CN109063907A - A kind of big step-ahead prediction method of line of high-speed railway extreme wind speed intelligence traversal - Google Patents
A kind of big step-ahead prediction method of line of high-speed railway extreme wind speed intelligence traversal Download PDFInfo
- Publication number
- CN109063907A CN109063907A CN201810846131.XA CN201810846131A CN109063907A CN 109063907 A CN109063907 A CN 109063907A CN 201810846131 A CN201810846131 A CN 201810846131A CN 109063907 A CN109063907 A CN 109063907A
- Authority
- CN
- China
- Prior art keywords
- wind speed
- prediction
- measuring station
- air measuring
- neural network
- 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 45
- 238000013528 artificial neural network Methods 0.000 claims abstract description 89
- 238000009423 ventilation Methods 0.000 claims abstract description 17
- 238000012549 training Methods 0.000 claims description 66
- 241000282461 Canis lupus Species 0.000 claims description 57
- 230000008569 process Effects 0.000 claims description 17
- 230000006870 function Effects 0.000 claims description 15
- 238000002922 simulated annealing Methods 0.000 claims description 13
- 238000000137 annealing Methods 0.000 claims description 12
- 230000006399 behavior Effects 0.000 claims description 9
- 238000005070 sampling Methods 0.000 claims description 9
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims description 7
- 230000005012 migration Effects 0.000 claims description 6
- 238000013508 migration Methods 0.000 claims description 6
- 230000035772 mutation Effects 0.000 claims description 6
- 238000005457 optimization Methods 0.000 claims description 4
- 230000017105 transposition Effects 0.000 claims description 4
- 241000282421 Canidae Species 0.000 claims description 3
- 241000288673 Chiroptera Species 0.000 claims description 3
- 230000000739 chaotic effect Effects 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims description 3
- 238000001816 cooling Methods 0.000 claims description 3
- 210000004218 nerve net Anatomy 0.000 claims description 3
- 239000002245 particle Substances 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- XEEYBQQBJWHFJM-UHFFFAOYSA-N Iron Chemical compound [Fe] XEEYBQQBJWHFJM-UHFFFAOYSA-N 0.000 claims 2
- 230000006978 adaptation Effects 0.000 claims 1
- 229910052742 iron Inorganic materials 0.000 claims 1
- 230000004044 response Effects 0.000 claims 1
- 230000008901 benefit Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 210000001367 artery Anatomy 0.000 description 2
- 230000006835 compression Effects 0.000 description 2
- 238000007906 compression Methods 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 210000003462 vein Anatomy 0.000 description 2
- 241001269238 Data Species 0.000 description 1
- 240000007594 Oryza sativa Species 0.000 description 1
- 235000007164 Oryza sativa Nutrition 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 238000007792 addition Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000004836 empirical method Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000013277 forecasting method Methods 0.000 description 1
- 239000011521 glass Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 239000002244 precipitate Substances 0.000 description 1
- 235000009566 rice Nutrition 0.000 description 1
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
- G06N3/08—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
- G06Q50/40—Business processes related to the transportation industry
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Health & Medical Sciences (AREA)
- Economics (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- General Business, Economics & Management (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Marketing (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Tourism & Hospitality (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Development Economics (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Game Theory and Decision Science (AREA)
- Quality & Reliability (AREA)
- Primary Health Care (AREA)
- Feedback Control In General (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention provides a kind of line of high-speed railway extreme wind speeds intelligently to traverse big step-ahead prediction method, according to recent wind velocity condition, by constructing target air measuring station and time shift air measuring station in target ventilation measuring point, after carrying out denoising to the data of air measuring station, utilize PID neural network, air speed data after denoising is trained, forecasting wind speed model of each air measuring station under a variety of step-lengths is constructed;It selects each model to carry out the optimum prediction combination of a variety of step-lengths, realizes multi-Step Iterations prediction, improve precision of prediction, reduce the interference of random error;Along Railway wind speed advanced prediction is realized, can learn the wind speed environments situation in Frequent Accidents region in advance, in time, effectively instructs train operation, ensures train operation safety.
Description
Technical field
The invention belongs to railway forecasting wind speed field, in particular to intelligently traversal is big for a kind of line of high-speed railway extreme wind speed
Step-ahead prediction method.
Background technique
Strong wind is the one of the major reasons for influencing safety of railway traffic, makees running train by very strong beam wind
With.When train operation to curve, derailing, capsizing case easily occur under the action of strong wind for the special rail section such as hills, threaten fortune
Defeated safety.On December 25th, 2005, modern architecture in Japan express train " spike of rice 14 " traveling to Japan the north chevron county village Nei Zhen
When, it meets with precipitate strange wind and attacks, train derailing is overturned, and causes 4 people dead, 33 people are injured.9 days to 11 April in 2006
Day, the 30 years one chance strong wind in the Xinjiang area Bai Lifeng, the gravel that strong wind is rolled all smashes T70 window on train glass, train
Stop wheel and surpasses 20 hours;On 2 28th, 2007, No. 5807 times 11 section compartment of train was turned over by strong wind, caused 3 passenger's death, and 2
Name passenger is severely injured.On April 23rd, 2011, near the mound town Boot province Ai Sikeer of Argentina south, blast causes a column to be just expert at
Derailing in sailing causes more than 20 name passengers injured.
The baneful influence of strong wind is more significant with the increase of train running speed, seriously restricts railway transport of passengers shipping
Speed-raising and expanding economy.
Summary of the invention
The present invention proposes a kind of very big wind of line of high-speed railway to realize the high-precision forecast of the short-term wind speed of Along Railway
Fast intelligence traverses big step-ahead prediction method and selects the optimum prediction of a variety of step-lengths of each model to combine according to recent wind velocity condition, real
Existing multi-Step Iterations prediction, improves precision of prediction, reduces the interference of random error.
