure BDA0002559447430000092
wherein NCWC represents an objective function, PIMSE represents a prediction interval mean square error, PICP represents a coverage rate, PINAW represents a normalized average width, N is the number of wind energy data, y is the number of wind energy dataiIs the wind energy data, LiIs a predicted inter-zone upper endpoint value, UiIs the predicted interval lower endpoint value, η and μ are preset constants.
It can be seen that the PIMSE and PINAW indicators only determine the quality of the prediction interval based on whether the historical data is within the prediction interval, and do not relate to the error relationship between the specific values of the historical data and the prediction interval. In order to better utilize known historical wind energy data, the embodiment defines the error quantization index of a prediction interval by adopting an error quantization index defining mode similar to that in point prediction, such as mean square error, standard deviation and the like, and defines an objective function by integrating the error quantization index, a PINAW index and a PIMSE index, so that the interval prediction problem is converted into a single-target minimization problem, namely the optimization degree of a model is judged by the value of the objective function, and the prediction accuracy of a final model can be improved.
Specifically, in the mean square error definition method in reference point prediction of this embodiment, a prediction interval mean square error PIMSE is defined based on historical wind energy data, and is used as a quantization index for measuring interval prediction quality. The prediction interval mean square error PIMSE, the traditional prediction interval coverage rate PICP and the prediction interval normalized average width PINAW are superposed, a prediction interval comprehensive quality index NCWC which can reflect the error between the prediction interval and historical data, the prediction interval coverage probability and the prediction interval width is defined, and the prediction interval comprehensive quality index NCWC is used as a target function for measuring the optimization degree of the wind energy interval prediction model. Wherein, the specific definition of NCWC is:
NCWC＝PINAW+γ(PICP)·e-η(PICP-μ)+PIMSE
Figure BDA0002559447430000093
where η and μ are preset control parameters, the former is typically a large constant to penalize invalid prediction intervals, and the latter can be determined based on a nominal confidence level.
According to the method, the mean square error of the prediction interval and a new comprehensive quality index NCWC of the prediction interval are defined, so that historical wind energy data can be better utilized in the model training process, and the prediction accuracy of the final wind energy interval prediction model is improved.
In one embodiment, the method for constructing the preset wind energy interval prediction model based on the feedback neural network comprises the following steps:
and acquiring the embedding dimension of the wind energy data sequence, and setting the number of input nodes of the feedback neural network according to the embedding dimension.
And setting a connection weight value of the feedback neural network according to a preset initialization value.
And inputting the sample set into a feedback neural network to obtain a prediction interval endpoint value output by the feedback neural network, and determining the number of hidden layer nodes and the number of carrying layer nodes of the feedback neural network according to a target function to obtain a wind energy interval prediction model.
Specifically, the present embodiment performs initialization setting on parameters of a preset wind energy interval prediction model before acquiring the model. The number of the input nodes is the embedding dimension of the wind energy data sequence, the connection weight value between the nodes in the feedback neural network can be set to be a smaller random number, and the number of the output nodes is 2.
When the node number of the hidden layer and the node number of the bearing layer are selected, different node numbers can be set, the same group of wind energy data sequences are input into wind energy interval prediction models which are set differently, and the values of the objective functions corresponding to the models are calculated according to the output prediction intervals, so that the optimal node numbers of the hidden layer and the bearing layer are obtained. Taking the 5-fold cross validation method as an example, the number of the optimal hidden layer nodes obtained in this embodiment is 5, and the number of the carrying layer nodes is also 5.
The embodiment provides a method for obtaining the final wind energy interval prediction model structure parameters, and the optimal wind energy prediction model can be obtained in the subsequent model training process.
In one embodiment, the steps of acquiring wind energy data from a preset data source, reconstructing the wind energy data into a wind energy data sequence containing wind energy time-varying information based on a delayed embedding theorem, and constructing a sample set for model training according to the wind energy data sequence include:
wind energy data is obtained from a preset data source.
Determining the value of time delay tau by using a mutual information function, determining the value of an embedding dimension m by using a false nearest neighbor method, and reconstructing wind energy data into a wind energy data sequence X containing wind energy time-varying information based on a delay embedding theorem0(t)：
X0(t)＝[X(t),X(t-τ),…,X(t-(m-1)τ)]
Wherein X (t) is the wind energy data at time t.
And constructing a sample set for model training according to the wind energy data sequence.
