CN110322050A - A kind of wind energy resources compensation data method - Google Patents
A kind of wind energy resources compensation data method Download PDFInfo
- Publication number
- CN110322050A CN110322050A CN201910481772.4A CN201910481772A CN110322050A CN 110322050 A CN110322050 A CN 110322050A CN 201910481772 A CN201910481772 A CN 201910481772A CN 110322050 A CN110322050 A CN 110322050A
- Authority
- CN
- China
- Prior art keywords
- formula
- whale
- neural network
- error
- compensation data
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 39
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 47
- 241000283153 Cetacea Species 0.000 claims abstract description 45
- 238000013528 artificial neural network Methods 0.000 claims abstract description 45
- 238000005457 optimization Methods 0.000 claims abstract description 29
- 238000013499 data model Methods 0.000 claims abstract description 4
- 230000007246 mechanism Effects 0.000 claims description 9
- 238000011478 gradient descent method Methods 0.000 claims description 7
- 230000008569 process Effects 0.000 claims description 7
- DRUQKRWRXOUEGS-NGERZBJRSA-N Samin Chemical compound C1=C2OCOC2=CC([C@H]2OC[C@H]3[C@@H]2CO[C@@H]3O)=C1 DRUQKRWRXOUEGS-NGERZBJRSA-N 0.000 claims description 4
- 238000012360 testing method Methods 0.000 claims description 3
- 230000008901 benefit Effects 0.000 abstract description 7
- 230000006870 function Effects 0.000 description 18
- 238000012549 training Methods 0.000 description 12
- 239000003795 chemical substances by application Substances 0.000 description 9
- 210000002569 neuron Anatomy 0.000 description 9
- 238000003062 neural network model Methods 0.000 description 7
- 241001456553 Chanodichthys dabryi Species 0.000 description 5
- 238000010248 power generation Methods 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 230000004913 activation Effects 0.000 description 3
- 230000005611 electricity Effects 0.000 description 3
- 238000010606 normalization Methods 0.000 description 3
- 238000004088 simulation Methods 0.000 description 3
- 230000003466 anti-cipated effect Effects 0.000 description 2
- 230000006399 behavior Effects 0.000 description 2
- 210000004027 cell Anatomy 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000000977 initiatory effect Effects 0.000 description 2
- 241001269238 Data Species 0.000 description 1
- 241000167880 Hirundinidae Species 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000005684 electric field Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000012804 iterative process Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000002715 modification method Methods 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000001902 propagating effect Effects 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Classifications
-
- 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
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E40/00—Technologies for an efficient electrical power generation, transmission or distribution
- Y02E40/70—Smart grids as climate change mitigation technology in the energy generation sector
-
- 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)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- Strategic Management (AREA)
- General Health & Medical Sciences (AREA)
- Human Resources & Organizations (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- Tourism & Hospitality (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Primary Health Care (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 discloses a kind of wind energy resources compensation data methods, are optimized using whale optimization algorithm to BP neural network, construct compensation data model, by inputting meteorological element, can be obtained corresponding output data, to compensate the missing data during observation.Using whale optimization algorithm, fast convergence rate, precision in terms of Optimum search are high and easily jump out the advantage of local optimum by the present invention, by way of carrying out whale optimizing while BP neural network updates weight threshold, solve the problems, such as that precision of prediction is low caused by BP neural network easily falls into local optimum.Meanwhile the problem slow for convergence rate present in BP neural network, the whale optimization algorithm dynamic convergence factor is improved, algorithm reliability and precision are improved.
Description
Technical field
The present invention relates to a kind of wind energy resources compensation data methods.
Background technique
It in the measurement process of wind energy resources, often will appear shortage of data problem, survey wind compared to ground, ocean is surveyed
Wind, upper air wind measuring, Complex Mountain survey wind energy resources data measured by the survey wind mode such as wind and are difficult to supplement immediately.And it is common pre-
Method of determining and calculating, such as BP neural network prediction algorithm, due to its initial weight have randomness, cause algorithm global reliability, when
Effect property and accuracy are poor, it is difficult to provide accurate Data safeguard.