A kind of big step-ahead prediction method of line of high-speed railway extreme wind speed intelligence traversal, comprising the following steps:
Step 1: in railway target ventilation measuring point, air measuring station, including target air measuring station and time shift air measuring station are set;
For the target air measuring station apart from 100 meters of railway target ventilation measuring point, the time shift air measuring station includes at least 3, and sets
Where setting railway target ventilation measuring point and target air measuring station on line, first time shift air measuring station is apart from railway target ventilation measuring point 500
Meter, spacing is 500 meters between adjacent time shift air measuring station;
Step 2: building training sample data;
Wind speed of each air measuring station in historical time section is acquired with identical sample frequency, successively by the history wind of each air measuring station
Speed obtains training sample data using the wind speed intermediate value in time interval T as the sample moment wind speed of each air measuring station;
Will multiple wind speed in each time interval T air speed value of the intermediate value as a sample moment, compression histories
Air speed data;
Step 3: using the prediction step of training sample data and setting, constructing the forecasting wind speed based on PID neural network
Model group;
Successively with any three air measuring stations in target air measuring station and all time shift air measuring stations in any historical juncture t0Wind
Speed value is used as input data, and remaining air measuring station is in t0The air speed value of+time Δt carries out PID neural network as output data
Training obtains the forecasting wind speed model based on PID neural network that each air measuring station prediction step is Δ t;
The value of the prediction step Δ t is followed successively by p, 2p, 3p, 4p, and p is the prediction step unit time, and value range is
1-5min, a kind of corresponding one group of forecasting wind speed model based on PID neural network of prediction step;
Four prediction steps, four groups of forecasting wind speed models, every group of forecasting wind speed model include the wind of four air measuring stations altogether
Fast prediction model;
The input data of every group of forecasting wind speed model is the wind speed of four air measuring stations at a certain moment, and output data is when passing through
Between after Δ t, the prediction of wind speed of four air measuring stations;
The forecasting wind speed model of some air measuring station, which refers to, in actually every group of forecasting wind speed model surveys wind using the other three
The wind speed at a certain moment stood predicts the air measuring station in the wind speed after time Δ t;
Step 4: according to the target prediction time, the prediction task for constructing all air measuring stations is iterative vectorized;
Target prediction time m is split as n sub- predicted time hi, according to sub- predicted time and forecasting wind speed model group
Step-length is corresponded to, and the corresponding forecasting wind speed model group of each sub- predicted time is selected, formed each air measuring station predict task iteration to
Measure l={ hi,j, hi,jIndicate prediction that i-th of sub- predicted time selects j-th of forecasting wind speed model group to carry out forecasting wind speed
Task, the value range of i are 1-n, and the value range of j is 1-4;
The target prediction time m refers to is carrying out forecasting wind speed after time m;
Every sub- predicted time need to select forecasting wind speed model group four air measuring stations of progress of a corresponding step-length
Forecasting wind speed when by sub- predicted time;
Each sub- predicted time selects a forecasting wind speed model group to carry out a subtask prediction, referred to as one prediction
Subtask, the input of a prediction subtask are the air speed data of four air measuring station synchronizations, four data altogether, export and are
Four air measuring stations correspond to the air speed data of synchronization after step delta t by sub- predicted time, altogether four data, export number
According to the input for being used directly for next prediction subtask, the prediction in conventional prediction technique to extra air speed value is avoided,
The number of iterations is reduced, precision of prediction is promoted;
Step 5: any one the prediction task obtained using step 4 is iterative vectorized, carries out forecasting wind speed;
Using the air speed data of four air measuring stations of current time t moment as selected prediction task it is iterative vectorized in first it is pre-
The input data of subtask is surveyed, is surveyed with the target in the last one iterative vectorized prediction subtask output data of selected prediction task
Wind speed value of the air speed data at wind station as target prediction time m target ventilation measuring point;
Input data of the output data of previous prediction subtask as the latter prediction subtask.
Further, using newest historical wind speed data, selection optimum prediction task is iterative vectorized, carries out forecasting wind speed,
Optimal wind speed prediction result is obtained, detailed process is as follows:
Step A: being based on target prediction time m, selects and historical wind speed data of the current time t within the m+4p period
According to the construction method of training sample data, obtain forecast sample data, and select from forecast sample data each air measuring station according to
The secondary wind speed in t, t-m, t-m-p, t-m-2p, t-m-3p, t-m-4p and t-p, t-2p, t-3p, t-4p;
Step B: by wind speed of four air measuring stations when constantly for t-m, t-m-p, t-m-2p, t-m-3p, t-m-4p, successively
The input data iterative vectorized as each prediction task, obtain each prediction task it is iterative vectorized t, t-p, t-2p, t-3p,
The target air measuring station prediction of wind speed obtained when t-4p;
Step C: the iterative vectorized target successively obtained in t, t-p, t-2p, t-3p, t-4p of each prediction task is calculated
The error of air measuring station prediction of wind speed and actual measurement wind speed, and seeks mean value to error, obtains iterative vectorized total of each prediction task
Body predicts error;
Step D: choose the smallest prediction task of macro-forecast error amount it is iterative vectorized as optimum prediction task iteration to
Amount carries out forecasting wind speed, obtains optimal wind speed prediction result.
Further, the iterative vectorized weight of prediction task is set, and building optimum prediction task iteration merges vector, carries out wind
Speed prediction, obtains optimal wind speed prediction result,;
By the iterative vectorized macro-forecast error amount of all prediction tasks, according to sorting from small to large, preceding 5 predictions are chosen
Task is iterative vectorized, and the macro-forecast error amount iterative vectorized according to 5 selected prediction tasks accounts for selected 5 prediction tasks
The iterative vectorized weight of each prediction task is arranged in the ratio of the sum of iterative vectorized macro-forecast error amount, constructs OPTIMAL TASK
Iteration merges vector, carries out forecasting wind speed, obtains optimal wind speed prediction result.
5 predictions of wind speed progress that 5 prediction tasks are iterative vectorized, to synchronization are chosen, according to each prediction task
5 prediction results are carried out weight combination, obtain the prediction of wind speed of target air measuring station by iterative vectorized weight;
Further, Kalman filtering processing is interacted to the sample data of each air measuring station, filtered data is used
In the iterative vectorized selection of model training and prediction task.
Further, the prediction task it is iterative vectorized in each prediction subtask sub- predicted time meet it is following public
Formula:
Wherein, hminValue be the prediction step unit time.
Further, on the direction parallel with railway, each air measuring station two sides spaced set has same type air measuring station,
Target air measuring station group and time shift air measuring station group are obtained, wherein time shift air measuring station group includes the first time shift air measuring station group, the second time shift
Air measuring station group and third time shift air measuring station group;
From the air speed value that each air measuring station of target air measuring station group measures, the maximum wind velocity value conduct of identical sampling instant is chosen
The air speed value of each sampling instant of virtual target air measuring station, using virtual target air measuring station as target sample air measuring station;
From each time shift air measuring station group, the air speed value of identical sampling instant and the wind to the virtual air measuring station corresponding moment are chosen
The speed value maximum time shift air measuring station of conspicuousness, obtains time shift target air measuring station.
Further, to the threshold value and power of the PID neural network in the forecasting wind speed model based on PID neural network
Value is optimized using following at least one method simultaneously:
1) wolf pack-simulated annealing, used PID neural network input layer node number are 3, and hidden layer number is
3, output layer node number is 1;Maximum number of iterations in training process is set as 200, and training learning rate is 0.1, and threshold value is
0.004;;
2) water round-robin algorithm, used PID neural network, used PID neural network input layer node number are
3, hidden layer number is 3, and output layer node number is 1;Maximum number of iterations in training process is set as 200, training study
Rate is 0.1, threshold value 0.004;
3) chaos difference bat algorithm, used PID neural network, used PID neural network input layer node
Number is 3, and hidden layer number is 3, and output layer node number is 1;Maximum number of iterations in training process is set as 200, instruction
Practicing learning rate is 0.1, threshold value 0.004.