Specifically, the wind energy data refers to historical wind energy data recorded within a period of time, and can be used as training data of a wind energy interval prediction model, and can also be input into the trained wind energy interval prediction model to predict an interval of a future wind energy data value. The wind energy data can adopt wind energy data per hour, and can be obtained from an IESO website from 1 month and 1 day in 2016 to 4 months and 7 days in 2017 as model training data, and specific data are shown in FIG. 3.
For a given wind energy data set, reconstructing the wind energy data set into a wind energy data sequence X containing wind energy time-varying information based on a delay embedding theorem0At (t), the value of its time delay τ can be determined using a mutual information function, and the embedding dimension m can be determined using a pseudo-nearest neighbor method. More specifically, this can be done using the mutual and false _ nearest utility functions in the TISEAN toolkit. The values of m and determined using the above method are 16 and 7, respectively, for the wind energy data set selected in this embodiment. Therefore, the number of input nodes of the wind energy interval prediction model based on the feedback neural network in the present embodiment is also 7 corresponding to the dimension of the delay vector.
The embodiment provides a specific method for reconstructing the wind energy data into a wind energy data sequence containing wind energy time-varying information based on the delayed embedding theorem, so that the number of input nodes of a wind energy interval prediction model based on a feedback neural network can be determined.
In one embodiment, the step of training a preset wind energy interval prediction model based on a feedback neural network according to the sample set and a preset target function comprises:
and inputting the sample set into a preset wind energy interval prediction model based on a feedback neural network to obtain a prediction interval endpoint value output by the wind energy interval prediction model.
And constructing a speed vector and a position vector of an individual at the 1 st moment in the dragonfly algorithm according to the initialized value of the inter-node connection weight of the wind energy interval prediction model.
And acquiring the speed vector and the position vector of the individual at the ith moment, calculating the speed vector and the position vector of the individual at the (i + 1) th moment by using a preset dragonfly algorithm rule, and acquiring the fitness of the individual according to a target function. Wherein i is more than or equal to 1 and less than or equal to N-1.
And obtaining the individual with the highest fitness at the Nth moment, and setting a wind energy interval prediction model according to the connection weight value corresponding to the individual to obtain a final wind energy interval prediction model.
The Dragonfly Algorithm (DA) chosen in this embodiment is a population algorithm inspired by dragonfly behavior, in which each dragonfly (i.e. individual) has two vectors, position (P) and velocity (V), where the position vector represents one potential solution to the optimization problem. Specifically, in this embodiment, a position vector and a velocity vector of each individual are constructed according to a connection weight value between nodes in the feedback neural network, that is, each individual corresponds to one interval prediction model, and an initial value of the position and velocity vector of each individual may be randomly set according to a preset value range. For an individual with a neighbor, the iterative update algorithm of the position vector and the velocity vector is defined as follows:
Piter+1＝Piter+Viter+1
Viter+1＝(sSei+aAli+cCoi+fFoi+eEni)+wViter
wherein, PiterAnd ViterRespectively representing that when the iteration times is iter times, the position vector and the speed vector of an individual, and s, a, c, f, e and w are corresponding iteration weight coefficients, and the values of the iteration weight coefficients are adaptively adjusted in the optimization process. The velocity vector contains five individual behavior vectors, namely separation (Se), union (Al), aggregate (Co), approaching food (Fo) and departing enemy (En), and is defined as follows:
Figure BDA0002559447430000121
Figure BDA0002559447430000122
Figure BDA0002559447430000123
Foi＝P+-Pi
Eni＝P-+Pi
wherein P isiIs the location of the current individual, PjAnd VjRespectively, the position and speed of the jth neighbor, Num the number of neighbor, and P + and P-the positions of dragonfly food and enemy, respectively.
For individuals without neighbors, a random walk algorithm, such as Levy (Levy) flight, can be applied to improve the exploration capability and random behavior, and the position updating method is as follows:
Piter+1＝Piter+Levy(dim)×Piter
Figure BDA0002559447430000124
where dim represents the dimension of the position vector, c1And c2Are each [0,1]λ is a constant value, the value in this embodiment is 1.5, and the parameter ρ can be calculated by the following formula:
Figure BDA0002559447430000131
wherein, (n) ═ n-1! .
Further, in the present embodiment, the population size of the dragonfly algorithm is taken to be 30, and the control parameter μ in the objective function NCWC is set to the nominal confidence level, the value thereof is 0.9, and the parameter η is set to 50.
In the training process, the target function NCWC is used as the fitness function of the individual, and the dragonfly algorithm is used for solving the interval prediction problem of the single target. Through calculating the fitness function corresponding to each individual in the iteration process, the individual with the optimal fitness in the training process is recorded as the food of the population, the individual with the worst fitness is recorded as the enemy of the population, and the aim of reserving the individual with better fitness is achieved through iteration. And when the preset maximum iteration times are reached, terminating the model training, wherein the obtained food position represents the optimal connection weight value between the nodes of the feedback neural network and corresponds to the optimal wind energy interval prediction model.