Such as the Chinese invention patent of Patent No. 201210460427.0, a kind of wind-powered electricity generation based on BP neural network is provided
Field short term power prediction technique, core concept are as follows: using wind speed, wind direction, atmospheric density and relative humidity as BP neural network
The power generation output power of input, wind power plant is exported as BP neural network, utilizes the wind of BP neural network prediction certain time period
The power generation output power of electric field.Specific work steps is as follows: what a. obtained wind power plant location includes that wind speed, wind direction and air are close
It the historical record of the meteorological element data of degree and records opposite wind power plant with each and generates electricity output power；B. wherein, by wind speed,
Wind direction and atmospheric density are modified to wind speed, wind direction and atmospheric density at wind-powered machine unit hub, to generate revised meteorology
Factor data；C. BP neural network is inputted using revised meteorological element data as input data, it will be with each meteorological element
The corresponding wind power plant power generation output power of data is trained BP neural network as the output of BP neural network；D. according to number
It is worth weather forecast and obtains wind power plant location in the meteorological element number including wind speed, wind direction and atmospheric density of predicted time section
According to, and wind speed, wind direction and atmospheric density are modified to wind speed, wind direction and atmospheric density at wind-powered machine unit hub, to generate
Revised meteorological element data；E. the BP by the resulting revised meteorological element data input of step d after step c training
Neural network, the data of BP neural network output are the power generation output power of the wind power plant of the predicted time section.Wheel hub
Relative humidity, to input BP neural network as input data.
Entire technical solution is still that the power generation output power prediction of wind power plant is carried out using BP neural network, and essence is still
The scope for belonging to BP neural network, by BP neural network algorithm with wind power plant short term power prediction on, since this method is deposited
The problem of easily falling into locally optimal solution, lead to that precision of prediction is low, prediction result is unreliable.Based on this, the invention proposes one
Kind wind energy resources compensation data method, it is intended to carry out accurate compensation for wind energy resources missing data.
Summary of the invention
The purpose of the present invention is to propose to a kind of wind energy resources compensation data methods, using whale optimization algorithm in Optimum search
Aspect fast convergence rate, precision are high and easily jump out the advantage of local optimum, while updating weight threshold by BP neural network
The mode for carrying out whale optimizing solves the problems, such as easily to fall into precision of prediction caused by local optimum as BP neural network low.Together
When, for the problem that convergence rate present in BP neural network is slow, using whale optimization algorithm in terms of global optimizing ability
Advantage, improve the whale optimization algorithm dynamic convergence factor, improve algorithm reliability.
For this purpose, passing through the BP mind based on whale optimization algorithm the present invention provides a kind of wind energy resources compensation data method
Through network model, compensation data model is constructed, to prevent BP neural network from falling into local optimum, so that the meteorology according to input is wanted
Prime model can be obtained corresponding output data, to compensate the missing data during observation.
The invention has the advantages that using whale optimization algorithm, fast convergence rate, precision are high in terms of Optimum search and easily jump
The advantage of local optimum out, BP neural network update weight threshold while carry out whale optimizing by way of, solve due to
BP neural network easily falls into the low problem of precision of prediction caused by local optimum.Meanwhile it being received for present in BP neural network
Slow-footed problem is held back, using advantage of the whale optimization algorithm in terms of global optimizing ability, improves whale optimization algorithm dynamic
Convergence factor improves algorithm reliability and precision.
The present invention is described in further details below with reference to attached drawing.
Detailed description of the invention
Fig. 1 is WOA-BP algorithm flow chart.
Fig. 2 is WOA-BP prediction-error image.
Fig. 3 is the training of WOA-BP wind speed and test comparison chart.
Fig. 4 is for typical three layers of BP neural network structure chart.
Fig. 5 is BP neural network algorithm flow chart.
Fig. 6 whale optimization algorithm flow chart.
Fig. 7 contractile ring is around mechanism.
Fig. 8 helical position update mechanism.
Specific embodiment
It present embodiments provides a kind of wind energy resources compensation data method and obtains this method by way of experiment simulation
Have many advantages, such as highly reliable, prediction result is accurate, and solves BP neural network local minimization and convergence speed of the algorithm
Slow problem, experimental data are as follows.
The China national grade earth station hour Value Data provided using China Meteorological Administration, using Sichuan Province's Panzhihua
City on March 16th, 2019 to March 18 is every 13 kinds of data such as 3 hours atmospheric pressures, relative humidity, mean temperature as mould
Type input, and obtain this area's moment actual wind speed data simultaneously as model output and carry out modeling using preceding 128 groups of data
Training pattern.
Using 13 kinds of data such as atmospheric pressure, relative humidity, mean temperature of the 20th group of data as mode input, after compensation
The air speed data of 10 groups of data.The data obtained is tested as shown in Fig. 2-3 and table 1.