Further, using wolf pack-simulated annealing in the forecasting wind speed model based on PID neural network
The step of threshold value and weight of PID neural network optimize simultaneously is as follows:
Step 1.1): using individual wolf position as described based on PID neural network in PID neural network forecasting wind speed model
Threshold value, weight, simultaneously wolf pack parameter is arranged in every individual wolf in random initializtion wolf pack:
Wolf pack scale value range are as follows: [50,200], step factor value range are as follows: [50,120] visit wolf scale factor
Value range are as follows: [2,6], maximum migration number value range are as follows: [10,40], range estimation factor value range are as follows: [40,
100], maximum long-range raid number value range are as follows: [4,16] update scale factor value range are as follows: [2,6], maximum number of iterations
Value range are as follows: [500,1000], maximum search precision value range are as follows: [0.001,0.005];Set simulated annealing
Annealing initial temperature is 100, annealing rate is φ=0.8, annealing the number of iterations t2=1, maximum anneal cycles under Current Temperatures
Number is Lmax=6;
Step 1.2): setting fitness function, and determine initial optimal head wolf position and the number of iterations t1, t1=1;
The threshold value of the corresponding PID neural network of individual wolf position, wind speed of the weight substitution based on PID neural network is pre-
The forecasting wind speed model output wind speed predicted value based on PID neural network surveyed in model, and determined using individual wolf position, will
Inverse fitness function F as artificial wolf of the obtained wind speed value with the mean square error between expected wind speed value1;
Wherein, M indicates frequency of training, xi、yiRespectively indicate the wind speed value and desired output of i-th training;
Step 1.3): migration behavior successively is carried out to all artificial wolves, long-range raid behavior, besieges behavior, according to individual wolf
Fitness updates wolf pack, obtains updated optimal head wolf position;
Step 1.4): judge whether that reaching optimization required precision or maximum number of iterations enables t if not reaching1=t1+1
Step 1.5) is gone to, if reaching, goes to step 1.7);
Step 1.5): simulated annealing operation is carried out to the head wolf individual in this generation, in obtained head wolf position giIn neighborhood
Randomly choose new position gjAnd calculate the difference Δ F of the two fitness1=F1(gi)-F1(gj), calculating select probability P=exp (-
ΔF1/Tei), TeiFor Current Temperatures;If P > random [0,1), then it will be when front wolf position is by giReplace with gj, and with gjMake
For the beginning of next optimizing, otherwise with giStart optimizing next time;
Step 1.6): t is enabled2=t2+ 1, according to Tei+1=Tei* φ carries out cooling annealing, if t2<Lmax, go to step
1.5) step 1.3), otherwise, is gone to;
Step 1.7): weight, threshold value used in the corresponding PID neural network in export head wolf position.
Further, using water round-robin algorithm to the PID nerve in the forecasting wind speed model based on PID neural network
The step of threshold value and weight of network optimize simultaneously is as follows:
Step 2.1): threshold value of each rainfall layer as the PID neural network, weight initialize rainfall layer;
The value range of rainfall layer population quantity is set as [30,80], ocean quantity is 1, the value range of river quantity
For [10,20], the value range of maximum number of iterations is [500,1000], and the value range of minimum is [0.001,0.005];
Step 2.2: the corresponding weight of rainfall layer, threshold value being substituted into the forecasting wind speed model based on PID neural network, benefit
It is calculated with the forecasting wind speed model based on PID neural network that rainfall layer determines pre- with the wind speed that wind speed training subsample is input
The inverse of the mean square error of measured value and wind speed training expectation sample is as the second fitness function;
Wherein, M indicates frequency of training, xi、yiRespectively indicate the wind speed value and desired output of i-th training;
Step 2.3: using the maximum rainfall layer of fitness as sea, with fitness inferior to sea and the larger drop of fitness
Rain layer is as river, remaining rainfall layer is as the streams flowed to river or sea;
Step 2.4: in flow process, if the fitness in streams is higher than the fitness in river, streams river is exchanged
Position, if the fitness in river is higher than the fitness in sea, river sea transposition finally makes streams flow into river, river
Stream flows into ocean;
Step 2.5: judging whether the absolute difference between river and ocean fitness is less than minimum, if so, going to
Step 2.6);If it is not, repeating step 2.5);
Step 2.6: judging whether to reach maximum number of iterations, if it is not, giving up from rainfall layer population into next iteration
The river is abandoned, rainfall is re-started, random rainfall layer is generated and population is added, go to step 2.3);If so, output ocean drop
Threshold value of the corresponding parameter of rain layer as the PID neural network, weight.
Further, using chaos difference bat algorithm in the forecasting wind speed model based on PID neural network
The step of threshold value and weight of PID neural network optimize simultaneously is as follows:
Step 3.1): refreshing using the position of bat individual as PID described in the forecasting wind speed model based on PID neural network
Threshold value and weight through network;
The value range of bat population scale is [100,500], bat individual maximum impulse frequency r0=0.5, maximum arteries and veins
Rush intensity of sound A0Value range be [0.3,0.8], bat search rate increase coefficient value range be [0.02,
0.05], the value range of intensity of sound attenuation coefficient is [0.75,0.95], and crossover probability is set as 0.5, mutation probability setting
It is 0.5, the value range of fitness variance threshold values is [0.01,0.06], and the value range of search pulse frequency is [0,1.5],
The value range of maximum number of iterations is [200,500], and the value range of maximum search precision is [0.02,0.1];
Step 3.2): the position of each bat individual, speed, frequency in bat population are initialized according to Chaotic map sequence
Rate;
Step 3.3): setting fitness function, and determine initial optimal bat body position and the number of iterations t3, t3=1;
The corresponding threshold value in bat body position and weight are substituted into the forecasting wind speed model based on PID neural network, utilized
It is the wind inputted that the forecasting wind speed model based on PID neural network that bat body position determines, which is calculated with wind speed training subsample,
The inverse of fast predicted value and the mean square error of wind speed training expectation sample is as third fitness function;
Wherein, M indicates frequency of training, xi、yiRespectively indicate the wind speed value and desired output of i-th training.
Step 3.4): search pulse frequency, the position and speed of bat are updated using the pulse frequency of setting;
Step 3.5): if Random1>ri, then random perturbation is carried out for the bat of personal best particle, generates optimal position
Set the disturbance location of bat;
Wherein Random1For equally distributed random number, r on closed interval [0,1]iFor the pulse frequency of i-th bat;
Step 3.6): if Random2>Ei, the fitness of the disturbance location of optimal bat individual is represented better than disturbance front position
Fitness, optimal bat individual is moved to disturbance location, otherwise optimal bat body position is motionless;
Wherein Random2For equally distributed random number, E on closed interval [0,1]iFor the intensity of sound of i-th bat;
Step 3.7): the fitness of all bats individual of current population and the population's fitness side of bat population are calculated
Difference;
Earliness is judged according to the population's fitness variance of bat population, is given if bat population's fitness variance is less than
Threshold value, all bat individuals intersect and mutation operation, and go to step 3.5), otherwise, select optimal bat individual,
Go to step 3.8);
Step 3.8): judging whether to reach maximum number of iterations or maximum search precision, if so, exporting optimal bat individual
The threshold value and weight of PID neural network described in the corresponding forecasting wind speed model based on PID neural network in position, if it is not, t3
=t3+ 1, go to step 3.4).
Beneficial effect
The present invention provides a kind of line of high-speed railway extreme wind speeds intelligently to traverse big step-ahead prediction method, according to recent wind
Fast situation selects the optimum prediction of a variety of step-lengths of each model to combine, realize multi-Step Iterations prediction, improve precision of prediction, reduce with
The interference of chance error difference;For compared with the prior art, advantage specifically includes the following:
1. obtaining respective counts by constructing a target air measuring station in target ventilation measuring point and selecting 3 time shift air measuring stations
According to, set several prediction steps, using three air measuring station synchronization air speed datas as input, a remaining air measuring station
The PID mind of the respective various step-lengths of four air measuring stations is respectively trained as output in air speed data after postponing a prediction step
Through network forecasting wind speed model.Compared to existing wind speed forecasting method, the air speed data established between four air measuring stations is closed
System, prediction can obtain the prediction data of four air measuring stations each time;By target prediction time decomposition at multiple combinations, every kind
Group is combined into the sum of several predicted times, and predicted time is corresponding with model prediction step-length, an each predicted time i.e. prediction
Task is completed to avoid to a large amount of process moment wind speed the forecasting wind speed of object time by multiple prediction task beat types
Value predicted, reduces the iteration prediction number of forecasting wind speed model, and to multiple combinations according to current wind speed environments into
Row optimizing, optimum combination carry out forecasting wind speed, significantly improve precision of prediction;
2. by using wolf pack-simulated annealing, water round-robin algorithm and chaos difference bat algorithm respectively to PID
The threshold value and weight of neural network carry out initial parameter optimization, avoid initial parameter and choose and improper do to model training process
The influence with predictive ability is disturbed, the limitation that empirical method determines initial parameter value is also avoided;
3. utilizing method proposed by the invention, when train operation to curve, the special rail section such as hills, and it is in big vane
When under border, Along Railway wind speed advanced prediction can be realized, can learn the wind speed environments situation in Frequent Accidents region in advance, and
When, effectively instruct train operation, ensure train operation safety.