In the embodiment, a dragonfly algorithm is adopted for training the wind energy interval prediction model, and compared with the traditional evolutionary algorithm, the dragonfly algorithm has the advantages of stability, high optimization speed and high global optimization capability, the model training process can be optimized, and the wind energy interval prediction model with higher accuracy is obtained.
In one embodiment, the step of obtaining the speed vector and the position vector of the individual at the ith moment, calculating the speed vector and the position vector of the individual at the (i + 1) th moment by using a preset dragonfly algorithm rule, and obtaining the fitness of the individual according to a target function includes:
and acquiring the speed vector and the position vector of the individual at the ith moment, and acquiring the number of neighbors of the individual according to a preset algorithm.
And for the individuals with the neighbor numbers larger than 0, calculating the speed vector and the position vector of the individual at the (i + 1) th moment by using a preset dragonfly algorithm rule.
And for the individual with the number of the neighbors equal to 0, calculating the speed vector and the position vector of the individual at the i +1 th moment by using a Levy flight algorithm.
And obtaining the fitness of the individual according to the objective function.
In the dragonfly algorithm, the initial value of the iterative weight coefficient w of the velocity vector of an individual is adaptively adjusted, typically by a linear decreasing manner. In this embodiment, the iteration updating method for improving the weight coefficient w of the individual without the neighbor by using the levy flight algorithm includes:
Figure BDA0002559447430000141
witer＝witer+Levy(dim)
wherein wmaxAnd wminThe maximum weight value and the minimum weight value are respectively preset, the values in this embodiment are 1 and 0.7, and Maxt is the maximum iteration number, and the value in this embodiment is 1000.
In the embodiment, the Rivie flight algorithm is adopted to improve the iterative updating mode of the weight coefficient of the individual without the neighbor in the dragonfly algorithm, so that the searching capability of the dragonfly algorithm can be enhanced, and the optimal model training effect can be ensured to be obtained.
In one embodiment, the steps of acquiring wind energy data from a preset data source, reconstructing the wind energy data into a wind energy data sequence containing wind energy time-varying information based on a delayed embedding theorem, and constructing a sample set for model training according to the wind energy data sequence include:
wind energy data are obtained from a preset data source, and normalization processing is carried out on the wind energy data to obtain normalized wind energy data with the numerical value interval of [ -1,1 ].
And reconstructing the normalized wind energy data into a wind energy data sequence containing wind energy time-varying information based on a delay embedding theorem, and constructing a sample set for model training according to the wind energy data sequence.
In this embodiment, normalization processing is performed on the sample set used for model training, so that comparability between data can be enhanced, dimension influence is eliminated, and the training process of the model is accelerated.
In one embodiment, wind energy data is acquired as shown in FIG. 3 and divided into training data and test data. And inputting the test data into a final wind energy interval prediction model obtained by training, and testing the prediction performance of the model. The implementation flow of the method provided by the embodiment is shown in fig. 4, and fig. 5 shows the wind energy interval prediction result based on the test data.
It should be understood that, although the steps in the flowchart of fig. 1 are shown in order as indicated by the arrows, the steps are not necessarily performed in order as indicated by the arrows. The steps are not performed in the exact order shown and described, and may be performed in other orders, unless explicitly stated otherwise. Moreover, at least a portion of the steps in fig. 1 may include multiple sub-steps or multiple stages that are not necessarily performed at the same time, but may be performed at different times, and the order of performance of the sub-steps or stages is not necessarily sequential, but may be performed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps.
A wind energy direct interval prediction device based on a feedback neural network, the device comprising:
the training sample set generation module is used for acquiring wind energy data from a preset data source, reconstructing the wind energy data into a wind energy data sequence containing wind energy time-varying information based on a delayed embedding theorem, and constructing a sample set for model training according to the wind energy data sequence.
And the wind energy interval prediction model training module is used for training a preset wind energy interval prediction model based on a feedback neural network according to the sample set and a preset target function. And the objective function is constructed according to the comprehensive evaluation index of the prediction interval output by the wind energy interval prediction model.
And the wind energy interval prediction module is used for acquiring a real-time data sequence corresponding to the real-time wind energy data, inputting the real-time data sequence into the trained wind energy interval prediction model, and obtaining a wind energy direct interval prediction result of the real-time wind energy data.
In one embodiment, the method further includes an objective function constructing module, configured to construct an objective function according to the prediction interval mean square error, the coverage rate, and the normalized mean width, where the objective function is:
NCWC＝PINAW+γ(PICP)·e-η(PICP-μ)+PIMSE