1 WOA-BP part of neural network wind speed of table trains table
13 kinds of data such as the atmospheric pressure of 10 groups of data, relative humidity, mean temperature are corresponded to as mode input, output afterwards
Air speed data.Experimental data shows that compensation data method of the present invention has high reliability, and error is low, and performance is good, can be wind-powered electricity generation
The siteselecting planning of field provides authentic data basis.
The wind energy resources compensation data method, mainly passes through the BP neural network model based on whale optimization algorithm, structure
Compensation data model is built, to prevent BP neural network from falling into local optimum, to can obtain according to the meteorological element model of input
Corresponding output data is obtained, to compensate the missing data during observation.
To solve the above problems, a kind of wind energy resources compensation data method is present embodiments provided, it is main to realize process such as
Shown in Fig. 1, main flow is: starting, initiation parameter establishes cell array, calculates fitness function (error), error in judgement
Whether optimal solution is less than, if it is, using the error as current optimal solution, and the parameters such as A, C, P are calculated, if it is not, throwing away
Continue to calculate the parameters such as A, C, P, carries out the traversal iteration of W1, B1, W2, B2 later；
Judge P whether less than 0.5, if so, calculate A absolute value whether be greater than 1, if it is, circulating type search with
Expand search range, and export current best initial weights threshold value, if the absolute value of A is no more than 1, search for predation, to optimal
Solution more new search, until exporting current optimal solution weight threshold；Then, current optimal solution is stored in cell array, recalculated
Current optimal solution updates W1, B1, W2, B2 using gradient descent method；Judge whether to meet maximum number of iterations at this time, if so,
Current optimal solution and error are then exported, is terminated；If it is not, then recalculating fitness function, and circulation is sequentially carried out, until
Terminate；
If whether P is not less than 0.5, spiral bubble net search pattern expands local search, until exporting current optimal power
It is worth threshold value.
Specific operation process is as follows:
Step 1: data normalization
The meteorological datas such as atmospheric pressure, mean temperature, relative humidity are normalized using premnmx normalized function
Processing is convenient for calculating, for example, input training sample atmospheric pressure is 761.7 hundred pas, mean temperature is -2.6 degrees Celsius, relatively wet
Degree is 65%, and training sample output wind speed is 0.7m/s, and generating one group of new input data SamIn is [1, -1, -0.8231]；
Wind speed is normalized, generating another group of new output data tnOut is [0.7].Wherein SamIn and tnOut can be with
The variation of sample size and change；
Step 2: initializing neural network parameter according to normalization result, i.e. initiation parameter shown in Fig. 1 establishes member
Born of the same parents' array.
The noise that numerical value is 0.01 is added to the output data after normalization, in order to prevent BP neural network excessively quasi-
It closes, the specific method is as follows:
A, noise is added with normalized output data, is denoted as SamOut, as shown in formula 1；
SamOut=tnOut+Noise formula 1
B, the noise Noise is multiplied to obtain by random number with noise intensity, as shown in formula 2；
Noise=NoiseVar*Y formula 2
C, the noise intensity NoiseVar is usually 0.01；
D, the random number is obtained by formula 3.
Y=rand () formula 3
Number according to step 1 initialization data establishes BP neural network model, number InDim (this of input layer
3) example is, the number HiddenUnitNum (this example 8) of hidden layer node, and exporting the number OutDim of node layer, (this example is
1)；
According to the BP neural network model of aforementioned foundation, initialization model parameter is specifically included:
The BP neural network model learning rate is dynamical learning rate, as shown in formula 4.
Lr=1-t/max_iteration formula 4
BP neural network model uses activation primitive for sigmoid function, as shown in formula 5.
Step 4: initialization whale optimization algorithm parameter
The present invention is for compensation data problem to the linear convergence factor of whale optimization algorithmIt improves, it is improved
The dynamic convergence factorAs shown in formula 6.
T is current iteration number, and max_iteration is maximum number of iterations.
The convergence factor for improving whale optimization algorithmAs the number of iterations variation is as shown in formula 7.
T is current iteration number, and max_iteration is maximum number of iterations.
The weight threshold initial method of the whale optimization algorithm, specific as follows:
Calculating to the whale optimization algorithm weight and threshold value is updated between input layer and hidden layer using following formula
Weight ω1, threshold value B between input layer and hidden layer1, weight ω between output layer and hidden layer2, output layer and implicit
Threshold value B between layer2, the formula as shown in formula 8-11 is respectively adopted and is initialized.