Detailed description of the invention
Fig. 1 is the prediction model training schematic diagram in the method for the invention;
Fig. 2 is the forecasting wind speed flow diagram of the method for the invention;
Fig. 3 is that schematic diagram is arranged in air measuring station.
Specific embodiment
Below in conjunction with drawings and examples, the present invention is described further.
As depicted in figs. 1 and 2, a kind of big step-ahead prediction method of line of high-speed railway extreme wind speed intelligence traversal, including with
Lower step:
Step 1: in railway target ventilation measuring point, air measuring station, including target air measuring station and time shift air measuring station are set;
For the target air measuring station apart from 100 meters of railway target ventilation measuring point, the time shift air measuring station includes at least 3, and sets
Where setting railway target ventilation measuring point and target air measuring station on line, first time shift air measuring station is apart from railway target ventilation measuring point 500
Meter, spacing is 500 meters between adjacent time shift air measuring station;
As shown in figure 3, each air measuring station two sides spaced set has same type to survey wind on the direction parallel with railway
It stands, obtains target air measuring station group and time shift air measuring station group, when wherein time shift air measuring station group includes the first time shift air measuring station group, second
Move air measuring station group and third time shift air measuring station group;
From the air speed value that each air measuring station of target air measuring station group measures, the maximum wind velocity value conduct of identical sampling instant is chosen
The air speed value of each sampling instant of virtual target air measuring station, using virtual target air measuring station as target sample air measuring station;
From each time shift air measuring station group, the air speed value of identical sampling instant and the wind to the virtual air measuring station corresponding moment are chosen
The speed value maximum time shift air measuring station of conspicuousness, obtains time shift target air measuring station.
Step 2: building training sample data;
Wind speed of each air measuring station in historical time section is acquired with identical sample frequency, successively by the history wind of each air measuring station
Speed obtains training sample data using the wind speed intermediate value in time interval T as the sample moment wind speed of each air measuring station;
Will multiple wind speed in each time interval T air speed value of the intermediate value as a sample moment, compression histories
Air speed data;
In this example, interval 3S acquires primary air velocity;
Kalman filtering processing is interacted to the sample data of each air measuring station, filtered data are used for model training
The iterative vectorized selection with prediction task.
Step 3: using the prediction step of training sample data and setting, constructing the forecasting wind speed based on PID neural network
Model group;
Successively with any three air measuring stations in target air measuring station and all time shift air measuring stations in any historical juncture t0Wind
Speed value is used as input data, and remaining air measuring station is in t0The air speed value of+time Δt carries out PID neural network as output data
Training obtains the forecasting wind speed model based on PID neural network that each air measuring station prediction step is Δ t;
The value of the prediction step Δ t is followed successively by p, 2p, 3p, 4p, and p is the prediction step unit time, and value range is
1-5min, a kind of corresponding one group of forecasting wind speed model based on PID neural network of prediction step;
Four prediction steps, four groups of forecasting wind speed models, every group of forecasting wind speed model include the wind of four air measuring stations altogether
Fast prediction model;
The input data of every group of forecasting wind speed model is wind speed of four air measuring stations in certain initial time, and output data is warp
Cross time Δ t1Afterwards, the prediction of wind speed of four air measuring stations, to pass through time Δ t1The prediction of wind speed of four air measuring stations is as wind later
The input data of fast prediction model group exports as by time Δ t1+Δt2The prediction of wind speed of four air measuring stations afterwards, is omitted pair
From initial time to by time Δ t1Forecasting wind speed is carried out at the time of during this, beat type completion is pre- to the object time
It surveys;
The forecasting wind speed model of some air measuring station, which refers to, in actually every group of forecasting wind speed model surveys wind using the other three
The wind speed at a certain moment stood predicts the air measuring station in the wind speed after time Δ t;
The threshold value and weight of PID neural network in the forecasting wind speed model based on PID neural network are adopted simultaneously
It is optimized with following at least one method:
1) wolf pack-simulated annealing, used PID neural network input layer node number are 3, and hidden layer number is
3, output layer node number is 1;Maximum number of iterations in training process is set as 200, and training learning rate is 0.1, and threshold value is
0.004;
2) water round-robin algorithm, used PID neural network input layer node number are 3, and hidden layer number is 3, output
Node layer number is 1;Maximum number of iterations in training process is set as 200, and training learning rate is 0.1, threshold value 0.004;;
3) chaos difference bat algorithm, used PID neural network input layer node number are 3, and hidden layer number is
3, output layer node number is 1;Maximum number of iterations in training process is set as 200, and training learning rate is 0.1, and threshold value is
0.004;.
Using wolf pack-simulated annealing to the PID nerve net in the forecasting wind speed model based on PID neural network
The step of threshold value and weight of network optimize simultaneously is as follows:
Step 1.1): using individual wolf position as described based on PID neural network in PID neural network forecasting wind speed model
Threshold value, weight, simultaneously wolf pack parameter is arranged in every individual wolf in random initializtion wolf pack:
Wolf pack scale value range are as follows: [50,200], step factor value range are as follows: [50,120] visit wolf scale factor
Value range are as follows: [2,6], maximum migration number value range are as follows: [10,40], range estimation factor value range are as follows: [40,
100], maximum long-range raid number value range are as follows: [4,16] update scale factor value range are as follows: [2,6], maximum number of iterations
Value range are as follows: [500,1000], maximum search precision value range are as follows: [0.001,0.005];Set simulated annealing
Annealing initial temperature is 100, annealing rate is φ=0.8, annealing the number of iterations t2=1, maximum anneal cycles under Current Temperatures
Number is Lmax=6;
Step 1.2): setting fitness function, and determine initial optimal head wolf position and the number of iterations t1, t1=1;
The threshold value of the corresponding PID neural network of individual wolf position, wind speed of the weight substitution based on PID neural network is pre-
The forecasting wind speed model output wind speed predicted value based on PID neural network surveyed in model, and determined using individual wolf position, will
Inverse fitness function F as artificial wolf of the obtained wind speed value with the mean square error between expected wind speed value1;
Wherein, M indicates frequency of training, xi、yiRespectively indicate the wind speed value and desired output of i-th training;
Step 1.3): migration behavior successively is carried out to all artificial wolves, long-range raid behavior, besieges behavior, according to individual wolf
Fitness updates wolf pack, obtains updated optimal head wolf position;
Step 1.4): judge whether that reaching optimization required precision or maximum number of iterations enables t if not reaching1=t1+1
Step 1.5) is gone to, if reaching, goes to step 1.7);
Step 1.5): simulated annealing operation is carried out to the head wolf individual in this generation, in obtained head wolf position giIn neighborhood
Randomly choose new position gjAnd calculate the difference Δ F of the two fitness1=F1(gi)-F1(gj), calculating select probability P=exp (-
ΔF1/Tei), TeiFor Current Temperatures;If P > random [0,1), then it will be when front wolf position is by giReplace with gj, and with gjMake
For the beginning of next optimizing, otherwise with giStart optimizing next time;
Step 1.6): t is enabled2=t2+ 1, according to Tei+1=Tei* φ carries out cooling annealing, if t2<Lmax, go to step
1.5) step 1.3), otherwise, is gone to;
Step 1.7): weight, threshold value used in the corresponding PID neural network in export head wolf position.