Figure BDA0002559447430000151
Figure BDA0002559447430000152
wherein NCWC represents an objective function, PIMSE represents a prediction interval mean square error, PICP represents a coverage rate, PINAW represents a normalized average width, N is the number of wind energy data, y is the number of wind energy dataiIs the wind energy data, LiIs a predicted inter-zone upper endpoint value, UiIs the predicted interval lower endpoint value, η and μ are preset constants.
In one embodiment, the wind energy interval prediction model building module is further included and is used for obtaining the embedding dimension of the wind energy data sequence and setting the number of input nodes of the feedback neural network according to the embedding dimension.
And setting a connection weight value of the feedback neural network according to a preset initialization value.
And inputting the sample set into a feedback neural network to obtain a prediction interval endpoint value output by the feedback neural network, and determining the number of hidden layer nodes and the number of carrying layer nodes of the feedback neural network according to a target function to obtain a wind energy interval prediction model.
In one embodiment, the training sample set generation module is configured to obtain wind energy data from a preset data source.
Determining the value of time delay tau by using a mutual information function, determining the value of an embedding dimension m by using a false nearest neighbor method, and reconstructing wind energy data into a wind energy data sequence X containing wind energy time-varying information based on a delay embedding theorem0(t)：
X0(t)＝[X(t),X(t-τ),…,X(t-(m-1)τ)]
Wherein X (t) is the wind energy data at time t.
And constructing a sample set for model training according to the wind energy data sequence.
In one embodiment, the wind energy interval prediction model training module is configured to: and inputting the sample set into a preset wind energy interval prediction model based on a feedback neural network to obtain a prediction interval endpoint value output by the wind energy interval prediction model.
And constructing a speed vector and a position vector of an individual at the 1 st moment in the dragonfly algorithm according to the initialized value of the inter-node connection weight of the wind energy interval prediction model.
And acquiring the speed vector and the position vector of the individual at the ith moment, calculating the speed vector and the position vector of the individual at the (i + 1) th moment by using a preset dragonfly algorithm rule, and acquiring the fitness of the individual according to a target function. Wherein i is more than or equal to 1 and less than or equal to N-1.
And obtaining the individual with the highest fitness at the Nth moment, and setting a wind energy interval prediction model according to the connection weight value corresponding to the individual to obtain a final wind energy interval prediction model.
In one embodiment, the wind energy interval prediction model training module is further configured to: and acquiring the speed vector and the position vector of the individual at the ith moment, and acquiring the number of neighbors of the individual according to a preset algorithm. And for the individuals with the neighbor numbers larger than 0, calculating the speed vector and the position vector of the individual at the (i + 1) th moment by using a preset dragonfly algorithm rule. And for the individual with the number of the neighbors equal to 0, calculating the speed vector and the position vector of the individual at the i +1 th moment by using a Levy flight algorithm. And obtaining the fitness of the individual according to the objective function.
In one embodiment, the training sample set generating module is configured to: wind energy data are obtained from a preset data source, and normalization processing is carried out on the wind energy data to obtain normalized wind energy data with the numerical value interval of [ -1,1 ].
And reconstructing the normalized wind energy data into a wind energy data sequence containing wind energy time-varying information based on a delay embedding theorem, and constructing a sample set for model training according to the wind energy data sequence.
For specific limitations of the wind energy direct interval prediction device based on the feedback neural network, reference may be made to the above limitations of the wind energy direct interval prediction method based on the feedback neural network, and details thereof are not repeated herein. All or part of the modules in the wind energy direct interval prediction device based on the feedback neural network can be realized by software, hardware and a combination thereof. The modules can be embedded in a hardware form or independent from a processor in the computer device, and can also be stored in a memory in the computer device in a software form, so that the processor can call and execute operations corresponding to the modules.
In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as shown in fig. 6. The computer device includes a processor, a memory, a network interface, and a database connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The database of the computer device is used for storing wind energy data, wind energy data sequences, interval prediction models, objective functions, model parameter settings and other data. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to implement a wind energy direct interval prediction method based on a feedback neural network.
Those skilled in the art will appreciate that the architecture shown in fig. 6 is merely a block diagram of some of the structures associated with the disclosed aspects and is not intended to limit the computing devices to which the disclosed aspects apply, as particular computing devices may include more or less components than those shown, or may combine certain components, or have a different arrangement of components.
