CN114139783A - Wind power short-term power prediction method and device based on nonlinear weighted combination - Google Patents

Wind power short-term power prediction method and device based on nonlinear weighted combination Download PDF

Info

Publication number
CN114139783A
CN114139783A CN202111388592.5A CN202111388592A CN114139783A CN 114139783 A CN114139783 A CN 114139783A CN 202111388592 A CN202111388592 A CN 202111388592A CN 114139783 A CN114139783 A CN 114139783A
Authority
CN
China
Prior art keywords
sparrow
wind power
prediction model
population
search algorithm
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
CN202111388592.5A
Other languages
Chinese (zh)
Inventor
杨继明
张澈
陈岩磊
曹利蒲
李丹阳
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.)
Beijing Huaneng Xinrui Control Technology Co Ltd
Original Assignee
Beijing Huaneng Xinrui Control Technology Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Beijing Huaneng Xinrui Control Technology Co Ltd filed Critical Beijing Huaneng Xinrui Control Technology Co Ltd
Priority to CN202111388592.5A priority Critical patent/CN114139783A/en
Publication of CN114139783A publication Critical patent/CN114139783A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/004—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
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply

Landscapes

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

Abstract

The invention provides a wind power short-term power prediction method and device based on nonlinear weighted combination. The method comprises the steps of obtaining historical output power of the wind power plant; decomposing historical output power into a plurality of inherent modal functions based on an empirical mode decomposition method; establishing a low-frequency prediction model by using an LSTM neural network according to a plurality of inherent modal functions; establishing a high-frequency prediction model through a deep confidence network by utilizing an improved sparrow search algorithm according to a plurality of inherent modal functions; and integrating the prediction result of the low-frequency prediction model and the prediction result of the high-frequency prediction model by utilizing a neural network established based on an improved sparrow search algorithm-deep confidence network to obtain a final prediction model of the wind power. The invention provides an effective method for extracting wind power characteristics, which can effectively improve the accuracy of wind power short-term power prediction and obtain better prediction precision.

Description

Wind power short-term power prediction method and device based on nonlinear weighted combination
Technical Field
The invention belongs to the technical field of wind power prediction, and particularly relates to a wind power short-term power prediction method and device based on nonlinear weighted combination.
Background
With the development of the energy revolution, renewable energy attracts the attention of the world due to the characteristics of low carbon, no pollution and inexhaustible energy. Among them, wind energy is one of the most potential alternative energy sources in the world due to its advantages of being clean, abundant and renewable. However, the inherent characteristics of randomness, volatility, intermittency and the like of wind energy cause wind power sequences to have high nonlinearity and non-stationarity, which may cause fluctuations of voltage and frequency in a power system, especially in the case of large-scale wind power integration into the power system. Therefore, accurate wind power prediction is crucial to improving the safety of the power system and fully utilizing wind energy resources.
To date, researchers have developed various wind power prediction models, mainly including two major classes, a physical driving model and a data driving model. The physical driving model depends on a numerical weather forecasting system (NWP), and needs to be subjected to fine modeling in the process of converting wind energy into electric energy, so that not only can the calculation burden be increased, but also the short-term wind power prediction is difficult to apply. Unlike the physical drive model, the data-driven model only considers the mapping between the input variables and the target variables, which is very suitable for short-term wind speed/power prediction. However, how to further improve the accuracy of the data-driven model for wind power prediction becomes a problem to be solved urgently.
Disclosure of Invention
The invention aims to at least solve one of the technical problems in the prior art and provides a wind power short-term power prediction method and device based on nonlinear weighted combination.
In one aspect of the invention, a wind power short-term power prediction method based on nonlinear weighted combination is provided, and the method comprises the following steps:
acquiring historical output power of a wind power plant;
decomposing the historical output power into a plurality of inherent modal functions based on an empirical mode decomposition method;
establishing a low-frequency prediction model by using an LSTM neural network according to the plurality of inherent modal functions;
establishing a high-frequency prediction model through a deep confidence network by utilizing an improved sparrow search algorithm according to the plurality of inherent modal functions;
and integrating the prediction result of the low-frequency prediction model and the prediction result of the high-frequency prediction model by utilizing a neural network established based on an improved sparrow search algorithm-deep confidence network to obtain a final prediction model of the wind power.
In some embodiments, the empirical mode decomposition based method decomposes the historical output power into a number of intrinsic mode functions, including:
determining the historical output power as a training sample to be an original sequence X (t), obtaining all local maximum values and local minimum values of X (t), and interpolating by using a cubic spline function to obtain an upper envelope line emax(t) and lower envelope emin(t), obtaining the average value of the two to obtain an envelope average curve m1(t)；
Subtracting the envelope mean curve m from the original sequence X (t)1(t) obtaining a range-like average value sequence h1(t)；
Judgment h1(t) whether or not a condition for the establishment of the natural mode function is satisfied, and if not, h1(t) repeating the above steps as new X (t) until h1(t) satisfies the condition that the natural mode function is established, and in this case, h is1(t) asDecomposed first order intrinsic mode function, and h1(t) separation from X (t) to give r1(t) wherein r1(t)＝X(t)-h1(t)；
By r1(t) replacing X (t) to decompose a new inherent modal function, wherein the inherent modal functions obtained after each decomposition are respectively as follows: r is2(t)＝r1(t)-h2(t),...,rn(t)＝rn-1(t)-hn(t)；
And (3) finishing the decomposition when any one of the following conditions is met: 1) r isn(t) or hn(t) less than a predetermined threshold, 2) rn(t) is a monotonic function from which no longer a natural mode function can be screened;
after the decomposition is finished, all the inherent modal functions and the margin are cumulatively added to obtain
Figure BDA0003367896640000021
In some embodiments, the building a low frequency prediction model using an LSTM neural network according to the plurality of intrinsic mode functions includes:
training the LSTM neural network by using a low-frequency inherent mode function in the inherent mode functions to obtain the low-frequency prediction model, wherein the mathematical expression of the LSTM neural network is as follows:
Figure BDA0003367896640000031
wherein f ist,it,otRespectively the outputs of the forgetting gate, the input gate and the output gate,
Figure BDA0003367896640000033
representing the state of the candidate cell, ct,htAnd ct-1,ht-1Representing the cell state and cell output, x, at the current time t and the previous time t-1, respectivelytIs the input to the LSTM unit and,
Figure BDA0003367896640000032
to predict the output, wf,wi,wc,woW and bf,bi,bc,boB is a weight matrix and a bias vector, o is scalar multiplication, and σ is a sigmoid activation function.
In some embodiments, the establishing a high frequency prediction model through a deep belief network using an improved sparrow search algorithm according to the plurality of intrinsic mode functions includes:
obtaining the optimal weight and bias of the deep belief network by utilizing the improved sparrow search algorithm;
and substituting the optimal weight and the bias into the deep confidence network, and training a high-frequency inherent mode function in the inherent mode functions to obtain the high-frequency prediction model.
In some embodiments, the improved sparrow search algorithm comprises the steps of:
initializing parameters of the improved sparrow search algorithm, wherein the parameters comprise a population scale N, the number pNum of discoverers, the number sNum of sparrows responsible for early warning, the dimension D of a target function, the upper and lower bounds of an initial value, the maximum iteration number and the solving precision;
calculating the fitness value of each sparrow, and selecting the current optimal fitness value and the corresponding position thereof as well as the worst fitness value and the corresponding position thereof;
selecting sparrows randomly from the population as sNum, calculating the fitness value of each sparrow again after one iteration is finished, and updating the positions of discoverers as follows:
Figure BDA0003367896640000041
where t denotes the current number of iterations, j denotes the dimension, itermaxDenotes the maximum number of iterations, α belongs to (0, 1)]Is compared with the random number of (a),
Figure BDA0003367896640000042
belonging to random numbers following a normal distribution, ST representing a safety value，R2Indicates an early warning value, R2ST indicates that the population is in a safe area and the finder can randomly forage, R2ST being more than or equal to represents that predators exist around the population and a safe area needs to be transferred immediately for foraging;
the positions of the update joiners are as follows:
Figure BDA0003367896640000043
in the formula, XPIndicates the optimal location, X, where the finder is currently locatedworseRepresenting the worst position in the current population, A representing a matrix of 1 × d, and each element is randomly assigned with 1 or-1, i > n/2 representing that i participants with low fitness values do not obtain food and need to go to other places to find food;
the scout positions are updated as follows:
Figure BDA0003367896640000044
wherein β represents a step control parameter, and is a random number following a normal distribution with a mean value of 0 and a variance of 1, and XbestIndicating that the sparrow is currently in the safest position, fiRepresenting the fitness value of the current sparrow individual, fiAnd fgRespectively representing the current best and worst fitness values, when fi＞fgWhen f is greater than f, the sparrow is in the edge position and the threat is greatesti＝fgWhen the sparrow in the middle is aware of danger, the sparrow is close to the nearby sparrow as much as possible, gamma is the minimum constant, and the denominator is avoided being zero;
performing elite reverse learning operation on the sparrow population to obtain the latest sparrow population as xi＝k*(ai+bi)-xiWherein x isi∈[ai,bi],k∈[0,1]To obey uniformly distributed random numbers, wherein ai、biRespectively as the maximum position and the minimum position of the x population;
updating the optimal position, the worst position and the fitness value of the whole population according to the current state of the sparrow population, judging whether the maximum iteration times or the solving condition is reached, outputting the optimal value if the maximum iteration times or the solving condition is reached, and repeating the steps after initializing the parameters of the improved sparrow searching algorithm if the maximum iteration times or the solving condition is not reached.
In some embodiments, the obtaining optimal weights and biases for the deep belief network using the improved sparrow search algorithm comprises:
optimizing the weight and the bias of the deep belief network by adopting the improved sparrow search algorithm, and setting the fitness of the optimization process as a root mean square error to obtain the optimal weight and the bias of the deep belief network, wherein the root mean square error is expressed as:
Figure BDA0003367896640000051
in the formula, yiIs true wind power, y'iThe predicted wind power is the wind power predicted by the extreme learning machine, and N is the predicted wind power point number.
In some embodiments, the method further comprises:
obtaining a test sample from the historical output power and validating the final predictive model using the test sample.
In another aspect of the present invention, a wind power short-term power prediction apparatus based on nonlinear weighted combination is provided, the apparatus includes:
the acquisition module is used for acquiring historical output power of the wind power plant;
the decomposition module is used for decomposing the historical output power into a plurality of inherent modal functions based on an empirical mode decomposition method;
the first establishing module is used for establishing a low-frequency prediction model by using an LSTM neural network according to the plurality of intrinsic mode functions;
the second establishing module is used for establishing a high-frequency prediction model through a deep confidence network by utilizing an improved sparrow searching algorithm according to the plurality of inherent modal functions;
and the integration module is used for integrating the prediction result of the low-frequency prediction model and the prediction result of the high-frequency prediction model by utilizing a neural network established based on an improved sparrow search algorithm-deep confidence network so as to obtain the final prediction result of the wind power.
In another aspect of the present invention, an electronic device is provided, including:
one or more processors;
a storage unit for storing one or more programs which, when executed by the one or more processors, enable the one or more processors to implement the method according to the preceding description.
In another aspect of the invention, a computer-readable storage medium is provided, on which a computer program is stored, which computer program, when being executed by a processor, is adapted to carry out the method according to the above.
According to the wind power short-term power prediction method and device based on nonlinear weighted combination, strong fluctuation and non-stationary characteristics of a wind power sequence are considered, an empirical mode decomposition method is used for extracting data characteristics of wind power and decomposing the data characteristics into a plurality of inherent modal functions, decomposition data are used for establishing a prediction model by taking an LSTM neural network and an improved sparrow search algorithm-depth confidence network as subsequence predictors respectively, and therefore an effective method for extracting wind power characteristics is provided; in view of the limitation of a single deep learning method, the sparrow search algorithm is improved, the performance of the algorithm is improved and optimized, the improved sparrow search algorithm is used for optimizing the weight and bias of the deep confidence network, the accuracy of wind power short-term power prediction can be effectively improved, and better prediction accuracy is obtained; in addition, in order to overcome the defect of linear combination, the invention provides a hybrid prediction model based on an improved sparrow search algorithm-deep confidence network nonlinear combination mechanism, so that the accuracy of wind power short-term power prediction is further improved.
Drawings
FIG. 1 is a flow chart of a wind power short-term power prediction method based on nonlinear weighted combination proposed by the present invention;
FIG. 2 is a flow chart of a method of empirical mode decomposition employed in the present invention;
FIG. 3 is a schematic diagram of the structure of the LSTM neural network employed in the present invention;
FIG. 4 is a schematic diagram of a deep belief network employed in the present invention;
FIG. 5 is a schematic structural diagram of a nonlinear combination mechanism based on ESSA-DBN network adopted in the present invention;
FIG. 6 is a schematic structural diagram of wind power short-term power prediction based on nonlinear weighted combination according to the present invention.
Detailed Description
In order to make the technical solutions of the present invention better understood, the present invention will be described in further detail with reference to the accompanying drawings and specific embodiments.
One aspect of the present embodiment, as shown in fig. 1, relates to a wind power short-term power prediction method S100 based on nonlinear weighted combination, where the method S100 includes:
and S110, acquiring historical output power of the wind power plant.
Specifically, in this step, historical output power data of a certain 50MW wind farm may be selected, where the data interval is 15min, the year-round data is divided into four seasons of spring, summer, autumn, and winter, the wind power data of the original four seasons is preprocessed, an interpolation method is used to remove abnormal data and fill up missing data, the first 70% of data of each season is used as a training sample, and the remaining 30% of data is used as a testing sample.
And S120, decomposing the historical output power into a plurality of inherent mode functions based on an empirical mode decomposition method.
In particular, the Empirical Mode Decomposition (EMD) method was proposed by american nationality in 1998, and can be used to analyze nonlinear, non-stationary signal sequences.
The flow of EMD is shown in fig. 2, and as a signal processing method, the empirical mode decomposition method decomposes a multi-component signal into a series of quasi-single signal components, i.e. Intrinsic Mode Functions (IMFs), according to features of different time scales. The decomposed IMF should satisfy two conditions: 1) the time sequence of the zero crossing points in the whole signal is the same as or different from the maximum and minimum point number by one. 2) The mean of the lower envelope fitted by the local maximum points to the upper envelope and the local minimum points is zero. The two conditions are introduced to overcome unnecessary fluctuation phenomenon introduced by asymmetric conditions, so that instantaneous frequency defined by the Hilbert spectrum has practical significance everywhere.
Step S120 may specifically include the following steps:
determining the historical output power as a training sample to be an original sequence X (t), obtaining all local maximum values and local minimum values of X (t), and interpolating by using a cubic spline function to obtain an upper envelope line emax(t) and lower envelope emin(t), obtaining the average value of the two to obtain an envelope average curve m1(t)；
Subtracting the envelope mean curve m from the original sequence X (t)1(t) obtaining a range-like average value sequence h1(t)；
Judgment h1(t) whether or not a condition for the establishment of the natural mode function is satisfied, and if not, h1(t) repeating the above steps as new X (t) until h1(t) satisfies the condition that the natural mode function is established, and in this case, h is1(t) as a decomposed first order natural mode function, and h1(t) separation from X (t) to give r1(t) wherein r1(t)＝X(t)-h1(t)；
By r1(t) replacing X (t) to decompose a new inherent modal function, wherein the inherent modal functions obtained after each decomposition are respectively as follows: r is2(t)＝r1(t)-h2(t),...,rn(t)＝rn-1(t)-hn(t)；
And (3) finishing the decomposition when any one of the following conditions is met: 1) r isn(t) or hn(t) less than a predetermined threshold, 2) rn(t) is a monotonic function from which no longer a natural mode function can be screened;
after the decomposition is finished, all the inherent modal functions and the margin are cumulatively added to obtain
Figure BDA0003367896640000081
And S130, establishing a low-frequency prediction model by using an LSTM neural network according to the plurality of inherent mode functions.
In particular, to alleviate the shortcomings of shallow neural networks, such as local optimization, overfitting, weak generalization performance, etc., deep learning networks have been applied as machine learning in the field of wind power prediction. The LSTM neural network is a special Recurrent Neural Network (RNN), and effectively solves the problems of gradient explosion and gradient disappearance of a classical RNN network. The LSTM neural network consists of "gated structures" (input gate, forget gate, output gate) with short-term memory and cell states with long-term memory, the structure of which is shown in fig. 3.
Step S130 may specifically include the following steps:
training the LSTM neural network by using a low-frequency inherent mode function in the inherent mode functions to obtain the low-frequency prediction model, wherein the mathematical expression of the LSTM neural network is as follows:
Figure BDA0003367896640000091
wherein f ist,it,otRespectively the outputs of the forgetting gate, the input gate and the output gate,
Figure BDA0003367896640000094
representing the state of the candidate cell, ct,htAnd ct-1,ht-1Representing the cell state and cell output, x, at the current time t and the previous time t-1, respectivelytIs the input to the LSTM unit and,
Figure BDA0003367896640000092
to predict the output, wf,wi,wc,woW and bf,bi,bc,boB is the sum of the weight matricesThe deviant vector, -, is a scalar multiplication, and σ is a sigmoid activation function.
And S140, establishing a high-frequency prediction model through a deep confidence network by utilizing an improved sparrow search algorithm according to the plurality of inherent mode functions.
Specifically, step S140 may include the steps of:
obtaining the optimal weight and bias of the deep belief network by utilizing the improved sparrow search algorithm;
and substituting the optimal weight and the bias into the deep confidence network, and training a high-frequency inherent mode function in the inherent mode functions to obtain the high-frequency prediction model.
Deep Belief Networks (DBNs) are a probabilistic generative model consisting of a series of Restricted Boltzmann Machines (RBMs) and logistic regression layers. Wherein, the RMB network consists of a visualization layer trained by data and a hidden layer with feature detection capability. More specifically, the underlying DBN prediction model is unsupervised learning trained using a greedy algorithm, specifically, parameters obtained by training the first layer meta-structure are used as input parameters of the second layer meta-structure. The purpose of this process is to extract sample features more efficiently, except that the top layer of the DBN employs classical bp (back propagation) neural network structure regression prediction. The structure of the DBN is shown in fig. 3. Wherein x ist,
Figure BDA0003367896640000093
Respectively input and predicted output. { w1,w2,...,wnIs the weight matrix, v1And { h1,h2,...,hnDenotes the weight matrix input to the hidden layer and the bias of the hidden layer, respectively.
The Sparrow Search Algorithm (SSA) is a novel group intelligent optimization algorithm which is inspired by the foraging behavior and the anti-predation behavior of sparrows in 2019 by schumann construction and Kai et al, in the SSA algorithm, the sparrows in a group are generally divided into discoverers and jointers, the identities of the discoverers and the jointers are dynamically changed, when one sparrow becomes a discoverer, the other sparrow must become a jointer, and in addition, the initial position of each sparrow and the fitness determined by a fitness function are given. In the SSA algorithm, the magnitude of the fitness value indicates the power of the finder searching for food, and the location of the finder in the iterative process is updated as follows:
Figure BDA0003367896640000101
in the formula: t represents the current iteration number; j represents a dimension; i.e. itermaxRepresenting the maximum number of iterations; alpha is (0, 1)]A random number of (2);
Figure BDA0003367896640000102
random numbers that are subject to a normal distribution; ST represents a security value; r2Indicates an early warning value, R2ST indicates that the population is in a safe area and the finder can randomly forage, R2ST ≧ represents the presence of predators around the population, requiring immediate transfer to a secure area for foraging.
Participants in the population immediately compete for food if they are aware that they are foraging to a better food. If the enrollees compete for success, they can obtain the finder's food, otherwise the finder continues to monitor for foraging. The location of the enrollee is updated as follows:
Figure BDA0003367896640000103
in the above formula: xPIndicating the optimal position where the finder is currently located; xworseRepresenting the worst position in the current population; a represents a 1 × d matrix, and each element is randomly assigned a value of 1 or-1; i > n/2 indicates that i participants with lower fitness values do not obtain food and need to go elsewhere for food.
In addition, predators also exist in the sparrow algorithm, the number of the predators in the whole population is found to be 10% -20%, the initial positions of the sparrows are randomly generated, and the expression form is as follows:
Figure BDA0003367896640000104
in the above formula: beta represents a step size control parameter, and is a random number which follows normal distribution with the mean value of 0 and the variance of 1; xbestIndicating that the sparrow is currently in the safest position; f. ofiRepresenting the fitness value of the current sparrow individual; f. ofiAnd fgRespectively representing the current best and worst fitness values, when fi＞fgWhen the sparrows are in the edge position, the sparrows are threatened most; f. ofi＝fgIndicating that the middle sparrow is aware of the danger and thus is as close as possible to the nearby sparrow; gamma is the minimum constant, avoiding denominators of zero.
The improved sparrow search algorithm (ESSA) introduces an inverse solution in the standard SSA algorithm, which expands the search range of the algorithm much more than before.
In order to effectively improve the convergence speed of the algorithm, firstly, the reverse solution of the current solution is solved, then, individuals with the adaptability values of the original solution larger than the adaptability values of the reverse solution are found out according to the adaptability values, the individuals with the adaptability values of the original solution larger than the adaptability values of the reverse solution are formed into an elite group, then, a new search space is generated in the elite group, and the reverse solution of the individuals with the adaptability values of the original solution smaller than the adaptability values of the reverse solution is solved. When the algorithm finds the optimal solution, the area where the optimal solution is located is necessarily found, and then a reverse solution is generated on the interval defined by the elite group, so that the optimal solution is searched.
Elite inverse Solution (EOS) in N-dimensional space, the Elite individuals x of the current populationbest＝(x1,x2,...,xN) Can be represented as x'best＝(x′1,x′2,...,x′N) Can be defined as xi＝k*(ai+bi)-xiWherein x isi∈[ai,bi],k∈[0,1]To obey uniformly distributed random numbers, wherein ai、biRespectively, the maximum position and the minimum position of the x population, and a plurality of inverse solutions of the elite individual can be generated by utilizing k, wherein the inverse solutions can haveThe mining capacity of the algorithm is effectively improved.
Based on the above, the improved sparrow search algorithm comprises the following steps:
initializing parameters of the improved sparrow search algorithm, wherein the parameters comprise a population scale N, the number pNum of discoverers, the number sNum of sparrows responsible for early warning, the dimension D of a target function, the upper and lower bounds of an initial value, the maximum iteration number and the solving precision;
calculating the fitness value of each sparrow, and selecting the current optimal fitness value and the corresponding position thereof as well as the worst fitness value and the corresponding position thereof;
selecting sparrows randomly from the population as sNum, calculating the fitness value of each sparrow again after one iteration is finished, and updating the positions of discoverers as follows:
Figure BDA0003367896640000111
where t denotes the current number of iterations, j denotes the dimension, itermaxDenotes the maximum number of iterations, α belongs to (0, 1)]Is compared with the random number of (a),
Figure BDA0003367896640000112
belonging to random numbers following a normal distribution, ST representing a safety value, R2Indicates an early warning value, R2ST indicates that the population is in a safe area and the finder can randomly forage, R2ST being more than or equal to represents that predators exist around the population and a safe area needs to be transferred immediately for foraging;
the positions of the update joiners are as follows:
Figure BDA0003367896640000121
in the formula, XPIndicates the optimal location, X, where the finder is currently locatedworseRepresenting the worst position in the current population, A representing a matrix of 1 × d, and each element is randomly assigned with 1 or-1, i > n/2 representing that i participants with low fitness values do not obtain food and need to go to other places to find food;
the scout positions are updated as follows:
Figure BDA0003367896640000122
wherein β represents a step control parameter, and is a random number following a normal distribution with a mean value of 0 and a variance of 1, and XbestIndicating that the sparrow is currently in the safest position, fiRepresenting the fitness value of the current sparrow individual, fiAnd fgRespectively representing the current best and worst fitness values, when fi＞fgWhen f is greater than f, the sparrow is in the edge position and the threat is greatesti＝fgWhen the sparrow in the middle is aware of danger, the sparrow is close to the nearby sparrow as much as possible, gamma is the minimum constant, and the denominator is avoided being zero;
performing elite reverse learning operation on the sparrow population to obtain the latest sparrow population as xi＝k*(ai+bi)-xiWherein x isi∈[ai,bi],k∈[0,1]To obey uniformly distributed random numbers, wherein ai、biRespectively as the maximum position and the minimum position of the x population;
updating the optimal position, the worst position and the fitness value of the whole population according to the current state of the sparrow population, judging whether the maximum iteration times or the solving condition is reached, outputting the optimal value if the maximum iteration times or the solving condition is reached, and repeating the steps after initializing the parameters of the improved sparrow searching algorithm if the maximum iteration times or the solving condition is not reached.
Further, the obtaining of the optimal weight and bias of the deep belief network by using the improved sparrow search algorithm includes:
optimizing the weight and the bias of the deep belief network by adopting the improved sparrow search algorithm, and setting the fitness of the optimization process as a root mean square error to obtain the optimal weight and the bias of the deep belief network, wherein the root mean square error is expressed as:
Figure BDA0003367896640000133
in the formula, yiIs true wind power, y'iThe predicted wind power is the wind power predicted by the extreme learning machine, and N is the predicted wind power point number.
Specifically, in the step, an improved sparrow search algorithm is used in the DBN to optimize the weight and bias of the DBN, and the fitness selected in the optimization process is Root Mean Square Error (RMSE), wherein the RMSE formula is as follows:
Figure BDA0003367896640000132
wherein, yiIs true wind power, y'iThe predicted wind power is the wind power predicted by ELM (Extreme Learning Machine), and N is the predicted wind power point number.
S150, integrating the prediction result of the low-frequency prediction model and the prediction result of the high-frequency prediction model by utilizing a neural network established based on an improved sparrow search algorithm-deep confidence network to obtain a final prediction model of the wind power.
Specifically, most of current researches show that a hybrid prediction model based on a decomposition strategy is superior to a single model, but is influenced by a combination mechanism, and the existing linear combination mechanism ignores the inherent nonlinear relation among models and cannot meet the requirement of high-precision prediction, so that the hybrid model based on the nonlinear combination mechanism is used for aggregating the prediction results of the single model. Specifically, after obtaining the prediction results of two individual models, the predictions of the two models are aggregated through a novel nonlinear combination mechanism based on the ESSA-DBN network. The structure of the nonlinear combined mechanism is shown in fig. 4. In this step, the essence of the nonlinear combination mechanism is a quadratic prediction of the prediction results of the high-frequency component and the low-frequency component of the wind farm output power.
The wind power short-term power prediction method based on nonlinear weighted combination is characterized in that strong volatility and non-stationary characteristics of a wind power sequence are considered, an empirical mode decomposition method is used for extracting data characteristics of wind power and decomposing the data characteristics into a plurality of inherent mode functions, and decomposed data are used for establishing a prediction model by taking an LSTM neural network and an improved sparrow search algorithm-deep confidence network as subsequence predictors respectively, so that an effective method for extracting the wind power characteristics is provided; in view of the limitation of a single deep learning method, the sparrow search algorithm is improved, the performance of the algorithm is improved and optimized, the improved sparrow search algorithm is used for optimizing the weight and bias of the deep confidence network, the accuracy of wind power short-term power prediction can be effectively improved, and better prediction accuracy is obtained; in addition, in order to overcome the defect of linear combination, the invention provides a hybrid prediction model based on an improved sparrow search algorithm-deep confidence network nonlinear combination mechanism, so that the accuracy of wind power short-term power prediction is further improved.
Further, the method S100 may further include the following steps:
obtaining a test sample from the historical output power and validating the final predictive model using the test sample.
In another aspect of the present invention, as shown in fig. 6, a wind power short-term power prediction 100 based on nonlinear weighted combination is provided, and the apparatus 100 is suitable for the method described above. The apparatus 100 comprises:
the obtaining module 110 is configured to obtain historical output power of the wind farm;
a decomposition module 120, configured to decompose the historical output power into a plurality of intrinsic mode functions based on an empirical mode decomposition method;
a first establishing module 130, configured to establish a low-frequency prediction model by using an LSTM neural network according to the plurality of intrinsic mode functions;
the second establishing module 140 is configured to establish a high-frequency prediction model through a deep belief network by using an improved sparrow search algorithm according to the plurality of intrinsic mode functions;
and the integration module 150 is used for integrating the prediction result of the low-frequency prediction model and the prediction result of the high-frequency prediction model by using a neural network established based on an improved sparrow search algorithm-deep confidence network so as to obtain a final prediction result of the wind power.
The wind power short-term power prediction device based on the nonlinear weighted combination is characterized in that strong fluctuation and non-stationary characteristics of a wind power sequence are considered, an empirical mode decomposition method is used for extracting data characteristics of wind power and decomposing the data characteristics into a plurality of inherent modal functions, and decomposed data are used for establishing a prediction model by taking an LSTM neural network and an improved sparrow search algorithm-deep confidence network as subsequence predictors respectively, so that an effective device for extracting the wind power characteristics is provided; in view of the limitation of a single deep learning method, the sparrow search algorithm is improved, the performance of the algorithm is improved and optimized, the improved sparrow search algorithm is used for optimizing the weight and bias of the deep confidence network, the accuracy of wind power short-term power prediction can be effectively improved, and better prediction accuracy is obtained; in addition, in order to overcome the defect of linear combination, the invention provides a hybrid prediction model based on an improved sparrow search algorithm-deep confidence network nonlinear combination mechanism, so that the accuracy of wind power short-term power prediction is further improved.
In another aspect of the present invention, an electronic device is provided, including:
one or more processors;
a storage unit for storing one or more programs which, when executed by the one or more processors, enable the one or more processors to implement the method according to the preceding description.
In another aspect of the invention, a computer-readable storage medium is provided, on which a computer program is stored, which computer program, when being executed by a processor, is adapted to carry out the method according to the above.
The computer readable medium may be included in the apparatus, device, system, or may exist separately.
The computer readable storage medium may be any tangible medium that can contain or store a program, and may be an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, more specific examples of which include, but are not limited to: an electrical connection having one or more wires, a portable computer diskette, a hard disk, an optical fiber, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.
The computer readable storage medium may also include a propagated data signal with computer readable program code embodied therein, for example, in a non-transitory form, such as in a carrier wave or in a carrier wave, wherein the carrier wave is any suitable carrier wave or carrier wave for carrying the program code.
It will be understood that the above embodiments are merely exemplary embodiments taken to illustrate the principles of the present invention, which is not limited thereto. It will be apparent to those skilled in the art that various modifications and improvements can be made without departing from the spirit and substance of the invention, and these modifications and improvements are also considered to be within the scope of the invention.

Claims (10)

1. A wind power short-term power prediction method based on nonlinear weighted combination is characterized by comprising the following steps:
acquiring historical output power of a wind power plant;
decomposing the historical output power into a plurality of inherent modal functions based on an empirical mode decomposition method;
establishing a low-frequency prediction model by using an LSTM neural network according to the plurality of inherent modal functions;
establishing a high-frequency prediction model through a deep confidence network by utilizing an improved sparrow search algorithm according to the plurality of inherent modal functions;
and integrating the prediction result of the low-frequency prediction model and the prediction result of the high-frequency prediction model by utilizing a neural network established based on an improved sparrow search algorithm-deep confidence network to obtain a final prediction model of the wind power.
2. The method of claim 1, wherein decomposing the historical output power into a number of intrinsic mode functions based on an empirical mode decomposition method comprises:
determining the historical output power as a training sample to be an original sequence X (t), obtaining all local maximum values and local minimum values of X (t), and interpolating by using a cubic spline function to obtain an upper envelope line emax(t) and lower envelope emin(t), obtaining the average value of the two to obtain an envelope average curve m1(t)；
Subtracting the envelope mean curve m from the original sequence X (t)1(t) obtaining a range-like average value sequence h1(t)；
Judgment h1(t) whether or not a condition for the establishment of the natural mode function is satisfied, and if not, h1(t) repeating the above steps as new X (t) until h1(t) satisfies the condition that the natural mode function is established, and in this case, h is1(t) as a decomposed first order natural mode function, and h1(t) separation from X (t) to give r1(t) wherein r1(t)＝X(t)-h1(t)；
By r1(t) replacing X (t) to decompose a new inherent modal function, wherein the inherent modal functions obtained after each decomposition are respectively as follows: r is2(t)＝r1(t)-h2(t),...,rn(t)＝rn-1(t)-hn(t)；
And (3) finishing the decomposition when any one of the following conditions is met: 1) r isn(t) or hn(t) less than a predetermined threshold, 2) rn(t) is a monotonic function from which no longer a natural mode function can be screened;
after the decomposition is finished, all the inherent modal functions and the margin are cumulatively added to obtain
Figure FDA0003367896630000021
3. The method according to claim 2, wherein the establishing a low frequency prediction model using an LSTM neural network according to the plurality of intrinsic mode functions comprises:
training the LSTM neural network by using a low-frequency inherent mode function in the inherent mode functions to obtain the low-frequency prediction model, wherein the mathematical expression of the LSTM neural network is as follows:
Figure FDA0003367896630000022
wherein f ist,it,otRespectively the outputs of the forgetting gate, the input gate and the output gate,
Figure FDA0003367896630000023
representing the state of the candidate cell, ct,htAnd ct-1,ht-1Representing the cell state and cell output, x, at the current time t and the previous time t-1, respectivelytIs the input to the LSTM unit and,
Figure FDA0003367896630000024
to predict the output, wf,wi,wc,woW and bf,bi,bc,boB is a weight matrix and a deviation vector,
Figure FDA0003367896630000025
for scalar multiplication, σ is a sigmoid activation function.
4. The method according to claim 3, wherein the establishing a high frequency prediction model through a deep belief network by using an improved sparrow search algorithm according to the plurality of intrinsic mode functions comprises:
obtaining the optimal weight and bias of the deep belief network by utilizing the improved sparrow search algorithm;
and substituting the optimal weight and the bias into the deep confidence network, and training a high-frequency inherent mode function in the inherent mode functions to obtain the high-frequency prediction model.
5. The method of claim 4, wherein the improved sparrow search algorithm comprises the steps of:
initializing parameters of the improved sparrow search algorithm, wherein the parameters comprise a population scale N, the number pNum of discoverers, the number sNum of sparrows responsible for early warning, the dimension D of a target function, the upper and lower bounds of an initial value, the maximum iteration number and the solving precision;
calculating the fitness value of each sparrow, and selecting the current optimal fitness value and the corresponding position thereof as well as the worst fitness value and the corresponding position thereof;
selecting sparrows randomly from the population as sNum, calculating the fitness value of each sparrow again after one iteration is finished, and updating the positions of discoverers as follows:
Figure FDA0003367896630000031
where t denotes the current number of iterations, j denotes the dimension, itermaxDenotes the maximum number of iterations, α belongs to (0, 1)]Is compared with the random number of (a),
Figure FDA0003367896630000032
belonging to random numbers following a normal distribution, ST representing a safety value, R2Indicates an early warning value, R2ST indicates that the population is in a safe area and the finder can randomly forage, R2ST being more than or equal to represents that predators exist around the population and a safe area needs to be transferred immediately for foraging;
the positions of the update joiners are as follows:
Figure FDA0003367896630000033
in the formula, XPIndicates the optimal location, X, where the finder is currently locatedworseRepresenting the worst position in the current population, A represents a 1 × d momentArray, and every element is assigned 1 or-1 randomly, i > n/2 represents that i participants with lower fitness value do not obtain food and need to go to other places to find food;
the scout positions are updated as follows:
Figure FDA0003367896630000034
wherein β represents a step control parameter, and is a random number following a normal distribution with a mean value of 0 and a variance of 1, and XbestIndicating that the sparrow is currently in the safest position, fiRepresenting the fitness value of the current sparrow individual, fiAnd fgRespectively representing the current best and worst fitness values, when fi＞fgWhen f is greater than f, the sparrow is in the edge position and the threat is greatesti＝fgWhen the sparrow in the middle is aware of danger, the sparrow is close to the nearby sparrow as much as possible, gamma is the minimum constant, and the denominator is avoided being zero;
performing elite reverse learning operation on the sparrow population to obtain the latest sparrow population as xi＝k*(ai+bi)-xiWherein x isi∈[ai,bi],k∈[0,1]To obey uniformly distributed random numbers, wherein ai、biRespectively as the maximum position and the minimum position of the x population;
updating the optimal position, the worst position and the fitness value of the whole population according to the current state of the sparrow population, judging whether the maximum iteration times or the solving condition is reached, outputting the optimal value if the maximum iteration times or the solving condition is reached, and repeating the steps after initializing the parameters of the improved sparrow searching algorithm if the maximum iteration times or the solving condition is not reached.
6. The method of claim 5, wherein the obtaining optimal weights and biases for the deep belief network using the improved sparrow search algorithm comprises:
using the improved sparrow search algorithm to search for the sparrowOptimizing the weight and the bias of the depth confidence network, and setting the fitness of the optimization process as a root mean square error to obtain the optimal weight and the bias of the depth confidence network, wherein the root mean square error is expressed as:
Figure FDA0003367896630000041
in the formula, yiIs true wind power, y'iThe predicted wind power is the wind power predicted by the extreme learning machine, and N is the predicted wind power point number.
7. The method according to any one of claims 1 to 6, further comprising:
obtaining a test sample from the historical output power and validating the final predictive model using the test sample.
8. A wind power short-term power prediction device based on nonlinear weighted combination is characterized by comprising the following components:
the acquisition module is used for acquiring historical output power of the wind power plant;
the decomposition module is used for decomposing the historical output power into a plurality of inherent modal functions based on an empirical mode decomposition method;
the first establishing module is used for establishing a low-frequency prediction model by using an LSTM neural network according to the plurality of intrinsic mode functions;
the second establishing module is used for establishing a high-frequency prediction model through a deep confidence network by utilizing an improved sparrow searching algorithm according to the plurality of inherent modal functions;
and the integration module is used for integrating the prediction result of the low-frequency prediction model and the prediction result of the high-frequency prediction model by utilizing a neural network established based on an improved sparrow search algorithm-deep confidence network so as to obtain the final prediction result of the wind power.
9. An electronic device, comprising:
one or more processors;
a storage unit to store one or more programs that, when executed by the one or more processors, cause the one or more processors to implement the method of any of claims 1 to 7.
10. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, is able to carry out a method according to any one of claims 1 to 7.
CN202111388592.5A 2021-11-22 2021-11-22 Wind power short-term power prediction method and device based on nonlinear weighted combination Pending CN114139783A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111388592.5A CN114139783A (en) 2021-11-22 2021-11-22 Wind power short-term power prediction method and device based on nonlinear weighted combination

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111388592.5A CN114139783A (en) 2021-11-22 2021-11-22 Wind power short-term power prediction method and device based on nonlinear weighted combination

Publications (1)

Publication Number Publication Date
CN114139783A true CN114139783A (en) 2022-03-04

Family

ID=80390614

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111388592.5A Pending CN114139783A (en) 2021-11-22 2021-11-22 Wind power short-term power prediction method and device based on nonlinear weighted combination

Country Status (1)

Country Link
CN (1) CN114139783A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114611808A (en) * 2022-03-18 2022-06-10 河海大学 Short-term offshore wind power prediction method based on CEEMDAN-SSA-BilSTM
CN116780776A (en) * 2023-06-29 2023-09-19 淮阴工学院 Chemical industry park photovoltaic monitoring system and method based on improved sparrow algorithm
CN118551467A (en) * 2024-07-30 2024-08-27 湖南科技大学 Structure surface wind load prediction method and device

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114611808A (en) * 2022-03-18 2022-06-10 河海大学 Short-term offshore wind power prediction method based on CEEMDAN-SSA-BilSTM
CN116780776A (en) * 2023-06-29 2023-09-19 淮阴工学院 Chemical industry park photovoltaic monitoring system and method based on improved sparrow algorithm
CN118551467A (en) * 2024-07-30 2024-08-27 湖南科技大学 Structure surface wind load prediction method and device

Similar Documents

Publication Publication Date Title
JP5888640B2 (en) Photovoltaic power generation prediction apparatus, solar power generation prediction method, and solar power generation prediction program
CN114139783A (en) Wind power short-term power prediction method and device based on nonlinear weighted combination
CN108985515B (en) New energy output prediction method and system based on independent cyclic neural network
CN107944648B (en) Large ship speed and oil consumption rate prediction method
Li et al. Multi-step ahead wind speed forecasting approach coupling maximal overlap discrete wavelet transform, improved grey wolf optimization algorithm and long short-term memory
CN112100911B (en) Solar radiation prediction method based on depth BILSTM
CN111461463A (en) Short-term load prediction method, system and equipment based on TCN-BP
CN116644970A (en) Photovoltaic power prediction method based on VMD decomposition and lamination deep learning
CN114792156A (en) Photovoltaic output power prediction method and system based on curve characteristic index clustering
CN111242355A (en) Photovoltaic probability prediction method and system based on Bayesian neural network
CN115907131B (en) Method and system for constructing electric heating load prediction model in northern area
CN115099461A (en) Solar radiation prediction method and system based on double-branch feature extraction
CN111985719A (en) Power load prediction method based on improved long-term and short-term memory network
CN114897204A (en) Method and device for predicting short-term wind speed of offshore wind farm
CN113128666A (en) Mo-S-LSTMs model-based time series multi-step prediction method
CN116799796A (en) Photovoltaic power generation power prediction method, device, equipment and medium
CN115713144A (en) Short-term wind speed multi-step prediction method based on combined CGRU model
CN116014722A (en) Sub-solar photovoltaic power generation prediction method and system based on seasonal decomposition and convolution network
CN116522594A (en) Time self-adaptive transient stability prediction method and device based on convolutional neural network
CN115766125A (en) Network flow prediction method based on LSTM and generation countermeasure network
CN116245019A (en) Load prediction method, system, device and storage medium based on Bagging sampling and improved random forest algorithm
Cao et al. Fast and explainable warm-start point learning for AC Optimal Power Flow using decision tree
CN118040678A (en) Short-term offshore wind power combination prediction method
CN116632834A (en) Short-term power load prediction method based on SSA-BiGRU-Attention
CN114239945B (en) Short-term power load prediction method, device, 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