Using water round-robin algorithm to the threshold of the PID neural network in the forecasting wind speed model based on PID neural network
The step of value and weight optimize simultaneously is as follows:
Step 2.1): threshold value of each rainfall layer as the PID neural network, weight initialize rainfall layer;
The value range of rainfall layer population quantity is set as [30,80], ocean quantity is 1, the value range of river quantity
For [10,20], the value range of maximum number of iterations is [500,1000], and the value range of minimum is [0.001,0.005];
Step 2.2: the corresponding weight of rainfall layer, threshold value being substituted into the forecasting wind speed model based on PID neural network, benefit
It is calculated with the forecasting wind speed model based on PID neural network that rainfall layer determines pre- with the wind speed that wind speed training subsample is input
The inverse of the mean square error of measured value and wind speed training expectation sample is as the second fitness function;
Wherein, M indicates frequency of training, xi、yiRespectively indicate the wind speed value and desired output of i-th training;
Step 2.3: using the maximum rainfall layer of fitness as sea, with fitness inferior to sea and the larger drop of fitness
Rain layer is as river, remaining rainfall layer is as the streams flowed to river or sea;
Step 2.4: in flow process, if the fitness in streams is higher than the fitness in river, streams river is exchanged
Position, if the fitness in river is higher than the fitness in sea, river sea transposition finally makes streams flow into river, river
Stream flows into ocean;
Step 2.5: judging whether the absolute difference between river and ocean fitness is less than minimum, if so, going to
Step 2.6);If it is not, repeating step 2.5);
Step 2.6: judging whether to reach maximum number of iterations, if it is not, giving up from rainfall layer population into next iteration
The river is abandoned, rainfall is re-started, random rainfall layer is generated and population is added, go to step 2.3);If so, output ocean drop
Threshold value of the corresponding parameter of rain layer as the PID neural network, weight.
Using chaos difference bat algorithm to the PID nerve net in the forecasting wind speed model based on PID neural network
The step of threshold value and weight of network optimize simultaneously is as follows:
Step 3.1): refreshing using the position of bat individual as PID described in the forecasting wind speed model based on PID neural network
Threshold value and weight through network;
The value range of bat population scale is [100,500], bat individual maximum impulse frequency r0=0.5, maximum arteries and veins
Rush intensity of sound A0Value range be [0.3,0.8], bat search rate increase coefficient value range be [0.02,
0.05], the value range of intensity of sound attenuation coefficient is [0.75,0.95], and crossover probability is set as 0.5, mutation probability setting
It is 0.5, the value range of fitness variance threshold values is [0.01,0.06], and the value range of search pulse frequency is [0,1.5],
The value range of maximum number of iterations is [200,500], and the value range of maximum search precision is [0.02,0.1];
Step 3.2): the position of each bat individual, speed, frequency in bat population are initialized according to Chaotic map sequence
Rate;
Step 3.3): setting fitness function, and determine initial optimal bat body position and the number of iterations t3, t3=1;
The corresponding threshold value in bat body position and weight are substituted into the forecasting wind speed model based on PID neural network, utilized
It is the wind inputted that the forecasting wind speed model based on PID neural network that bat body position determines, which is calculated with wind speed training subsample,
The inverse of fast predicted value and the mean square error of wind speed training expectation sample is as third fitness function;
Wherein, M indicates frequency of training, xi、yiRespectively indicate the wind speed value and desired output of i-th training.
Step 3.4): search pulse frequency, the position and speed of bat are updated using the pulse frequency of setting;
Step 3.5): if Random1>ri, then random perturbation is carried out for the bat of personal best particle, generates optimal position
Set the disturbance location of bat;
Wherein Random1For equally distributed random number, r on closed interval [0,1]iFor the pulse frequency of i-th bat;
Step 3.6): if Random2>Ei, the fitness of the disturbance location of optimal bat individual is represented better than disturbance front position
Fitness, optimal bat individual is moved to disturbance location, otherwise optimal bat body position is motionless;
Wherein Random2For equally distributed random number, E on closed interval [0,1]iFor the intensity of sound of i-th bat;
Step 3.7): the fitness of all bats individual of current population and the population's fitness side of bat population are calculated
Difference;
Earliness is judged according to the population's fitness variance of bat population, is given if bat population's fitness variance is less than
Threshold value, all bat individuals intersect and mutation operation, and go to step 3.5), otherwise, select optimal bat individual,
Go to step 3.8);
Step 3.8): judging whether to reach maximum number of iterations or maximum search precision, if so, exporting optimal bat individual
The threshold value and weight of PID neural network described in the corresponding forecasting wind speed model based on PID neural network in position, if it is not, t3
=t3+ 1, go to step 3.4).
Step 4: according to the target prediction time, the prediction task for constructing all air measuring stations is iterative vectorized;
Target prediction time m is split as n sub- predicted time hi, according to sub- predicted time and forecasting wind speed model group
Step-length is corresponded to, and the corresponding forecasting wind speed model group of each sub- predicted time is selected, formed each air measuring station predict task iteration to
Measure l={ hi,j, hi,jIndicate prediction that i-th of sub- predicted time selects j-th of forecasting wind speed model group to carry out forecasting wind speed
Task, the value range of i are 1-n, and the value range of j is 1-4;
The target prediction time m refers to is carrying out forecasting wind speed after time m;
Every sub- predicted time need to select forecasting wind speed model group four air measuring stations of progress of a corresponding step-length
Forecasting wind speed when by sub- predicted time;
Each sub- predicted time selects a forecasting wind speed model group to carry out a subtask prediction, referred to as one prediction
Subtask, the input of a prediction subtask are the air speed data of four air measuring station synchronizations, four data altogether, export and are
Four air measuring stations correspond to the air speed data of synchronization after step delta t by sub- predicted time, altogether four data, export number
According to the input for being used directly for next prediction subtask, the prediction in conventional prediction technique to extra air speed value is avoided,
The number of iterations is reduced, precision of prediction is promoted;
Using newest historical wind speed data, selection optimum prediction task is iterative vectorized, carries out forecasting wind speed, obtains optimal wind
Fast prediction result, detailed process is as follows:
Step A: being based on target prediction time m, selects and historical wind speed data of the current time t within the m+4p period
According to the construction method of training sample data, obtain forecast sample data, and select from forecast sample data each air measuring station according to
The secondary wind speed in t, t-m, t-m-p, t-m-2p, t-m-3p, t-m-4p and t-p, t-2p, t-3p, t-4p;
Step B: by wind speed of four air measuring stations when constantly for t-m, t-m-p, t-m-2p, t-m-3p, t-m-4p, successively
The input data iterative vectorized as each prediction task, obtain each prediction task it is iterative vectorized t, t-p, t-2p, t-3p,
The target air measuring station prediction of wind speed obtained when t-4p;
Step C: the iterative vectorized target successively obtained in t, t-p, t-2p, t-3p, t-4p of each prediction task is calculated
The error of air measuring station prediction of wind speed and actual measurement wind speed, and seeks mean value to error, obtains iterative vectorized total of each prediction task
Body predicts error;
Step D: it is obtained using following two method iterative vectorized for carrying out the prediction task of forecasting wind speed;
1) it is iterative vectorized iterative vectorized as optimum prediction task to choose the smallest prediction task of macro-forecast error amount, into
Row forecasting wind speed obtains optimal wind speed prediction result;
2) the iterative vectorized weight of prediction task is set, building optimum prediction task iteration merges vector, carries out forecasting wind speed,
Obtain optimal wind speed prediction result;
By the iterative vectorized macro-forecast error amount of all prediction tasks, according to sorting from small to large, preceding 5 predictions are chosen
Task is iterative vectorized, and the macro-forecast error amount iterative vectorized according to 5 selected prediction tasks accounts for selected 5 prediction tasks
The iterative vectorized weight of each prediction task is arranged in the ratio of the sum of iterative vectorized macro-forecast error amount, constructs OPTIMAL TASK
Iteration merges vector, carries out forecasting wind speed, obtains optimal wind speed prediction result.