In one embodiment, there is provided a computer device comprising a memory storing a computer program and a processor implementing the following steps when the processor executes the computer program:
the method comprises the steps of obtaining wind energy data from a preset data source, reconstructing the wind energy data into a wind energy data sequence containing wind energy time-varying information based on a delay embedding theorem, and constructing a sample set for model training according to the wind energy data sequence.
And training a preset wind energy interval prediction model based on a feedback neural network according to the sample set and a preset target function. The objective function is constructed according to comprehensive evaluation indexes of the prediction interval output by the wind energy interval prediction model.
And acquiring a real-time data sequence corresponding to the real-time wind energy data, and inputting the real-time data sequence into the trained wind energy interval prediction model to obtain a wind energy direct interval prediction result of the real-time wind energy data.
In one embodiment, the processor, when executing the computer program, further performs the steps of: the objective function constructed according to the mean square error of the prediction interval, the coverage rate and the normalized mean width is as follows:
NCWC＝PINAW+γ(PICP)·e-η(PICP-μ)+PIMSE
Figure BDA0002559447430000181
Figure BDA0002559447430000182
wherein NCWC represents an objective function, PIMSE represents a prediction interval mean square error, PICP represents a coverage rate, PINAW represents a normalized average width, N is the number of wind energy data, y is the number of wind energy dataiIs the wind energy data, LiIs a predicted inter-zone upper endpoint value, UiIs the predicted interval lower endpoint value, η and μ are preset constants.
In one embodiment, the processor, when executing the computer program, further performs the steps of: and acquiring the embedding dimension of the wind energy data sequence, and setting the number of input nodes of the feedback neural network according to the embedding dimension. And setting a connection weight value of the feedback neural network according to a preset initialization value. And inputting the sample set into a feedback neural network to obtain a prediction interval endpoint value output by the feedback neural network, and determining the number of hidden layer nodes and the number of carrying layer nodes of the feedback neural network according to a target function to obtain a wind energy interval prediction model.
In one embodiment, the processor, when executing the computer program, further performs the steps of: wind energy data is obtained from a preset data source. Determining the value of time delay tau by using a mutual information function, determining the value of an embedding dimension m by using a false nearest neighbor method, and reconstructing wind energy data into a wind energy data sequence X containing wind energy time-varying information based on a delay embedding theorem0(t)：
X0(t)＝[X(t),X(t-τ),…,X(t-(m-1)τ)]
Wherein X (t) is the wind energy data at time t. And constructing a sample set for model training according to the wind energy data sequence.
In one embodiment, the processor, when executing the computer program, further performs the steps of: and inputting the sample set into a preset wind energy interval prediction model based on a feedback neural network to obtain a prediction interval endpoint value output by the wind energy interval prediction model. And constructing a speed vector and a position vector of an individual at the 1 st moment in the dragonfly algorithm according to the initialized value of the inter-node connection weight of the wind energy interval prediction model. And acquiring the speed vector and the position vector of the individual at the ith moment, calculating the speed vector and the position vector of the individual at the (i + 1) th moment by using a preset dragonfly algorithm rule, and acquiring the fitness of the individual according to a target function. Wherein i is more than or equal to 1 and less than or equal to N-1. And obtaining the individual with the highest fitness at the Nth moment, and setting a wind energy interval prediction model according to the connection weight value corresponding to the individual to obtain a final wind energy interval prediction model.
In one embodiment, the processor, when executing the computer program, further performs the steps of: and acquiring the speed vector and the position vector of the individual at the ith moment, and acquiring the number of neighbors of the individual according to a preset algorithm. And for the individuals with the neighbor numbers larger than 0, calculating the speed vector and the position vector of the individual at the (i + 1) th moment by using a preset dragonfly algorithm rule. And for the individual with the number of the neighbors equal to 0, calculating the speed vector and the position vector of the individual at the i +1 th moment by using a Levy flight algorithm. And obtaining the fitness of the individual according to the objective function.
In one embodiment, the processor, when executing the computer program, further performs the steps of: wind energy data are obtained from a preset data source, and normalization processing is carried out on the wind energy data to obtain normalized wind energy data with the numerical value interval of [ -1,1 ]. And reconstructing the normalized wind energy data into a wind energy data sequence containing wind energy time-varying information based on a delay embedding theorem, and constructing a sample set for model training according to the wind energy data sequence.
In one embodiment, a computer-readable storage medium is provided, having a computer program stored thereon, which when executed by a processor, performs the steps of:
the method comprises the steps of obtaining wind energy data from a preset data source, reconstructing the wind energy data into a wind energy data sequence containing wind energy time-varying information based on a delay embedding theorem, and constructing a sample set for model training according to the wind energy data sequence.
And training a preset wind energy interval prediction model based on a feedback neural network according to the sample set and a preset target function. The objective function is constructed according to comprehensive evaluation indexes of the prediction interval output by the wind energy interval prediction model.
And acquiring a real-time data sequence corresponding to the real-time wind energy data, and inputting the real-time data sequence into the trained wind energy interval prediction model to obtain a wind energy direct interval prediction result of the real-time wind energy data.
In one embodiment, the computer program when executed by the processor further performs the steps of: the objective function constructed according to the mean square error of the prediction interval, the coverage rate and the normalized mean width is as follows:
NCWC＝PINAW+γ(PICP)·e-η(PICP-μ)+PIMSE
Figure BDA0002559447430000191
Figure BDA0002559447430000201
wherein NCWC represents an objective function, PIMSE represents a prediction interval mean square error, PICP represents a coverage rate, PINAW represents a normalized average width, N is the number of wind energy data, y is the number of wind energy dataiIs the wind energy data, LiIs a predicted inter-zone upper endpoint value, UiIs the predicted interval lower endpoint value, η and μ are preset constants.
In one embodiment, the computer program when executed by the processor further performs the steps of: and acquiring the embedding dimension of the wind energy data sequence, and setting the number of input nodes of the feedback neural network according to the embedding dimension. And setting a connection weight value of the feedback neural network according to a preset initialization value. And inputting the sample set into a feedback neural network to obtain a prediction interval endpoint value output by the feedback neural network, and determining the number of hidden layer nodes and the number of carrying layer nodes of the feedback neural network according to a target function to obtain a wind energy interval prediction model.
In one embodiment, the computer program when executed by the processor further performs the steps of: wind energy data is obtained from a preset data source. Determining the value of time delay tau by using a mutual information function, determining the value of an embedding dimension m by using a false nearest neighbor method, and reconstructing wind energy data into a wind energy data sequence X containing wind energy time-varying information based on a delay embedding theorem0(t)：
X0(t)＝[X(t),X(t-τ),…,X(t-(m-1)τ)]
Wherein X (t) is the wind energy data at time t. And constructing a sample set for model training according to the wind energy data sequence.
In one embodiment, the computer program when executed by the processor further performs the steps of: and inputting the sample set into a preset wind energy interval prediction model based on a feedback neural network to obtain a prediction interval endpoint value output by the wind energy interval prediction model. And constructing a speed vector and a position vector of an individual at the 1 st moment in the dragonfly algorithm according to the initialized value of the inter-node connection weight of the wind energy interval prediction model. And acquiring the speed vector and the position vector of the individual at the ith moment, calculating the speed vector and the position vector of the individual at the (i + 1) th moment by using a preset dragonfly algorithm rule, and acquiring the fitness of the individual according to a target function. Wherein i is more than or equal to 1 and less than or equal to N-1. And obtaining the individual with the highest fitness at the Nth moment, and setting a wind energy interval prediction model according to the connection weight value corresponding to the individual to obtain a final wind energy interval prediction model.
In one embodiment, the computer program when executed by the processor further performs the steps of: and acquiring the speed vector and the position vector of the individual at the ith moment, and acquiring the number of neighbors of the individual according to a preset algorithm. And for the individuals with the neighbor numbers larger than 0, calculating the speed vector and the position vector of the individual at the (i + 1) th moment by using a preset dragonfly algorithm rule. And for the individual with the number of the neighbors equal to 0, calculating the speed vector and the position vector of the individual at the i +1 th moment by using a Levy flight algorithm. And obtaining the fitness of the individual according to the objective function.
In one embodiment, the computer program when executed by the processor further performs the steps of: wind energy data are obtained from a preset data source, and normalization processing is carried out on the wind energy data to obtain normalized wind energy data with the numerical value interval of [ -1,1 ]. And reconstructing the normalized wind energy data into a wind energy data sequence containing wind energy time-varying information based on a delay embedding theorem, and constructing a sample set for model training according to the wind energy data sequence.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in the embodiments provided herein may include non-volatile and/or volatile memory, among others. Non-volatile memory can include read-only memory (ROM), Programmable ROM (PROM), Electrically Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus Direct RAM (RDRAM), direct bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.
The above-mentioned embodiments only express several embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, which falls within the scope of protection of the present application. Therefore, the protection scope of the present patent shall be subject to the appended claims.