ω1=0.5*rand (HiddenUnitNum, InDim)) -0.1 formula 8
B1- 0.1 formula 9 of=0.5*rand (HiddenUnitNum, 1)
ω2- 0.1 formula 10 of=0.5*rand (OutDim, HiddenUnitNum)
B2- 0.1 formula 11 of=0.5*rand (OutDim, 1)
Step 5: calculating fitness function (error) and current best search agent according to BP neural network model
Fitness function (error function) is as shown in formula 11-14.
HiddenOut=sigmoid (ω1*SamIn+repmat(B1, 1, SamNum)) formula 12
NetworkOut=ω2*HiddenOut+repmat(B2, 1, SamNum) and formula 13
Error=SamOut-NetworkOut formula 14
SSE=sumsqr (Error) formula 15
HiddenOut is hidden layer node output, and NetworkOut is output node layer output, and Error is network output
Error between value and test data, SSE is error of sum square.
Step 6: updating weight threshold using the whale optimization algorithm
By in the BP neural network one group of weight and threshold value be denoted as a search agent, i.e. a whale；
Using the error sum of squares SSE as fitness function, the minimum fitness value of whale population is calculated；
The fitness value of every whale is compared with the optimal whale of individual, obtains current optimal whale position；
Whale population position is updated around mechanism and helical position update mechanism according to contractile ring.
Step 6: updating weight threshold using gradient descent method
Using gradient descent method (neural network backpropagation), the weight threshold of current optimal whale position, gradient are updated
Descent method weight threshold more new formula is as shown in formula 16-21.
Δ2=Error formula 16
Δ1=ω2 T*Δ2* HiddenOut* (1-HiddenOut) formula 17
In weight threshold more new formula, ω1' it is ω1Weight after updated, ω2' it is ω2Weight after updated, B1'
For B1Threshold value after updated, B2' it is B2Threshold value after updated.
Step 5: training terminates judgement
Trained termination condition is as follows:
Target error reaches preset value；
The number of iterations reaches predetermined number of times.
If so, model training pressure terminates.If it is not, then returning, " step 2 calculates fitness function and current best search
Agency ", continues to train
Step 6: output training pattern carries out compensation data
After method reaches termination condition, predictive compensation model is generated.
Using data such as any atmospheric pressure, relative humidity, mean temperatures as the predictive compensation mode input, can obtain
Corresponding air speed data output is obtained, to compensate the missing data during observation.
The wind energy resources compensation data method that the above various embodiments provides for ease of understanding, the present embodiment is just related to be known substantially
Knowledge is introduced in conjunction with attached drawing:
Fig. 4 show typical three layers of BP neural network structure chart, and its essence is a kind of calculations for solving peak optimizating network weight
Method.The objective function (cost function) of optimization problem is the error letter that network exports that error is constituted between desired output at this time
Number, variable to be optimized is exactly all weights and threshold value in network, and BP neural network is to solve energy using gradient descent method
Error function is enough set to reach the modification method of minimum network weight and threshold value, BP neural network algorithm flow chart is as shown in Figure 5.
Step 1: initialization weight threshold
Network connection weight between every two neuron is initialized to the random number of a very little, while each nerve
Member has a bias, is also initialized to a random number.To each input sample X, handled according to step 2.
Step 2: propagating input (calculating the input and output of hidden layer, each neuron of output layer) forward
The input layer of network is provided according to training sample X, the output of each neuron is obtained by calculation, such as 22 institute of formula
Show.
Note: ωijIt is by the network weight of upper one layer of unit i to this unit j；OiIt is the output of upper one layer of unit i；A
For the biasing of this unit, for serving as threshold value, thus it is possible to vary the activity of unit.
It can be seen that, the output of neuron j depends on its total input S from the equations abovej=∑ ωij*Oi+Ai, then lead to
Cross activation primitiveFinal output is obtained, this activation primitive is known as sigmoid function, can will be biggish
Input value is mapped as a value between section 0~1, since the function is nonlinear and can be micro-, but also BP is refreshing
The classification problem of linearly inseparable can be modeled through network algorithm, expand its application range significantly.
Step 3: back-propagation
By step 2, reality output finally is obtained in output layer, it can be by obtaining each output compared with anticipated output
The error of unit: (TjIt is the anticipated output of output unit), as shown in formula 23.