5 predictions of wind speed progress that 5 prediction tasks are iterative vectorized, to synchronization are chosen, according to each prediction task
5 prediction results are carried out weight combination, obtain the prediction of wind speed of target air measuring station by iterative vectorized weight;
The sub- predicted time of each prediction subtask meets following formula during the prediction task is iterative vectorized:
Wherein, hminValue be the prediction step unit time.
Step 5: any one the prediction task obtained using step 4 is iterative vectorized, carries out forecasting wind speed;
Using the air speed data of four air measuring stations of current time t moment as selected prediction task it is iterative vectorized in first it is pre-
The input data of subtask is surveyed, is surveyed with the target in the last one iterative vectorized prediction subtask output data of selected prediction task
Wind speed value of the air speed data at wind station as target prediction time m target ventilation measuring point;
Input data of the output data of previous prediction subtask as the latter prediction subtask.
Specific embodiment described herein is only an example for the spirit of the invention.The neck of technology belonging to the present invention
The technical staff in domain can make various modifications or additions to the described embodiments or replace by a similar method
In generation, however, it does not deviate from the spirit of the invention or beyond the scope of the appended claims.
Claims (10)
1. a kind of line of high-speed railway extreme wind speed intelligently traverses big step-ahead prediction method, which comprises the following steps:
Step 1: in railway target ventilation measuring point, air measuring station, including target air measuring station and time shift air measuring station are set;
For the target air measuring station apart from 100 meters of railway target ventilation measuring point, the time shift air measuring station includes at least 3, and iron is arranged
Where road target ventilation measuring point and target air measuring station on line, first time shift air measuring station is apart from 500 meters of railway target ventilation measuring point, phase
Spacing is 500 meters between adjacent time shift air measuring station;
Step 2: building training sample data;
Wind speed of each air measuring station in historical time section is acquired with identical sample frequency, successively by the historical wind speed of each air measuring station,
Using the wind speed intermediate value in time interval T as the sample moment wind speed of each air measuring station, training sample data are obtained;
Step 3: using the prediction step of training sample data and setting, constructing the forecasting wind speed model based on PID neural network
Group;
Successively with any three air measuring stations in target air measuring station and all time shift air measuring stations in any historical juncture t0Air speed value make
For input data, remaining air measuring station is in t0The air speed value of+time Δt is trained PID neural network as output data,
Obtain the forecasting wind speed model based on PID neural network that each air measuring station prediction step is Δ t;
The value of the prediction step Δ t is followed successively by p, 2p, 3p, 4p, and p is prediction step unit time, value range 1-
5min, a kind of corresponding one group of forecasting wind speed model based on PID neural network of prediction step;
Step 4: according to the target prediction time, the prediction task for constructing all air measuring stations is iterative vectorized;
Target prediction time m is split as n sub- predicted time hi, step-length according to sub- predicted time and forecasting wind speed model group
It is corresponded to, selects the corresponding forecasting wind speed model group of each sub- predicted time, form each air measuring station prediction iterative vectorized l=of task
{hi,j, hi,jIndicate the prediction subtask that i-th of sub- predicted time selects j-th of forecasting wind speed model group to carry out forecasting wind speed, i
Value range be 1-n, the value range of j is 1-4;
Step 5: any one the prediction task obtained using step 4 is iterative vectorized, carries out forecasting wind speed;
Using the air speed data of four air measuring stations of current time t moment as selected prediction task it is iterative vectorized in first prediction
The input data of task, with the target air measuring station in the last one iterative vectorized prediction subtask output data of selected prediction task
Wind speed value of the air speed data as target prediction time m target ventilation measuring point;
Input data of the output data of previous prediction subtask as the latter prediction subtask.
2. the method according to claim 1, wherein choosing optimum prediction using newest historical wind speed data and appointing
It is engaged in iterative vectorized, carries out forecasting wind speed, obtain optimal wind speed prediction result, detailed process is as follows:
Step A: be based on target prediction time m, select with historical wind speed data of the current time t within the m+4p period according to
The construction method of training sample data obtains forecast sample data, and selects each air measuring station from forecast sample data and successively exist
T, wind speed when t-m, t-m-p, t-m-2p, t-m-3p, t-m-4p and t-p, t-2p, t-3p, t-4p;
Step B: by wind speed of four air measuring stations when constantly for t-m, t-m-p, t-m-2p, t-m-3p, t-m-4p, successively conduct
The iterative vectorized input data of each prediction task, it is iterative vectorized in t, t-p, t-2p, t-3p, t-4p to obtain each prediction task
When the target air measuring station prediction of wind speed that obtains;
Step C: it calculates the iterative vectorized target successively obtained in t, t-p, t-2p, t-3p, t-4p of each prediction task and surveys wind
It stands and prediction of wind speed and surveys the error of wind speed, and mean value is sought to error, obtain iterative vectorized overall pre- of each prediction task
Survey error;
Step D: the selection the smallest prediction task of macro-forecast error amount is iterative vectorized iterative vectorized as optimum prediction task, into
Row forecasting wind speed obtains optimal wind speed prediction result.
3. according to the method described in claim 2, it is characterized in that, the setting iterative vectorized weight of prediction task, constructs optimal pre-
Survey task iteration merges vector, carries out forecasting wind speed, obtains optimal wind speed prediction result,;
The iterative vectorized macro-forecast error amount of all prediction tasks is chosen into preceding 5 predictions task according to sorting from small to large
Iterative vectorized and iterative vectorized according to 5 selected prediction tasks macro-forecast error amount accounts for selected 5 prediction task iteration
The iterative vectorized weight of each prediction task is arranged in the ratio of the sum of the macro-forecast error amount of vector, constructs OPTIMAL TASK iteration
Vector is merged, forecasting wind speed is carried out, obtains optimal wind speed prediction result.
4. according to the method described in claim 2, it is characterized in that, interacting Kalman's filter to the sample data of each air measuring station
Filtered data are used for model training and the iterative vectorized selection of prediction task by wave processing.
5. the method according to claim 1, wherein each prediction subtask during the prediction task is iterative vectorized
Sub- predicted time meet following formula:
Wherein, hminValue be the prediction step unit time.
6. method according to claim 1-5, which is characterized in that on the direction parallel with railway, each survey
Wind station two sides spaced set has same type air measuring station, obtains target air measuring station group and time shift air measuring station group, and wherein wind is surveyed in time shift
Group of standing includes the first time shift air measuring station group, the second time shift air measuring station group and third time shift air measuring station group;
From the air speed value that each air measuring station of target air measuring station group measures, the maximum wind velocity value of identical sampling instant is chosen as virtual
The air speed value of each sampling instant of target air measuring station, using virtual target air measuring station as target sample air measuring station;
From each time shift air measuring station group, the air speed value of identical sampling instant and the air speed value to the virtual air measuring station corresponding moment are chosen
The maximum time shift air measuring station of conspicuousness obtains time shift target air measuring station.
7. according to the method described in claim 6, it is characterized in that, to the forecasting wind speed model based on PID neural network
In PID neural network threshold value and weight optimized simultaneously using following at least one method:
1) wolf pack-simulated annealing, used PID neural network input layer node number are 3, and hidden layer number is 3, defeated
Node layer number is 1 out;Maximum number of iterations in training process is set as 200, and training learning rate is 0.1, and threshold value is
0.004;;
2) water round-robin algorithm, used PID neural network, used PID neural network input layer node number is 3, hidden
Number containing layer is 3, and output layer node number is 1;Maximum number of iterations in training process is set as 200, and training learning rate is
0.1, threshold value 0.004;
3) chaos difference bat algorithm, used PID neural network, used PID neural network input layer node number
It is 3, hidden layer number is 3, and output layer node number is 1;Maximum number of iterations in training process is set as 200, and training is learned
Habit rate is 0.1, threshold value 0.004.
8. the method according to the description of claim 7 is characterized in that using wolf pack-simulated annealing to described based on PID mind
The step of threshold value and weight of PID neural network in forecasting wind speed model through network optimize simultaneously is as follows:
Step 1.1): using individual wolf position as the threshold based on PID neural network in PID neural network forecasting wind speed model
It is worth, weight, simultaneously wolf pack parameter is arranged in the individual wolf of every in random initializtion wolf pack:
Wolf pack scale value range are as follows: [50,200], step factor value range are as follows: [50,120] visit wolf scale factor value
Range are as follows: [2,6], maximum migration number value range are as follows: [10,40], range estimation factor value range are as follows: [40,100],
Maximum long-range raid number value range are as follows: [4,16] update scale factor value range are as follows: [2,6], maximum number of iterations value model
It encloses are as follows: [500,1000], maximum search precision value range are as follows: [0.001,0.005];At the beginning of the annealing for setting simulated annealing
Beginning temperature is 100, annealing rate is φ=0.8, annealing the number of iterations t2=1, maximum anneal cycles number is under Current Temperatures
Lmax=6;
Step 1.2): setting fitness function, and determine initial optimal head wolf position and the number of iterations t1, t1=1;
The threshold value of the corresponding PID neural network of individual wolf position, weight are substituted into the forecasting wind speed mould based on PID neural network
In type, and the forecasting wind speed model output wind speed predicted value based on PID neural network determined using individual wolf position, it will obtain
Wind speed value with the mean square error between expected wind speed value fitness function F of the inverse as artificial wolf1;
Wherein, M indicates frequency of training, xi、yiRespectively indicate the wind speed value and desired output of i-th training;
Step 1.3): migration behavior successively is carried out to all artificial wolves, long-range raid behavior, besieges behavior, according to the adaptation of individual wolf
Degree updates wolf pack, obtains updated optimal head wolf position;
Step 1.4): judge whether that reaching optimization required precision or maximum number of iterations enables t if not reaching1=t1+ 1 goes to
Step 1.5) goes to step 1.7) if reaching;
Step 1.5): simulated annealing operation is carried out to the head wolf individual in this generation, in obtained head wolf position giIt is selected at random in neighborhood
Select new position gjAnd calculate the difference Δ F of the two fitness1=F1(gi)-F1(gj), calculate select probability P=exp (- Δ F1/
Tei), TeiFor Current Temperatures;If P > random [0,1), then it will be when front wolf position is by giReplace with gj, and with gjAs under
The beginning of secondary optimizing, otherwise with giStart optimizing next time;
Step 1.6): t is enabled2=t2+ 1, according to Tei+1=Tei* φ carries out cooling annealing, if t2<Lmax, step 1.5) is gone to, it is no
Then, step 1.3) is gone to;
Step 1.7): weight, threshold value used in the corresponding PID neural network in export head wolf position.
9. the method according to the description of claim 7 is characterized in that being based on PID neural network to described using water round-robin algorithm
Forecasting wind speed model in PID neural network threshold value and weight the step of optimizing, is as follows simultaneously:
Step 2.1): threshold value of each rainfall layer as the PID neural network, weight initialize rainfall layer;
The value range of rainfall layer population quantity is set as [30,80], ocean quantity is 1, and the value range of river quantity is
[10,20], the value range of maximum number of iterations are [500,1000], and the value range of minimum is [0.001,0.005];
Step 2.2: the corresponding weight of rainfall layer, threshold value being substituted into the forecasting wind speed model based on PID neural network, drop is utilized
It is the wind speed value inputted that the forecasting wind speed model based on PID neural network that rain layer determines, which is calculated with wind speed training subsample,
Inverse with the mean square error of wind speed training expectation sample is as the second fitness function;
Wherein, M indicates frequency of training, xi、yiRespectively indicate the wind speed value and desired output of i-th training;
Step 2.3: using the maximum rainfall layer of fitness as sea, with fitness inferior to sea and the larger rainfall layer of fitness
As river, remaining rainfall layer is as the streams flowed to river or sea;
Step 2.4: in flow process, if the fitness in streams be higher than river fitness, streams river transposition,
If the fitness in river is higher than the fitness in sea, river sea transposition finally makes streams flow into river, and river flows into
Ocean;
Step 2.5: judging whether the absolute difference between river and ocean fitness is less than minimum, if so, going to step
2.6);If it is not, repeating step 2.5);
Step 2.6: judging whether to reach maximum number of iterations, if it is not, giving up this from rainfall layer population into next iteration
River re-starts rainfall, generates random rainfall layer and population is added, go to step 2.3);If so, output ocean rainfall layer
Threshold value of the corresponding parameter as the PID neural network, weight.
10. the method according to the description of claim 7 is characterized in that using chaos difference bat algorithm to described based on PID mind
The step of threshold value and weight of PID neural network in forecasting wind speed model through network optimize simultaneously is as follows:
Step 3.1): using the position of bat individual as PID nerve net described in the forecasting wind speed model based on PID neural network
The threshold value and weight of network;
The value range of bat population scale is [100,500], bat individual maximum impulse frequency r0=0.5, maximum impulse sound
Intensity A0Value range be [0.3,0.8], bat search rate increase coefficient value range be [0.02,0.05], sound
The value range of strength retrogression's coefficient is [0.75,0.95], and crossover probability is set as 0.5, and mutation probability is set as 0.5, adapts to
The value range for spending variance threshold values is [0.01,0.06], and the value range of search pulse frequency is [0,1.5], greatest iteration time
Several value ranges is [200,500], and the value range of maximum search precision is [0.02,0.1];
Step 3.2): the position of each bat individual, speed, frequency in bat population are initialized according to Chaotic map sequence;
Step 3.3): setting fitness function, and determine initial optimal bat body position and the number of iterations t3, t3=1;
The corresponding threshold value in bat body position and weight are substituted into the forecasting wind speed model based on PID neural network, utilize bat
The forecasting wind speed model based on PID neural network that a body position determines calculates pre- with the wind speed that wind speed training subsample is input
The inverse of the mean square error of measured value and wind speed training expectation sample is as third fitness function;
Wherein, M indicates frequency of training, xi、yiRespectively indicate the wind speed value and desired output of i-th training.
Step 3.4): search pulse frequency, the position and speed of bat are updated using the pulse frequency of setting;
Step 3.5): if Random1>ri, then random perturbation is carried out for the bat of personal best particle, generates optimal location bat
The disturbance location of bat;
Wherein Random1For equally distributed random number, r on closed interval [0,1]iFor the pulse frequency of i-th bat;
Step 3.6): if Random2>Ei, represent fitness the fitting better than disturbance front position of the disturbance location of optimal bat individual
Optimal bat individual is moved to disturbance location by response, and otherwise optimal bat body position is motionless;
Wherein Random2For equally distributed random number, E on closed interval [0,1]iFor the intensity of sound of i-th bat;
Step 3.7): the fitness of all bats individual of current population and the population's fitness variance of bat population are calculated;
Earliness is judged according to the population's fitness variance of bat population, if bat population's fitness variance is less than given threshold
Value carries out intersection and mutation operation to all bat individuals, and goes to step 3.5), otherwise, selects optimal bat individual, goes to
Step 3.8);
Step 3.8): judge whether to reach maximum number of iterations or maximum search precision, if so, exporting optimal bat body position
The threshold value and weight of PID neural network described in the corresponding forecasting wind speed model based on PID neural network, if it is not, t3=t3+
1, go to step 3.4).
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810846131.XA CN109063907B (en) | 2018-07-27 | 2018-07-27 | Intelligent traversal large-step-length prediction method for maximum wind speed along high-speed railway |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810846131.XA CN109063907B (en) | 2018-07-27 | 2018-07-27 | Intelligent traversal large-step-length prediction method for maximum wind speed along high-speed railway |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109063907A true CN109063907A (en) | 2018-12-21 |
CN109063907B CN109063907B (en) | 2020-06-05 |
Family
ID=64836724
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810846131.XA Active CN109063907B (en) | 2018-07-27 | 2018-07-27 | Intelligent traversal large-step-length prediction method for maximum wind speed along high-speed railway |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109063907B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109615027A (en) * | 2018-12-29 | 2019-04-12 | 中南大学 | A kind of line of high-speed railway wind speed feature extraction intelligent Forecasting |
CN113482853A (en) * | 2021-08-06 | 2021-10-08 | 贵州大学 | Yaw control method, system, electronic equipment and storage medium |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2006083943A2 (en) * | 2005-02-01 | 2006-08-10 | Windlogics, Inc. | System and method for enhanced measure-correlate-predict for a wind farm location |
CN104612898A (en) * | 2014-11-27 | 2015-05-13 | 江苏科技大学 | Wind power variable-pitch multi-variable fuzzy neural network PID control method |
CN106772695A (en) * | 2016-11-14 | 2017-05-31 | 中南大学 | A kind of Along Railway wind speed forecasting method for merging many air measuring station measured datas |
CN106779151A (en) * | 2016-11-14 | 2017-05-31 | 中南大学 | A kind of line of high-speed railway wind speed multi-point multi-layer coupling prediction method |
CN106779148A (en) * | 2016-11-14 | 2017-05-31 | 中南大学 | A kind of method for forecasting wind speed of high speed railway line of multi-model multiple features fusion |
-
2018
- 2018-07-27 CN CN201810846131.XA patent/CN109063907B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2006083943A2 (en) * | 2005-02-01 | 2006-08-10 | Windlogics, Inc. | System and method for enhanced measure-correlate-predict for a wind farm location |
CN104612898A (en) * | 2014-11-27 | 2015-05-13 | 江苏科技大学 | Wind power variable-pitch multi-variable fuzzy neural network PID control method |
CN106772695A (en) * | 2016-11-14 | 2017-05-31 | 中南大学 | A kind of Along Railway wind speed forecasting method for merging many air measuring station measured datas |
CN106779151A (en) * | 2016-11-14 | 2017-05-31 | 中南大学 | A kind of line of high-speed railway wind speed multi-point multi-layer coupling prediction method |
CN106779148A (en) * | 2016-11-14 | 2017-05-31 | 中南大学 | A kind of method for forecasting wind speed of high speed railway line of multi-model multiple features fusion |
Non-Patent Citations (1)
Title |
---|
HUI LIU ET AL.: "Smart wind speed forecasting using EWT decomposition, GWO evolutionary optimization, RELM learning and IEWT reconstruction", 《ENERGY CONVERSION AND MANAGEMENT》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109615027A (en) * | 2018-12-29 | 2019-04-12 | 中南大学 | A kind of line of high-speed railway wind speed feature extraction intelligent Forecasting |
CN113482853A (en) * | 2021-08-06 | 2021-10-08 | 贵州大学 | Yaw control method, system, electronic equipment and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN109063907B (en) | 2020-06-05 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN112216124B (en) | Traffic signal control method based on deep reinforcement learning | |
CN103581188B (en) | A kind of network security situation prediction method and system | |
CN106779148B (en) | A kind of method for forecasting wind speed of high speed railway line of multi-model multiple features fusion | |
CN106205156B (en) | A kind of intersection self-healing combination control method for the mutation of part lane flow | |
CN103489039B (en) | There is the freeway traffic flow amount fusion forecasting method of online self-tuning optimization ability | |
CN105513350A (en) | Time-phased multi-parameter short-term traffic flow prediction method based on time-space characteristics | |
CN109215355A (en) | A kind of single-point intersection signal timing optimization method based on deeply study | |
CN106910337A (en) | A kind of traffic flow forecasting method based on glowworm swarm algorithm Yu RBF neural | |
Araghi et al. | Intelligent traffic light control of isolated intersections using machine learning methods | |
CN109214581A (en) | A kind of Along Railway wind speed forecasting method considering wind direction and confidence interval | |
CN106781465A (en) | A kind of road traffic Forecasting Methodology | |
CN109064748A (en) | Traffic average speed prediction method based on temporal clustering analysis and variable convolution neural network | |
CN104809895A (en) | Adjacent intersection arterial road coordinate control model and optimization method thereof | |
CN111582541A (en) | Firefly algorithm-based inland inundation model prediction method | |
CN104598765A (en) | Building energy consumption prediction method based on elastic adaptive neural network | |
CN109034476A (en) | A kind of line of high-speed railway extreme wind speeds big data Forecast method | |
CN109063907A (en) | A kind of big step-ahead prediction method of line of high-speed railway extreme wind speed intelligence traversal | |
CN112215435A (en) | Urban congestion propagation mode prediction method based on cyclic autoregressive model | |
CN109492768A (en) | Air Quality Forecast model training method and device | |
CN113436443A (en) | Distributed traffic signal control method based on generation of countermeasure network and reinforcement learning | |
CN109034478A (en) | A kind of line of high-speed railway strong wind iteration competition high-precision forecasting method | |
CN106781464A (en) | A kind of congestion in road situation method of testing | |
CN106228232A (en) | A kind of dynamic multi-objective based on fuzzy reasoning Population forecast strategy teaching optimization method | |
CN114495499A (en) | Multi-target intelligent networked vehicle collaborative optimization control method | |
CN112907970A (en) | Variable lane steering control method based on vehicle queuing length change rate |
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 |