Claims (10)

1. A wind energy direct interval prediction method based on a feedback neural network, the method comprising:
acquiring wind energy data from a preset data source, reconstructing the wind energy data into a wind energy data sequence containing wind energy time-varying information based on a delay embedding theorem, and constructing a sample set for model training according to the wind energy data sequence;
training a preset wind energy interval prediction model based on a feedback neural network according to the sample set and a preset target function; the objective function is constructed according to comprehensive evaluation indexes of a prediction interval output by a wind energy interval prediction model;
and acquiring a real-time data sequence corresponding to the real-time wind energy data, and inputting the real-time data sequence into the trained wind energy interval prediction model to obtain a wind energy direct interval prediction result of the real-time wind energy data.
2. The method of claim 1, wherein the composite evaluation index comprises: predicting the mean square error, the coverage rate and the standardized average width of the interval;
the method for constructing the objective function according to the comprehensive evaluation index of the prediction interval output by the wind energy interval prediction model comprises the following steps:
the objective function constructed according to the prediction interval mean square error, the coverage rate and the normalized average width is as follows:
NCWC＝PINAW+γ(PICP)·e-η(PICP-μ)+PIMSE
Figure FDA0002559447420000011
Figure FDA0002559447420000012
wherein NCWC represents an objective function, PIMSE represents a prediction interval mean square error, PICP represents a coverage rate, PINAW represents a normalized average width, N is the number of wind energy data, y is the number of wind energy dataiIs the wind energy data, LiIs a predicted inter-zone upper endpoint value, UiIs the predicted interval lower endpoint value, η and μ are preset constants.
3. The method of claim 1, wherein constructing the pre-set feedback neural network-based wind energy interval prediction model comprises:
acquiring the embedding dimension of the wind energy data sequence, and setting the number of input nodes of a feedback neural network according to the embedding dimension;
setting a connection weight value of the feedback neural network according to a preset initialization value;
and inputting the sample set into the feedback neural network to obtain a prediction interval endpoint value output by the feedback neural network, and determining the number of hidden layer nodes and the number of carrying layer nodes of the feedback neural network according to the target function to obtain a wind energy interval prediction model.
4. The method according to any one of claims 1 to 3, wherein the steps of obtaining wind energy data from a preset data source, reconstructing the wind energy data into a wind energy data sequence containing wind energy time-varying information based on a delayed embedding theorem, and constructing a sample set for model training according to the wind energy data sequence comprise:
acquiring wind energy data from a preset data source;
determining the value of time delay tau by using a mutual information function, determining the value of an embedding dimension m by using a false nearest neighbor method, and reconstructing the wind energy data into a wind energy data sequence X containing wind energy time-varying information based on a delay embedding theorem0(t)：
X0(t)＝[X(t),X(t-τ),…,X(t-(m-1)τ)]
Wherein X (t) is the wind energy data at time t;
and constructing a sample set for model training according to the wind energy data sequence.
5. The method of claim 1, wherein the step of training a preset feedback neural network-based wind energy interval prediction model according to the sample set and a preset objective function comprises:
inputting the sample set into a preset wind energy interval prediction model based on a feedback neural network to obtain a prediction interval endpoint value output by the wind energy interval prediction model;
according to the initialized value of the inter-node connection weight of the wind energy interval prediction model, constructing a speed vector and a position vector of an individual at the 1 st moment in a dragonfly algorithm;
acquiring a speed vector and a position vector of the individual at the ith moment, calculating the speed vector and the position vector of the individual at the (i + 1) th moment by using a preset dragonfly algorithm rule, and acquiring the fitness of the individual according to the target function; wherein i is more than or equal to 1 and less than or equal to N-1;
and obtaining the individual with the highest fitness at the Nth moment, and setting the wind energy interval prediction model according to the connection weight value corresponding to the individual to obtain a final wind energy interval prediction model.
6. The method as claimed in claim 5, wherein the step of obtaining the speed vector and the position vector of the individual at the ith time, calculating the speed vector and the position vector of the individual at the (i + 1) th time by using a preset dragonfly algorithm rule, and obtaining the fitness of the individual according to the objective function comprises:
acquiring a speed vector and a position vector of the individual at the ith moment, and acquiring the number of neighbors of the individual according to a preset algorithm;
for the individuals with the neighbor number larger than 0, calculating the speed vector and the position vector of the individual at the (i + 1) th moment by using a preset dragonfly algorithm rule;
for the individual with the number of the neighbors equal to 0, calculating a speed vector and a position vector of the individual at the i +1 th moment by using a Levy flight algorithm;
and obtaining the fitness of the individual according to the objective function.
7. The method of claim 1, wherein the steps of obtaining wind energy data from a pre-set data source, reconstructing the wind energy data into a wind energy data sequence containing wind energy time-varying information based on a delayed embedding theorem, and constructing a sample set for model training from the wind energy data sequence comprise:
acquiring wind energy data from a preset data source, and carrying out normalization processing on the wind energy data to obtain normalized wind energy data with a numerical value interval of [ -1,1 ];
reconstructing the normalized wind energy data into a wind energy data sequence containing wind energy time-varying information based on a delay embedding theorem, and constructing a sample set for model training according to the wind energy data sequence.
8. A wind energy direct interval prediction device based on a feedback neural network, the device comprising:
the training sample set generation module is used for acquiring wind energy data from a preset data source, reconstructing the wind energy data into a wind energy data sequence containing wind energy time-varying information based on a delayed embedding theorem, and constructing a sample set for model training according to the wind energy data sequence;
the wind energy interval prediction model training module is used for training a preset wind energy interval prediction model based on a feedback neural network according to the sample set and a preset target function; the objective function is constructed according to comprehensive evaluation indexes of a prediction interval output by a wind energy interval prediction model;
and the wind energy interval prediction module is used for acquiring a real-time data sequence corresponding to the real-time wind energy data, inputting the real-time data sequence into the trained wind energy interval prediction model and obtaining a wind energy direct interval prediction result of the real-time wind energy data.
9. A computer device comprising a memory and a processor, the memory storing a computer program, wherein the processor implements the steps of the method of any one of claims 1 to 7 when executing the computer program.
10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method of any one of claims 1 to 7.
CN202010602413.2A 2020-06-29 2020-06-29 Wind energy direct interval prediction method and device based on feedback neural network Pending CN111784535A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010602413.2A CN111784535A (en) 2020-06-29 2020-06-29 Wind energy direct interval prediction method and device based on feedback neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010602413.2A CN111784535A (en) 2020-06-29 2020-06-29 Wind energy direct interval prediction method and device based on feedback neural network

Publications (1)

Publication Number Publication Date
CN111784535A true CN111784535A (en) 2020-10-16

Family

ID=72760932

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010602413.2A Pending CN111784535A (en) 2020-06-29 2020-06-29 Wind energy direct interval prediction method and device based on feedback neural network

Country Status (1)

Country Link
CN (1) CN111784535A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112729411A (en) * 2021-01-14 2021-04-30 金陵科技学院 Distributed drug warehouse environment monitoring method based on GA-RNN

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105512745A (en) * 2015-11-05 2016-04-20 天津大学 Wind power section prediction method based on particle swarm-BP neural network
CN106875041A (en) * 2017-01-16 2017-06-20 广东电网有限责任公司揭阳供电局 A kind of short-term wind speed forecasting method

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105512745A (en) * 2015-11-05 2016-04-20 天津大学 Wind power section prediction method based on particle swarm-BP neural network
CN106875041A (en) * 2017-01-16 2017-06-20 广东电网有限责任公司揭阳供电局 A kind of short-term wind speed forecasting method

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
ZHICHAO SHI ET AL.: ""Direct Interval Forecast of Uncertain Wind Power Based on Recurrent Neural Networks"", 《IEEE TRANSACTIONS ON SUSTAINABLE ENERGY》, pages 1177 - 1187 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112729411A (en) * 2021-01-14 2021-04-30 金陵科技学院 Distributed drug warehouse environment monitoring method based on GA-RNN
CN112729411B (en) * 2021-01-14 2022-09-13 金陵科技学院 Distributed drug warehouse environment monitoring method based on GA-RNN

Similar Documents

Publication Publication Date Title
Yuan et al. Monthly runoff forecasting based on LSTM–ALO model
Alencar et al. Hybrid approach combining SARIMA and neural networks for multi-step ahead wind speed forecasting in Brazil
CN106022521B (en) Short-term load prediction method of distributed BP neural network based on Hadoop architecture
CN112910711A (en) Wireless service flow prediction method, device and medium based on self-attention convolutional network
CN114065653A (en) Construction method of power load prediction model and power load prediction method
CN116644970A (en) Photovoltaic power prediction method based on VMD decomposition and lamination deep learning
CN115347571A (en) Photovoltaic power generation power short-term prediction method and device based on transfer learning
CN115329930A (en) Flood process probability forecasting method based on mixed deep learning model
CN111126707A (en) Energy consumption equation construction and energy consumption prediction method and device
CN111784535A (en) Wind energy direct interval prediction method and device based on feedback neural network
CN111612648B (en) Training method and device for photovoltaic power generation prediction model and computer equipment
CN116316617B (en) Multi-station intelligent fusion new energy generation power region prediction method and system
CN114611799B (en) Time sequence neural network new energy output multi-step prediction method based on supervised learning
CN109447843B (en) Photovoltaic output prediction method and device
CN116911178A (en) Method and system for predicting capacity of small and medium-sized reservoirs based on weather forecast
CN116706888A (en) Intelligent power grid load prediction method based on federal learning
KR102515608B1 (en) Photovoltaic forecasting method and system
Sahu et al. Forecasting of Precipitation in India by Different Data Types using Investigation of Radial Basis Function Neural Network Model
CN115630316A (en) Ultrashort-term wind speed prediction method based on improved long-term and short-term memory network
CN112667394A (en) Computer resource utilization rate optimization method
Doudkin et al. Spacecraft Telemetry Time Series Forecasting With Ensembles of Neural Networks
Alcantara et al. Pareto optimal prediction intervals with hypernetworks
CN116227738B (en) Method and system for predicting traffic interval of power grid customer service
CN113344260B (en) Solar photovoltaic power prediction method based on data fusion
CN113609097B (en) Fingerprint library generation method, device, computer equipment and storage medium

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