Ej=Oj(1-Oj)(Tj-Oj) formula 23
Obtained error needs to propagate from back to front, the error of previous layer unit can by with it connect behind one layer
All units error calculation obtained by, as shown in formula 24.
Ej=Oj(1-Oj)∑kωjkEkFormula 24
Successively obtain the last one hidden layer to first each neuron of hidden layer error.
Step 4: network weight and neuron biasing adjustment (weight threshold of amendment hidden layer, output layer)
Weight is adjusted since the connection weight of input layer and the first hidden layer, is successively carried out backward, each connection weight
It is adjusted as follows, as shown in formula 25.
ωij=ωij+Δωij=ωij+(lr)OiEjFormula 25
Note: wherein lr is dynamical learning rate, as shown in formula 4.
Lr=1-t/max_iteration formula 4
The method of adjustment of neuron biasing is to carry out the update as shown in formula 26 to each neuron.
θj=θj+Δθj=θj+(lr)EjFormula 26
Step 5: judgement terminates
For each sample, if final output error is less than acceptable range or the number of iterations has reached certain
Threshold value, then choose next sample, go to step 2 and continue to execute again；Otherwise, the number of iterations adds 1, then turns to step 2
Current sample is continued to use to be trained.
BP neural network has self study and adaptive ability, fault-tolerant ability and non-linear mapping capability well.But it is same
The sample algorithm has certain defect, and wherein major defect includes following two points:
1, local minimization problem: in terms of mathematical angle, traditional BP neural network is a kind of optimization side of local search
Method, it is to solve a complex nonlinear problem, and the weight of network is by gradually carrying out along the direction of minor betterment
Adjustment, algorithm can be made to fall into local extremum, weight convergence to local minimum point, so as to cause network training failure in this way.Add
Upper BP neural network is very sensitive to initial network weight, with different weights initialisation networks, often converges on difference
Local minimum, this is also that each training obtains the basic reason of Different Results.
2, convergence speed of the algorithm is slow: since BP neural network algorithm is essentially gradient descent method, what it to be optimized
Objective function be it is extremely complex, therefore, necessarily will appear " zigzag phenomenon ", this makes BP algorithm inefficient；Again due to optimization
Objective function it is very complicated, some flat regions will necessarily occur, in these regions in the case where neuron is exported close to 0 or 1
Interior, weight changes very little, and training process is made almost to pause；It, cannot in order to make network execute BP algorithm in BP neural network model
The step-length of each iteration is sought using traditional linear search method, and must assign in advance network the update rule of step-length, it is this
Method also can cause algorithm inefficient.It is above various, result in the slow phenomenon of BP neural network algorithm the convergence speed.
(2) whale optimization algorithm
Whale optimization algorithm is nature heuristic algorithm, imitates the predation of humpback, and humpback is from about seabed depths
Make spiral gesture to move about upwards and the bubble that spues, when the bubble finally to spue and first bubble to spue rise to water simultaneously
When face, a kind of bubble net is formed, prey is tightly surrounded as huge net, and prey is forced to net center, it
Just big mouth is almost erectly opened in bubble circle, swallows the prey of net collection.Algorithm flow chart is as shown in Figure 6.
Specifically:
Current minimal error is calculated using the fitness function of individual, and by the weight threshold of minimal error (best whale
Fish) it is used as the best search agent of epicycle iteration, it scans for acting on behalf of individual location updating, update method using whale optimization algorithm
It is as follows:
A. the prey stage is surrounded:
Humpback can identify the position of prey and surround them.Since the position of the optimal design in search space is not
A priori known, therefore WOA algorithm assumes that current optimal candidate solution is target prey or close to optimal.Defining best search
After agency, other search agents will be attempted to update its position to best search agent.This behavior is as shown in formula 27,28.
In equation,Distance for search agent away from best search agent, t are current iteration number,WithFor coefficient,
X*Position vector for the optimum solution obtained at present, X are position vector, | | to take absolute value.Here it is worth noting that such as
Fruit has more excellent solution, then X*It will update in each iteration.VectorWithCalculation formula as shown in formula 29,30.
B. bubble net phase of the attack:
Mathematical simulation has been carried out for the behavior of the bubble net attack prey of humpback, has devised two methods:
Contractile ring is around mechanism:
It will be gradually decreased to 0 in an iterative process.Therefore between [- a, a] in for A be arranged random value, search agent
New position can agency home position and currently any position between the position most preferably acted on behalf of defines.Fig. 7 is shown
It is arrived by (X, Y)The possible position that 0≤A≤1 obtains in two-dimensional space.
Helical position update mechanism
Helical position renewal process is calculated first positioned at the whale of (X, Y) and positioned at (X*,Y*) the distance between prey.
Then a helical position renewal equation is established between whale and the position of prey, as shown in figure 8, the spiral of simulation humpback
Shape is moved as shown in formula 31.
In motion process,Indicate (the current best search of the distance between i-th whale and prey
Agency), b is the constant for defining logarithmic spiral shape, and l is the random number in [- 1,1].
Contractile ring updates during loop iteration around mechanism and helical position, is carried out respectively with 0.5 probability, such as public
Shown in formula 32.
It is completed after acting on behalf of searching position update by surrounding prey stage and bubble net phase of the attack, recalculates adaptation
Function is spent, current best search agent is updated.
Claims (4)
1. a kind of wind energy resources compensation data method characterized by comprising
BP neural network is optimized by improving whale optimization algorithm, neural network weight threshold value after being optimized；Root
Neural network weight threshold value constructs compensation data model after according to the optimization, is wanted based on the compensation data mode input meteorology
Element, to obtain corresponding output data, to obtain the missing data during compensation observation.
2. wind energy resources compensation data method as described in claim 1, which is characterized in that the improvement whale optimization algorithm,
Specifically:
To the linear convergence factor of whale optimization algorithmIt improves, as shown in formula 6,
T is current iteration number, and max_iteration is maximum number of iterations.
3. wind energy resources compensation data method as claimed in claim 2, which is characterized in that the improvement whale optimization algorithm is also wrapped
Include following process:
Using error sum of squares SSE as fitness function, the minimum fitness value of whale population is calculated；
The fitness value of every whale is compared with the optimal whale of individual, obtains current optimal whale position；
The whale population location updating is completed around mechanism and helical position update mechanism according to contractile ring；
Using gradient descent method, the weight threshold of current optimal whale position is updated, gradient descent method updates weight threshold, updates
Formula as shown in formula 16-21,
Δ2=Error formula 16
Δ1=ω2 T*Δ2* HiddenOut* (1-HiddenOut) formula 17
In weight threshold more new formula, ω1' it is ω1Weight after updated, ω2' it is ω2Weight after updated, B1' it is B1
Threshold value after updated, B2' it is B2Threshold value after updated.
4. wind energy resources compensation data method as claimed in claim 3, which is characterized in that specific as follows: the error sum of squares
The conduct fitness function value of SSE, it is specific to calculate as shown in formula 12-15,
HiddenOut=sigmoid (ω1*SamIn+repmat(B1, 1, SamNum)) formula 12
NetworkOut=ω2*HiddenOut+repmat(B2, 1, SamNum) and formula 13
Error=SamOut-NetworkOut formula 14
SSE=sumsqr (Error) formula 15
HiddenOut be hidden layer node output, NetworkOut be output node layer output, Error be network output valve with
Error between test data, SSE are the quadratic sum of error.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910481772.4A CN110322050B (en) | 2019-06-04 | 2019-06-04 | Wind energy resource data compensation method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910481772.4A CN110322050B (en) | 2019-06-04 | 2019-06-04 | Wind energy resource data compensation method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110322050A true CN110322050A (en) | 2019-10-11 |
| CN110322050B CN110322050B (en) | 2023-04-07 |
Family
ID=68119396
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910481772.4A Active CN110322050B (en) | 2019-06-04 | 2019-06-04 | Wind energy resource data compensation method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110322050B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111948283A (en) * | 2020-07-07 | 2020-11-17 | 上海交通大学 | System and method for detecting oxidation stability of transformer oil |
| CN113240307A (en) * | 2021-05-24 | 2021-08-10 | 上海电力大学 | Power system economic dispatching method based on improved differential evolution algorithm |
Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP0295424A2 (en) * | 1987-06-08 | 1988-12-21 | International Business Machines Corporation | Method for managing subpage concurrency control and partial transaction rollback in a transaction-oriented system of the write-ahead logging type |
| US20090287620A1 (en) * | 2008-05-16 | 2009-11-19 | Samsung Electronics Co., Ltd. | System and method for object detection and classification with multiple threshold adaptive boosting |
| US8407020B1 (en) * | 2010-09-28 | 2013-03-26 | The United States Of America As Represented By The Secretary Of The Navy | Fast method to search for linear frequency-modulated signals |
| US20160117587A1 (en) * | 2014-10-27 | 2016-04-28 | Zhicheng Yan | Hierarchical deep convolutional neural network for image classification |
| CN107292051A (en) * | 2017-07-07 | 2017-10-24 | 湘潭大学 | A kind of carbide chip chemically-mechanicapolish polishes the Forecasting Methodology of surface roughness |
| CN107911300A (en) * | 2017-10-25 | 2018-04-13 | 西南交通大学 | Multicast routing optimization method based on whale algorithm and its application on Spark platforms |
| CN108021658A (en) * | 2017-12-01 | 2018-05-11 | 湖北工业大学 | A kind of big data intelligent search method and system based on whale optimization algorithm |
| CN108205698A (en) * | 2017-06-15 | 2018-06-26 | 广东工业大学 | A kind of cloud resource load predicting method based on the double string whale optimization algorithms of just remaining chaos |
| CN108830431A (en) * | 2018-08-03 | 2018-11-16 | 广东工业大学 | A kind of Electricity price forecasting solution and relevant apparatus based on whale optimization algorithm |
| CN109101738A (en) * | 2018-08-24 | 2018-12-28 | 河北工业大学 | A kind of IGBT module degree of aging appraisal procedure |
| CN109214449A (en) * | 2018-08-28 | 2019-01-15 | 华北电力大学 | A kind of electric grid investment needing forecasting method |
| CN109345005A (en) * | 2018-09-12 | 2019-02-15 | 中国电力科学研究院有限公司 | A kind of integrated energy system multidimensional optimization method based on improvement whale algorithm |
| CN109765893A (en) * | 2019-01-17 | 2019-05-17 | 重庆邮电大学 | Method for planning path for mobile robot based on whale optimization algorithm |
-
2019
- 2019-06-04 CN CN201910481772.4A patent/CN110322050B/en active Active
Patent Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP0295424A2 (en) * | 1987-06-08 | 1988-12-21 | International Business Machines Corporation | Method for managing subpage concurrency control and partial transaction rollback in a transaction-oriented system of the write-ahead logging type |
| US20090287620A1 (en) * | 2008-05-16 | 2009-11-19 | Samsung Electronics Co., Ltd. | System and method for object detection and classification with multiple threshold adaptive boosting |
| US8407020B1 (en) * | 2010-09-28 | 2013-03-26 | The United States Of America As Represented By The Secretary Of The Navy | Fast method to search for linear frequency-modulated signals |
| US20160117587A1 (en) * | 2014-10-27 | 2016-04-28 | Zhicheng Yan | Hierarchical deep convolutional neural network for image classification |
| CN108205698A (en) * | 2017-06-15 | 2018-06-26 | 广东工业大学 | A kind of cloud resource load predicting method based on the double string whale optimization algorithms of just remaining chaos |
| CN107292051A (en) * | 2017-07-07 | 2017-10-24 | 湘潭大学 | A kind of carbide chip chemically-mechanicapolish polishes the Forecasting Methodology of surface roughness |
| CN107911300A (en) * | 2017-10-25 | 2018-04-13 | 西南交通大学 | Multicast routing optimization method based on whale algorithm and its application on Spark platforms |
| CN108021658A (en) * | 2017-12-01 | 2018-05-11 | 湖北工业大学 | A kind of big data intelligent search method and system based on whale optimization algorithm |
| CN108830431A (en) * | 2018-08-03 | 2018-11-16 | 广东工业大学 | A kind of Electricity price forecasting solution and relevant apparatus based on whale optimization algorithm |
| CN109101738A (en) * | 2018-08-24 | 2018-12-28 | 河北工业大学 | A kind of IGBT module degree of aging appraisal procedure |
| CN109214449A (en) * | 2018-08-28 | 2019-01-15 | 华北电力大学 | A kind of electric grid investment needing forecasting method |
| CN109345005A (en) * | 2018-09-12 | 2019-02-15 | 中国电力科学研究院有限公司 | A kind of integrated energy system multidimensional optimization method based on improvement whale algorithm |
| CN109765893A (en) * | 2019-01-17 | 2019-05-17 | 重庆邮电大学 | Method for planning path for mobile robot based on whale optimization algorithm |
Non-Patent Citations (9)
| Title |
|---|
| IBRAHIM ALJARAH 等: "Optimizing connection weights in neural networks using the whale optimization algorithm", 《SOFT COMPUTING》 * |
| P. DINAKARA PRASAD REDDY 等: "Optimal renewable resources placement in distribution networks by combined power loss index and whale optimization algorithms", 《JOURNAL OF ELECTRICAL SYSTEMS AND INFORMATION TECHNOLOGY》 * |
| SEYEDALI MIRJALILI 等: "The Whale Optimization Algorithm", 《ADVANCES IN ENGINEERING SOFTWARE》 * |
| WANG CHUNZHI 等: "Credit Card Fraud Detection Based on Whale Algorithm Optimized BP Neural Network", 《THE 13TH INTERNATIONAL CONFERENCE ON COMPUTER SCIENCE & EDUCATION》 * |
| 李岸达 等: "基于多目标鲸鱼优化的关键质量特性识别方法", 《系统工程》 * |
| 杜沛: "基于多目标鲸鱼优化算法和Elman神经网络的短期风速预测模型的研究与应用", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
| 牛培峰: "基于鲸鱼优化算法的汽轮机热耗率模型预测", 《化工学报》 * |
| 王家君 等: "基于多预测模型与非线性组合的神经网络风速预测", 《物联网学报》 * |
| 龙文 等: "收敛因子非线性变化的鲸鱼优化算法", 《兰州理工大学学报》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111948283A (en) * | 2020-07-07 | 2020-11-17 | 上海交通大学 | System and method for detecting oxidation stability of transformer oil |
| CN113240307A (en) * | 2021-05-24 | 2021-08-10 | 上海电力大学 | Power system economic dispatching method based on improved differential evolution algorithm |
| CN113240307B (en) * | 2021-05-24 | 2022-09-27 | 上海电力大学 | Power system economic dispatching method based on improved differential evolution algorithm |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110322050B (en) | 2023-04-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111310889B (en) | Evaporation waveguide profile estimation method based on deep neural network | |
| Kavousi-Fard | Modeling uncertainty in tidal current forecast using prediction interval-based SVR | |
| CN109001736B (en) | Radar echo extrapolation method based on deep space-time prediction neural network | |
| CN113361777B (en) | Runoff prediction method and system based on VMD decomposition and IHHO optimization LSTM | |
| CN113326969B (en) | Short-term wind speed prediction method and system based on improved whale optimization algorithm and ELM | |
| CN105260786B (en) | A kind of simulation credibility of electric propulsion system assessment models comprehensive optimization method | |
| CN113568055B (en) | Aviation transient electromagnetic data inversion method based on LSTM network | |
| CN108985515B (en) | New energy output prediction method and system based on independent cyclic neural network | |
| CN104636985A (en) | Method for predicting radio disturbance of electric transmission line by using improved BP (back propagation) neural network | |
| CN114362175B (en) | Wind power prediction method and system based on depth certainty strategy gradient algorithm | |
| Syama et al. | A hybrid extreme learning machine model with lévy flight chaotic whale optimization algorithm for wind speed forecasting | |
| CN114462718A (en) | CNN-GRU wind power prediction method based on time sliding window | |
| CN110322050A (en) | A kind of wind energy resources compensation data method | |
| Wang et al. | Quantile deep learning model and multi-objective opposition elite marine predator optimization algorithm for wind speed prediction | |
| Yao et al. | Wave height forecast method with multi-step training set extension LSTM neural network | |
| CN105138729B (en) | Based on PSO GRNN wind power plant wind turbine defect air speed value fill methods | |
| CN115877483A (en) | Typhoon path forecasting method based on random forest and GRU | |
| CN114154401A (en) | Soil erosion modulus calculation method and system based on machine learning and observation data | |
| CN116454863A (en) | Optimal weight determining method of wind power combination prediction model based on improved hawk optimization algorithm | |
| CN114897204A (en) | Method and device for predicting short-term wind speed of offshore wind farm | |
| WO2022032873A1 (en) | Adversarial neural network-based hydrological parameter calibration method for data-lacking region | |
| CN113486591A (en) | Gravity multi-parameter data density weighted inversion method for convolutional neural network result | |
| CN112529684A (en) | Customer credit assessment method and system based on FWA _ DBN | |
| CN111914488A (en) | Data regional hydrological parameter calibration method based on antagonistic neural network | |
| CN116070498A (en) | Handheld rapid damage assessment device |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |