CN110197278A - Based on the Air Quality Forecast method for improving chicken group's algorithm optimization BP neural network - Google Patents
Based on the Air Quality Forecast method for improving chicken group's algorithm optimization BP neural network Download PDFInfo
- Publication number
- CN110197278A CN110197278A CN201910480366.6A CN201910480366A CN110197278A CN 110197278 A CN110197278 A CN 110197278A CN 201910480366 A CN201910480366 A CN 201910480366A CN 110197278 A CN110197278 A CN 110197278A
- Authority
- CN
- China
- Prior art keywords
- individual
- chicken group
- chicken
- neural network
- air quality
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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"
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Economics (AREA)
- General Engineering & Computer Science (AREA)
- Human Resources & Organizations (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Strategic Management (AREA)
- Software Systems (AREA)
- Entrepreneurship & Innovation (AREA)
- Mathematical Physics (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Development Economics (AREA)
- Biophysics (AREA)
- Game Theory and Decision Science (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of based on the Air Quality Forecast method for improving chicken group's algorithm optimization BP neural network, the main indicator PM2.5 and PM10 that influence air quality index AQI have been extracted as prediction target, and has carried out PM2.5 or PM10 respectively in connection with 5 kinds of meteorological factors and predicts.The disadvantages of present invention is slow, convergence precision is not high for existing BP neural network convergence rate, utilize improvement chicken group's algorithm optimization BP neural network, BP neural network is set to effectively improve neural network prediction speed and precision according to the weight and threshold value of the prediction result adaptively changing neural network of every step.The method of self-adaptive BP neural networks prediction air quality of the invention has many advantages, such as precision of prediction height, fast convergence rate.
Description
Technical field
The invention belongs to Air Quality Forecast technical fields, and in particular to one kind is neural based on chicken group's algorithm optimization BP is improved
The design of the Air Quality Forecast method of network.
Background technique
In recent years, with urbanization and industrialized fast development, so that the air pollution situation day in many areas in China
Beneficial severe, air quality allows of no optimist, these air pollutants cause very big influence to human health.Therefore, how effectively
Prevent the mankind from having received widespread attention far from the harm of air pollution.However, realizing the improvement to air pollution in a short time
It is very difficult.Therefore, by predicting the air quality in following a period of time, air pollution is prevented and treated in time, is prevented
The generation only seriously polluted improves extensive concern of the public life quality by society.
Currently, predicting air quality usually using BP neural network, but it is to restrain the shortcomings that BP neural network
Speed is slow and convergence precision is not high, and it is not accurate enough that this will lead to final Air Quality Forecast result.It is existing for this problem
Technology is mostly optimized using parameter of the grid-search algorithms to BP neural network, but grid-search algorithms search speed
Slowly, effect of optimization is undesirable.
Summary of the invention
The purpose of the present invention is to solve existing BP neural networks to the problem of the prediction result inaccuracy of air quality,
It proposes a kind of based on the Air Quality Forecast method for improving chicken group's algorithm optimization BP neural network.
The technical solution of the present invention is as follows: based on the Air Quality Forecast method for improving chicken group's algorithm optimization BP neural network,
The following steps are included:
S1, original meteorological data is obtained, and original meteorological data is pre-processed, obtain training sample.
The weight and threshold value of S2, Initialize installation list hidden layer BP neural network.
S3, training sample is inputted into current single hidden layer BP neural network, obtains training result.
S4, judge whether current single hidden layer BP neural network restrains according to training result, it is no if then entering step S6
Then enter step S5.
S5, the weight and threshold value of single hidden layer BP neural network are updated using improvement chicken group's algorithm, return step
S3。
S6, will currently single hidden layer BP neural network it is and meteorological data to be predicted is defeated as Air Quality Forecast model
Enter Air Quality Forecast model, obtains Air Quality Forecast result.
Further, original meteorological data and meteorological data to be predicted include air pressure, wind speed, relative humidity, precipitation,
Temperature, PM2.5 parameter and PM10 parameter, training result and Air Quality Forecast result include PM2.5 parameter and PM10 parameter.
Further, pretreated method is carried out to original meteorological data in step S1 specifically:
Interpolation supplement, interpolation formula are carried out to the abnormal data in original meteorological data using Lagrange's interpolation algorithm
Are as follows:
Wherein L (a) indicates Lagrange's interpolation as a result, a indicates interpolation moment, ai,ajRespectively indicate i-th of moment and jth
A moment, biIndicate the original meteorological data at i-th of moment, i=0,1,2 ..., n0；J=0,1,2 ..., n0, n0It is original
The quantity of meteorological data.
Further, the whether convergent method of current single hidden layer BP neural network is judged in step S4 specifically:
Whether the error between training of judgement result and actual air mass parameter is less than preset error threshold, if then
Current list hidden layer BP neural network convergence, otherwise current single hidden layer BP neural network does not restrain.
Further, step S5 include it is following step by step:
S501, the weight of single hidden layer BP neural network and threshold value are arranged in order, obtain m dimensional vector X, wherein m is
The sum of weight and number of thresholds of single hidden layer BP neural network.
S502, a vector X is individual as a chicken group, the individual amount N of Initialize installation chicken group, building chicken group.
S503, the fitness value for calculating each individual in chicken group, and by the sequence of fitness value from small to large to chicken group
Body is ranked up, and regard preceding RN chicken group individual as cock, rear CN chicken group individual is used as chicken, and intermediate HN chicken group individual is made
For hen, wherein RN+HN+CN=N.
S504, the mother and sons' hierarchical relationship for determining chicken and hen in chicken group, hen quantity MN of the Initialize installation with chicken,
Wherein MN≤HN.
The position of each individual in S505, Initialize installation chicken group.
S506, setting maximum number of iterations tmaxWith iteration step length G, and Initialize installation the number of iterations t=1.
S507, judge whether current iteration number can be iterated step-length G and divide exactly, if then entering step S509, otherwise into
Enter step S508.
S508, the number of iterations t is enabled to add 1, return step S507.
S509, the position for successively updating each cock, and calculate each cock and update the fitness value behind position.
S510, judge whether the fitness value after each cock updates position is less than current adaptive optimal control angle value, if then
Adaptive optimal control angle value is updated, S511 is entered step, is otherwise directly entered step S511.
S511, the position for successively updating each hen, and calculate each hen and update the fitness value behind position.
S512, judge whether the fitness value after each hen updates position is less than current adaptive optimal control angle value, if then
Adaptive optimal control angle value is updated, S513 is entered step, is otherwise directly entered step S513.
S513, the position for successively updating each chicken, and calculate each chicken and update the fitness value behind position.
S514, judge whether the fitness value after each chicken updates position is less than current adaptive optimal control angle value, if then
Adaptive optimal control angle value is updated, S515 is entered step, is otherwise directly entered step S515.
S515, judge whether current iteration number reaches maximum number of iterations tmaxIf then entering step S516, otherwise
Return step S507.
The weight and threshold value of single hidden layer BP neural network corresponding to S516, chicken group's individual by adaptive optimal control angle value are made
It updates for weight and threshold value as a result, return step S3.
Further, the calculation formula of chicken group ideal adaptation angle value are as follows:
Wherein fiIndicate the fitness value of i-th of chicken group's individual,Indicate implicit using the corresponding list of i-th of chicken group's individual
The weight and the obtained training result of threshold value of layer BP neural network, yiFor corresponding actual result, i=1,2 ..., N.
Further, in step S505 in Initialize installation chicken group each body position formula are as follows:
WhereinIndicate the initial position of jth dimension variable in i-th of chicken group's individual, i=1,2 ..., N；J=1,2 ...,
M, [xdown,xup] beDomain, β be chaos random number, value range be (0,1).
Further, in step S509 cock position more new formula are as follows:
WhereinIndicate the position of jth dimension variable in i-th of chicken group's individual when the t times iteration,It indicates to change for the t-1 times
For when i-th chicken group's individual in jth dimension variable position, i=1,2 ..., RN；J=1,2 ..., m, Randn (0, δ2) it is clothes
It is 0 from mean value, standard deviation δ2Gaussian Profile, in which:
Wherein exp () is exponential function, fi,fkRespectively indicate the adaptation of i-th of chicken group individual and k-th of chicken group's individual
Angle value, k ∈ [1, RN], k ≠ i, ε are constant.
Further, in step S511 hen position more new formula are as follows:
WhereinIndicate the position of jth dimension variable in i-th of chicken group's individual when the t times iteration,It indicates to change for the t-1 times
For when i-th chicken group's individual in jth dimension variable position,Indicate that jth is tieed up in the r1 chicken group's individual when the t-1 times iteration
The position of variable,Indicate the position of jth dimension variable in the r2 chicken group's individual when the t-1 times iteration, r1 is i-th hen
Cock in group where itself, r2 are any individual randomly selected in cock and hen in entire chicken group, r1 ≠ r2, i=1,
2,...,HN；J=1,2 ..., m, C (x0, γ) and it is that Cauchy is distributed, x0, γ is respectively the location parameter and scale ginseng of Cauchy's distribution
Number, S1 and S2 are personal best particle control parameter and global optimum's position control parameter, and:
S1=exp ((fi-fr1)/(abs(fi)+ε))
S2=exp (fr2-fi)
Wherein exp () is exponential function, and abs () is ABS function, fi,fr1,fr2Respectively indicate i-th of chicken group
The fitness value of body, the r1 chicken group individual and the r2 chicken group's individual, ε is constant.
Further, in step S513 chicken position more new formula are as follows:
WhereinIndicate the position of jth dimension variable in i-th of chicken group's individual when the t times iteration,It indicates to change for the t-1 times
For when i-th chicken group's individual in jth dimension variable position,Indicate that jth dimension becomes in h-th of chicken group's individual when the t-1 times iteration
The position of amount, h are mother of i-th of chicken, i=1,2 ..., CN；H=1,2 ..., MN, FL are the random number between 0~2.
The beneficial effects of the present invention are:
(1) present invention optimizes the weight and threshold value of BP neural network using chicken group's algorithm is improved, and efficiently solves
The problem of BP neural network learning process convergence rate is slow, and network training is easily trapped into locally optimal solution, simultaneously because improving chicken
Group's algorithm can be adaptive Optimized BP Neural Network weight and threshold value, therefore by the BP neural network after weight and threshold optimization
As Air Quality Forecast model, more accurate Air Quality Forecast result can be obtained.
(2) the Air Quality Forecast model that the present invention obtains can be in conjunction with 5 kinds of meteorological factors, to influence air quality
PM2.5 and PM10 parameter is effectively predicted, and obtained Air Quality Forecast result is more accurate and objective.
Detailed description of the invention
Fig. 1 show provided in an embodiment of the present invention pre- based on the air quality for improving chicken group's algorithm optimization BP neural network
Survey method flow diagram.
Fig. 2 show single hidden layer BP neural network structural schematic diagram provided in an embodiment of the present invention.
Fig. 3 show the flow chart step by step of step S5 provided in an embodiment of the present invention.
Specific embodiment
Carry out detailed description of the present invention illustrative embodiments with reference to the drawings.It should be appreciated that shown in attached drawing and
The embodiment of description is only exemplary, it is intended that is illustrated the principle and spirit of the invention, and is not limited model of the invention
It encloses.
The embodiment of the invention provides a kind of based on the Air Quality Forecast side for improving chicken group's algorithm optimization BP neural network
Method, as shown in Figure 1, including the following steps S1~S6:
S1, original meteorological data is obtained, and original meteorological data is pre-processed, obtain training sample.
Air quality index (Air Quality Index, AQI) is the zero dimension index for comprehensively considering air quality, is fitted
Conjunction illustrates Air Quality Change Trend short-term in city, its calculation formula is:
A QI=max { IAQI1,IAQI2,IAQI3,…,IAQIn} (1)
Wherein IAQIiIndicate that air quality separate index number, i=0,1,2 ..., n, n are pollutant project sum.AQI is exactly
Maximum value in the IAQI of every pollutant.Select the main indicator PM2.5 and PM10 for wherein influencing AQI pre- as air quality
The target of survey, and combine air pressure, wind speed, relative humidity, precipitation and temperature this five kinds of meteorological factors as the defeated of neural network
Enter to be predicted, and then obtains air quality AQI.Therefore, in the embodiment of the present invention, original meteorological data include air pressure, wind speed,
Relative humidity, precipitation, temperature, PM2.5 parameter and PM10 parameter.
Due to there may be more abnormal data, for example missing values or not meeting data standard in original meteorological data
Data, it is therefore desirable to it be pre-processed, in the embodiment of the present invention, it is specific that pretreated method is carried out to original meteorological data
Are as follows:
Interpolation supplement, interpolation formula are carried out to the abnormal data in original meteorological data using Lagrange's interpolation algorithm
Are as follows:
Wherein L (a) indicates Lagrange's interpolation as a result, a indicates interpolation moment, ai,ajRespectively indicate i-th of moment and jth
A moment, biIndicate the original meteorological data at i-th of moment, i=0,1,2 ..., n0；J=0,1,2 ..., n0, n0It is original
The quantity of meteorological data.
The weight and threshold value of S2, Initialize installation list hidden layer BP neural network.
BP neural network (Back Propagation Neural Network, BPNN) is reversely passed as one kind by error
The feedforward neural network broadcast has very strong interpretability to complicated non-linear relation, its own also possesses good self-organizing
Ability and robustness.The embodiment of the present invention is solved relevant meteorological factor and can be inhaled using single hidden layer BP neural network model
Enter the Nonlinear Mapping relationship between particulate matter PM10, fine particle PM2.5, particle concentration is predicted, such as Fig. 2 institute
Show.
S3, training sample is inputted into current single hidden layer BP neural network, obtains training result.
In the embodiment of the present invention, training result includes PM2.5 parameter and PM10 parameter.
S4, judge whether current single hidden layer BP neural network restrains according to training result, it is no if then entering step S6
Then enter step S5.
In the embodiment of the present invention, the whether convergent method of current single hidden layer BP neural network is judged specifically:
Whether the error between training of judgement result and actual air mass parameter is less than preset error threshold, if then
Current list hidden layer BP neural network convergence, otherwise current single hidden layer BP neural network does not restrain.
S5, the weight and threshold value of single hidden layer BP neural network are updated using improvement chicken group's algorithm, return step
S3。
As a kind of domestic animal that distribution is most wide, chicken itself and their egg are saved mainly as food source.Family chicken is
Gregarious birds, live in a group together.Their cognitive ability is very strong, can be mutual after some months of living apart
Identification, there is 30 multiple and different exchange sound, such as cackle, chirp and crying, mutually transmits many in relation to nesting, eating
The information of object discovery, mating and danger.Except through attempting also use for reference previous experience and other chickens with mistake study, chicken
Decision experience.Hierarchy plays an important role in the social life of chicken, and the chicken in advantage chicken group can dominate weak person.In population
There is the hens of more prevailing close head cock, and cock of the hen being more obedient to station in group periphery.It moves
Chicken temporary breaks one existing mass society order of meeting is removed or adds, until establishing new hierarchy.Advantage individual can
Preferentially to obtain food, and cock may allow their companion first to eat when finding food.This magnanimous behavior there is also
In hen when raising child.However, such case is not present between the individual of different groups, as other chicken groups
When individual invades their territory, chicken group will issue huge shriek.In general, the behavior of chicken is different because of gender, and head is public
Chicken is certain to search of food, fights with the chicken for invading the tissue territory.As for chicken, they are in mother's search of food at one's side, often
Chicken is all too simple, cannot work in coordination.However, they may coordinate certainly as a team as a group
Oneself, the search of food under specific order of grade.This colony intelligence can be associated with target problem to be optimized.
In order to simplify the simulation to chicken group's behavior, chicken group behavior will be carried out as follows pumping in the embodiment of the present invention
As:
(1) there are several groups in Ji Qunzhong, each group includes a head cock, several hens and several small
Chicken.How chicken group is divided into several groups, and determines the identity of chicken (cock, hen and chicken), both depends on the fitness value of chicken.
The best chicken of several fitness values can be taken as cock, wherein every chicken is all one group of head cock, several fitness values are worst
Chicken will be designated as chicken, remaining is hen.Which group hen random selection will live in, between hen and chicken
Mother and sons' hierarchical relationship is also to establish at random.
(2) order of grade in group, dominance relation and mother and sons' hierarchical relationship are constant, these states are only in each time step
It updates once, which is iteration step length G.
(3) chicken follows its companion's cock search of food, but other companions may be prevented to eat oneself food.Assuming that chicken meeting
Randomly steal the good food that others has found, chicken search of food around mother (hen).Advantage individual is in flood competition
In have advantage.Assuming that RN, HN, CN and MN respectively indicate the quantity of cock, hen, chicken and the hen with chicken.Possess phase
Cock will be considered to RN chicken of degree of being preferably adapted to, chicken will be considered by possessing CN chicken of relatively worst fitness value, be remained
Under will be considered hen.
Based on above-mentioned setting, as shown in figure 3, step S5 includes following S501~S516 step by step:
S501, the weight of single hidden layer BP neural network and threshold value are arranged in order, obtain m dimensional vector X, wherein m is
The sum of weight and number of thresholds of single hidden layer BP neural network.
In the embodiment of the present invention, as illustrated in fig. 2, it is assumed that single hidden layer BP neural network has n1A input layer, nNode
A hidden layer node and 1 output node layer, then have n between input layer and hidden layer1* nNode weight, hidden layer and output
There is nNode weight between layer, there is nNode threshold value, therefore m=n in hidden layer1* nNode+nNode+nNode=(n1+2)*
nNode。
S502, a vector X is individual as a chicken group, the individual amount N of Initialize installation chicken group, building chicken group.
I-th of individual is in the corresponding vector of the t times iteration in chicken groupWherein i=
1,2,...N；J=1,2 ..., m.When building chicken group it should also be expressly that chicken group's migration boundary, i.e.,
S503, the fitness value for calculating each individual in chicken group, and by the sequence of fitness value from small to large to chicken group
Body is ranked up, and regard preceding RN chicken group individual as cock, rear CN chicken group individual is used as chicken, and intermediate HN chicken group individual is made
For hen, wherein RN+HN+CN=N.
In the embodiment of the present invention, using the fitness value of chicken group's individual as objective function, its calculation formula is:
Wherein fiIndicate the fitness value of i-th of chicken group's individual,Indicate implicit using the corresponding list of i-th of chicken group's individual
The weight and the obtained training result of threshold value of layer BP neural network, yiFor corresponding actual result, i=1,2 ..., N.
S504, the mother and sons' hierarchical relationship for determining chicken and hen in chicken group, hen quantity MN of the Initialize installation with chicken,
Wherein MN≤HN.
Artificial rearing chicken is primarily to food, and as food itself, only hen could lay eggs, this is also coming for food
Source.Therefore foster mother chicken is relatively beneficial to the mankind than feeding cock, therefore HN should be bigger than RN.In view of individual difference, and not all mother
Chicken can simultaneously incubation, therefore, HN should also be as be greater than MN.Although every hen can raise one or more chicken, this hair
In bright embodiment assume Adult Chicken quantity by be more than chicken quantity.Therefore, in the embodiment of the present invention, the number parameter of chicken group
Setting is as shown in table 1.
Table 1
The position of each individual in S505, Initialize installation chicken group.
After determining chicken group migration boundary, chicken group's initialized location has larger impact to algorithm optimizing convergence rate,
Chicken group's distribution is more uniform, and algorithmic statement performance totally turns for the better.Therefore in the embodiment of the present invention, per each and every one in Initialize installation chicken group
The formula of body position are as follows:
WhereinIndicate the initial position of jth dimension variable in i-th of chicken group's individual, i=1,2 ..., N；J=1,2 ...,
M, [xdown,xup] beDomain, β be chaos random number, value range be (0,1), than rand random number ergodic,
Randomness is stronger.
S506, setting maximum number of iterations tmaxWith iteration step length G, and Initialize installation the number of iterations t=1.
Maximum number of iterations t in the embodiment of the present inventionmax=300, iteration step length G should be set as a value appropriate, such as
Fruit G value is very big, then is unfavorable for algorithm and rapidly converges to global optimum, and when G value very little, which may fall into part most again
It is excellent.Through overtesting it is found that G ∈ [2,20] can obtain good effect in most problems.
S507, judge whether current iteration number can be iterated step-length G and divide exactly, if then entering step S509, otherwise into
Enter step S508.
S508, the number of iterations t is enabled to add 1, return step S507.
S509, the position for successively updating each cock, and calculate each cock and update the fitness value behind position.
The cock for possessing optimal adaptation angle value is easier to get food compared to the cock for possessing poor fitness value, because
This, the cock for possessing preferable fitness value can arrive wider region and find, and placement optimization more new formula is as follows:
WhereinIndicate the position of jth dimension variable in i-th of chicken group's individual when the t times iteration,It indicates to change for the t-1 times
For when i-th chicken group's individual in jth dimension variable position, i=1,2 ..., RN；J=1,2 ..., m, Randn (0, δ2) it is clothes
It is 0 from mean value, standard deviation δ2Gaussian Profile, in which:
Wherein exp () is exponential function, fi,fkRespectively indicate the adaptation of i-th of chicken group individual and k-th of chicken group's individual
Angle value, k ∈ [1, RN], k ≠ i, ε are the constant of very little, are avoided the occurrence of except 0 mistake.
S510, judge whether the fitness value after each cock updates position is less than current adaptive optimal control angle value, if then
Adaptive optimal control angle value is updated, S511 is entered step, is otherwise directly entered step S511.
S511, the position for successively updating each hen, and calculate each hen and update the fitness value behind position.
Hen can follow their companion's cock search of food.It is found in addition, they can also steal other chickens at random
Good food, although other chickens can suppress them.The hen that more hen of advantage possesses optimal adaptation angle value is scrambling for food
Aspect is advantageously than more docile hen.Therefore, these phenomenons can be simulated as follows with mathematical formulae:
WhereinIndicate the position of jth dimension variable in i-th of chicken group's individual when the t times iteration,It indicates to change for the t-1 times
For when i-th chicken group's individual in jth dimension variable position,Indicate that jth is tieed up in the r1 chicken group's individual when the t-1 times iteration
The position of variable,Indicate the position of jth dimension variable in the r2 chicken group's individual when the t-1 times iteration, r1 is i-th hen
Cock in group where itself, r2 are any individual randomly selected in cock and hen in entire chicken group, r1 ≠ r2, i=1,
2,...,HN；The normal distribution of j=1,2 ..., m, Rand between [0,1], S1 and S2 are personal best particle control parameter
With global optimum's position control parameter, be used for and:
S1=exp ((fi-fr1)/(abs(fi)+ε)) (8)
S2=exp (fr2-fi) (9)
Wherein exp () is exponential function, and abs () is ABS function, fi,fr1,fr2Respectively indicate i-th of chicken group
The fitness value of body, the r1 chicken group individual and the r2 chicken group's individual, ε is the constant of very little, is avoided the occurrence of except 0 mistake.
S1 and S2 is used to control the close optimal location and global optimum position in oneself group of chicken group, does not stop to group
Interior optimal location and global optimum are located proximate to, but still lack randomness, therefore can improve hen move mode, enhance
The global optimizing ability of hen and random optimizing ability then improve mother due to the Cauchy function randomness that more preferable control is mobile
The mobile update mode of chicken position is as follows:
Wherein C (x0, γ) and it is that Cauchy is distributed, x0, γ is respectively the location parameter and scale parameter of Cauchy's distribution.
S512, judge whether the fitness value after each hen updates position is less than current adaptive optimal control angle value, if then
Adaptive optimal control angle value is updated, S513 is entered step, is otherwise directly entered step S513.
S513, the position for successively updating each chicken, and calculate each chicken and update the fitness value behind position.
For chicken around its mother's search of food, location update formula is as follows:
WhereinIndicate the position of jth dimension variable in i-th of chicken group's individual when the t times iteration,It indicates to change for the t-1 times
For when i-th chicken group's individual in jth dimension variable position,Indicate that jth dimension becomes in h-th of chicken group's individual when the t-1 times iteration
The position of amount, h are mother of i-th of chicken, i=1,2 ..., CN；H=1,2 ..., MN, FL are the random number between 0~2.
In the embodiment of the present invention, FL=g*Rand, g ∈ [0.4,1], normal distribution of the Rand between [0,1].
S514, judge whether the fitness value after each chicken updates position is less than current adaptive optimal control angle value, if then
Adaptive optimal control angle value is updated, S515 is entered step, is otherwise directly entered step S515.
S515, judge whether current iteration number reaches maximum number of iterations tmaxIf then entering step S516, otherwise
Return step S507.
The weight and threshold value of single hidden layer BP neural network corresponding to S516, chicken group's individual by adaptive optimal control angle value are made
It updates for weight and threshold value as a result, return step S3.
S6, will currently single hidden layer BP neural network it is and meteorological data to be predicted is defeated as Air Quality Forecast model
Enter Air Quality Forecast model, obtains Air Quality Forecast result.
In the embodiment of the present invention, as shown in Fig. 2, meteorological data to be predicted include air pressure, wind speed, relative humidity, precipitation,
Temperature, PM2.5 parameter and PM10 parameter, Air Quality Forecast result include PM2.5 parameter and PM10 parameter.
Those of ordinary skill in the art will understand that the embodiments described herein, which is to help reader, understands this hair
Bright principle, it should be understood that protection scope of the present invention is not limited to such specific embodiments and embodiments.This field
Those of ordinary skill disclosed the technical disclosures can make according to the present invention and various not depart from the other each of essence of the invention
The specific variations and combinations of kind, these variations and combinations are still within the scope of the present invention.
Claims (10)
1. based on the Air Quality Forecast method for improving chicken group's algorithm optimization BP neural network, which is characterized in that including following step
It is rapid:
S1, original meteorological data is obtained, and original meteorological data is pre-processed, obtain training sample；
The weight and threshold value of S2, Initialize installation list hidden layer BP neural network；
S3, training sample is inputted into current single hidden layer BP neural network, obtains training result；
S4, judge whether current single hidden layer BP neural network restrains according to training result, if then entering step S6, otherwise into
Enter step S5；
S5, the weight and threshold value of single hidden layer BP neural network are updated using improvement chicken group's algorithm, return step S3；
S6, will currently single hidden layer BP neural network is as Air Quality Forecast model, and by meteorological data to be predicted input sky
Gas quality prediction model obtains Air Quality Forecast result.
2. Air Quality Forecast method according to claim 1, which is characterized in that the original meteorological data with it is to be predicted
Meteorological data includes air pressure, wind speed, relative humidity, precipitation, temperature, PM2.5 parameter and PM10 parameter, the training result
It include PM2.5 parameter and PM10 parameter with Air Quality Forecast result.
3. Air Quality Forecast method according to claim 1, which is characterized in that original meteorological number in the step S1
According to the pretreated method of progress specifically:
Interpolation supplement, interpolation formula are carried out to the abnormal data in original meteorological data using Lagrange's interpolation algorithm are as follows:
Wherein L (a) indicates Lagrange's interpolation as a result, a indicates interpolation moment, ai,ajWhen respectively indicating i-th of moment and j-th
It carves, biIndicate the original meteorological data at i-th of moment, i=0,1,2 ..., n0；J=0,1,2 ..., n0, n0For original meteorology
The quantity of data.
4. Air Quality Forecast method according to claim 1, which is characterized in that judge in the step S4 current single hidden
The whether convergent method of BP neural network containing layer specifically:
Whether the error between training of judgement result and actual air mass parameter is less than preset error threshold, if then current
Single hidden layer BP neural network convergence, otherwise current single hidden layer BP neural network does not restrain.
5. Air Quality Forecast method according to claim 1, which is characterized in that the step S5 includes following substep
It is rapid:
S501, the weight of single hidden layer BP neural network and threshold value are arranged in order, obtain m dimensional vector X, wherein m is single hidden
The sum of the weight of the BP neural network containing layer and number of thresholds；
S502, a vector X is individual as a chicken group, the individual amount N of Initialize installation chicken group, building chicken group；
S503, the fitness value for calculating each individual in chicken group, and by the sequence of fitness value from small to large it is individual to chicken group into
Row sequence regard preceding RN chicken group individual as cock, and rear CN chicken group individual is used as chicken, and intermediate HN chicken group individual is as female
Chicken, wherein RN+HN+CN=N；
S504, the mother and sons' hierarchical relationship for determining chicken and hen in chicken group, hen quantity MN of the Initialize installation with chicken, wherein
MN≤HN；
The position of each individual in S505, Initialize installation chicken group；
S506, setting maximum number of iterations tmaxWith iteration step length G, and Initialize installation the number of iterations t=1；
S507, judge whether current iteration number can be iterated step-length G and divide exactly, if then entering step S509, otherwise enter step
Rapid S508；
S508, the number of iterations t is enabled to add 1, return step S507；
S509, the position for successively updating each cock, and calculate each cock and update the fitness value behind position；
S510, judge whether the fitness value after each cock updates position is less than current adaptive optimal control angle value, if then updating
Adaptive optimal control angle value, enters step S511, is otherwise directly entered step S511；
S511, the position for successively updating each hen, and calculate each hen and update the fitness value behind position；
S512, judge whether the fitness value after each hen updates position is less than current adaptive optimal control angle value, if then updating
Adaptive optimal control angle value, enters step S513, is otherwise directly entered step S513；
S513, the position for successively updating each chicken, and calculate each chicken and update the fitness value behind position；
S514, judge whether the fitness value after each chicken updates position is less than current adaptive optimal control angle value, if then updating
Adaptive optimal control angle value, enters step S515, is otherwise directly entered step S515；
S515, judge whether current iteration number reaches maximum number of iterations tmaxIf then entering step S516, otherwise return
Step S507；
S516, using the weight of single hidden layer BP neural network corresponding to the chicken of adaptive optimal control angle value group's individual and threshold value as weighing
Weight and threshold value update as a result, return step S3.
6. Air Quality Forecast method according to claim 5, which is characterized in that the meter of the chicken group ideal adaptation angle value
Calculate formula are as follows:
Wherein fiIndicate the fitness value of i-th of chicken group's individual,It indicates using the corresponding single hidden layer BP of i-th of chicken group's individual
The obtained training result of weight and threshold value of neural network, yiFor corresponding actual result, i=1,2 ..., N.
7. Air Quality Forecast method according to claim 5, which is characterized in that Initialize installation in the step S505
The formula of each body position in chicken group are as follows:
WhereinIndicate the initial position of jth dimension variable in i-th of chicken group's individual, i=1,2 ..., N；J=1,2 ..., m,
[xdown,xup] beDomain, β be chaos random number, value range be (0,1).
8. Air Quality Forecast method according to claim 5, which is characterized in that cock position in the step S509
More new formula are as follows:
WhereinIndicate the position of jth dimension variable in i-th of chicken group's individual when the t times iteration,It indicates when the t-1 times iteration
The position of jth dimension variable, i=1,2 ..., RN in i chicken group's individual；J=1,2 ..., m, Randn (0, δ2) it is to obey mean value
It is 0, standard deviation δ2Gaussian Profile, in which:
Wherein exp () is exponential function, fi,fkThe fitness value of i-th of chicken group individual and k-th of chicken group's individual is respectively indicated,
K ∈ [1, RN], k ≠ i, ε are constant.
9. Air Quality Forecast method according to claim 5, which is characterized in that hen position in the step S511
More new formula are as follows:
WhereinIndicate the position of jth dimension variable in i-th of chicken group's individual when the t times iteration,When indicating the t-1 times iteration
The position of jth dimension variable in i-th of chicken group's individual,Indicate that jth ties up variable in the r1 chicken group's individual when the t-1 times iteration
Position,Indicate the position of jth dimension variable in the r2 chicken group's individual when the t-1 times iteration, r1 is i-th hen itself
Cock in the group of place, r2 are any individual randomly selected in cock and hen in entire chicken group, r1 ≠ r2, i=1,
2,...,HN；J=1,2 ..., m, C (x0, γ) and it is that Cauchy is distributed, x0, γ is respectively the location parameter and scale ginseng of Cauchy's distribution
Number, S1 and S2 are personal best particle control parameter and global optimum's position control parameter, and:
S1=exp ((fi-fr1)/(abs(fi)+ε))
S2=exp (fr2-fi)
Wherein exp () is exponential function, and abs () is ABS function, fi,fr1,fr2Respectively indicate i-th chicken group individual,
The fitness value of the r1 chicken group individual and the r2 chicken group's individual, ε is constant.
10. Air Quality Forecast method according to claim 5, which is characterized in that chicken position in the step S513
More new formula are as follows:
WhereinIndicate the position of jth dimension variable in i-th of chicken group's individual when the t times iteration,It indicates when the t-1 times iteration
The position of jth dimension variable in i chicken group's individual,Indicate the position of jth dimension variable in h-th of chicken group's individual when the t-1 times iteration
It sets, h is mother of i-th of chicken, i=1,2 ..., CN；H=1,2 ..., MN, FL are the random number between 0~2.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910480366.6A CN110197278A (en) | 2019-06-04 | 2019-06-04 | Based on the Air Quality Forecast method for improving chicken group's algorithm optimization BP neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910480366.6A CN110197278A (en) | 2019-06-04 | 2019-06-04 | Based on the Air Quality Forecast method for improving chicken group's algorithm optimization BP neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110197278A true CN110197278A (en) | 2019-09-03 |
Family
ID=67753853
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910480366.6A Pending CN110197278A (en) | 2019-06-04 | 2019-06-04 | Based on the Air Quality Forecast method for improving chicken group's algorithm optimization BP neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110197278A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110796234A (en) * | 2019-10-21 | 2020-02-14 | 苏州浪潮智能科技有限公司 | Method and device for predicting computer state |
| CN110889228A (en) * | 2019-11-28 | 2020-03-17 | 国网吉林省电力有限公司电力科学研究院 | Transformer oil paper insulation aging prediction method based on chicken swarm optimization BP neural network |
| CN111400963A (en) * | 2020-03-04 | 2020-07-10 | 山东师范大学 | Crowd evacuation simulation method and system based on chicken swarm algorithm and social force model |
| CN114839317A (en) * | 2021-12-27 | 2022-08-02 | 北京雪迪龙科技股份有限公司 | Atmosphere gridding comprehensive early warning method and system thereof |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101247039A (en) * | 2007-12-14 | 2008-08-20 | 南方电网技术研究中心 | Method for power system wave record playback based on real-time simulation system |
| CN101656538A (en) * | 2009-08-21 | 2010-02-24 | 北京大学深圳研究生院 | Lagrange interpolation method-based time mismatch real time compensation algorithm of time interlace analog digital converter (TIADC) system |
| CN103197123A (en) * | 2013-03-19 | 2013-07-10 | 中国西电电气股份有限公司 | Method of computing synthetic test attenuation-type recovery voltage |
| KR20130090278A (en) * | 2012-02-03 | 2013-08-13 | 이메트릭스 주식회사 | Real time cotrolling system and method based area modeling |
| CN105044524A (en) * | 2015-08-10 | 2015-11-11 | 许继集团有限公司 | Lightning arrester monitoring method applicable to intelligent transformer station and system |
| CN105353255A (en) * | 2015-11-27 | 2016-02-24 | 南京邮电大学 | Transformer fault diagnosis method based on neural network |
| CN106499656A (en) * | 2016-11-04 | 2017-03-15 | 安徽理工大学 | A kind of fan wind speed intelligent control method |
| CN108875793A (en) * | 2018-05-25 | 2018-11-23 | 云南电网有限责任公司电力科学研究院 | A kind of dirty area's grade appraisal procedure based on CSO-LSSVM |
| CN109492768A (en) * | 2018-12-20 | 2019-03-19 | 成都四方伟业软件股份有限公司 | Air Quality Forecast model training method and device |
| CN109819030A (en) * | 2019-01-22 | 2019-05-28 | 西北大学 | A kind of preparatory dispatching method of data resource based on edge calculations |
-
2019
- 2019-06-04 CN CN201910480366.6A patent/CN110197278A/en active Pending
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101247039A (en) * | 2007-12-14 | 2008-08-20 | 南方电网技术研究中心 | Method for power system wave record playback based on real-time simulation system |
| CN101656538A (en) * | 2009-08-21 | 2010-02-24 | 北京大学深圳研究生院 | Lagrange interpolation method-based time mismatch real time compensation algorithm of time interlace analog digital converter (TIADC) system |
| KR20130090278A (en) * | 2012-02-03 | 2013-08-13 | 이메트릭스 주식회사 | Real time cotrolling system and method based area modeling |
| CN103197123A (en) * | 2013-03-19 | 2013-07-10 | 中国西电电气股份有限公司 | Method of computing synthetic test attenuation-type recovery voltage |
| CN105044524A (en) * | 2015-08-10 | 2015-11-11 | 许继集团有限公司 | Lightning arrester monitoring method applicable to intelligent transformer station and system |
| CN105353255A (en) * | 2015-11-27 | 2016-02-24 | 南京邮电大学 | Transformer fault diagnosis method based on neural network |
| CN106499656A (en) * | 2016-11-04 | 2017-03-15 | 安徽理工大学 | A kind of fan wind speed intelligent control method |
| CN108875793A (en) * | 2018-05-25 | 2018-11-23 | 云南电网有限责任公司电力科学研究院 | A kind of dirty area's grade appraisal procedure based on CSO-LSSVM |
| CN109492768A (en) * | 2018-12-20 | 2019-03-19 | 成都四方伟业软件股份有限公司 | Air Quality Forecast model training method and device |
| CN109819030A (en) * | 2019-01-22 | 2019-05-28 | 西北大学 | A kind of preparatory dispatching method of data resource based on edge calculations |
Non-Patent Citations (4)
| Title |
|---|
| DONG LIU ET AL.: "Projection pursuit evaluation model of regional surface water environment based on improved chicken swarm optimization algorithm", 《WATER RESOURCES MANAGEMENT》 * |
| 张莹杰等: "改进约束鸡群算法在神经网络的应用", 《计算机工程与科学》 * |
| 蒋婷婷: "基于鸡群算法的小波神经网络短时交通流预测研究", 《中国优秀硕士学位论文全文数据库 工程科技II辑》 * |
| 郭德龙等: "基于柯西变异的果蝇优化算法", 《科技通报》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110796234A (en) * | 2019-10-21 | 2020-02-14 | 苏州浪潮智能科技有限公司 | Method and device for predicting computer state |
| CN110796234B (en) * | 2019-10-21 | 2022-07-12 | 苏州浪潮智能科技有限公司 | Method and device for predicting computer state |
| CN110889228A (en) * | 2019-11-28 | 2020-03-17 | 国网吉林省电力有限公司电力科学研究院 | Transformer oil paper insulation aging prediction method based on chicken swarm optimization BP neural network |
| CN110889228B (en) * | 2019-11-28 | 2023-04-18 | 国网吉林省电力有限公司电力科学研究院 | Transformer oil paper insulation aging prediction method based on chicken swarm optimization BP neural network |
| CN111400963A (en) * | 2020-03-04 | 2020-07-10 | 山东师范大学 | Crowd evacuation simulation method and system based on chicken swarm algorithm and social force model |
| CN114839317A (en) * | 2021-12-27 | 2022-08-02 | 北京雪迪龙科技股份有限公司 | Atmosphere gridding comprehensive early warning method and system thereof |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110197278A (en) | Based on the Air Quality Forecast method for improving chicken group's algorithm optimization BP neural network | |
| Nabe-Nielsen et al. | Effects of noise and by-catch on a Danish harbour porpoise population | |
| Rijnsdorp et al. | Modelling the spatial dynamics and fisheries of North Sea plaice (Pleuronectes platessa L.) based on tagging data | |
| CN109634098A (en) | A kind of fattening house environment conditioning system and method | |
| CN103336503B (en) | Aquaculture pond dissolved oxygen DO control system | |
| CN103971160B (en) | particle swarm optimization method based on complex network | |
| An et al. | Mussels wandering optimization: an ecologically inspired algorithm for global optimization | |
| Lu et al. | Vegetable price prediction based on pso-bp neural network | |
| CN109325580A (en) | A kind of adaptive cuckoo searching method for Services Composition global optimization | |
| CN103309370A (en) | Incubating room humidity control method based on BP (back-propagation) neural network | |
| CN115374689A (en) | Air quality index prediction method based on improved sparrow search algorithm optimization | |
| Føre et al. | Modelling of Atlantic salmon (Salmo salar L.) behaviour in sea-cages: Using artificial light to control swimming depth | |
| Araneda et al. | Optimal harvesting time of farmed aquatic populations with nonlinear size‐heterogeneous growth | |
| CN110147890A (en) | A kind of method and system based on lion group's algorithm optimization extreme learning machine integrated study | |
| CN110414115A (en) | A kind of wavelet neural network tomato yield prediction technique based on genetic algorithm | |
| CN115640956A (en) | Future water resource supply and demand balance analysis method | |
| CN109711592A (en) | A kind of pond water temperature prediction technique based on genetic algorithm optimization extreme learning machine | |
| Lough et al. | Spawning pattern variability and its effect on retention, larval growth and recruitment in Georges Bank cod and haddock | |
| CN114818324A (en) | Method, device, medium and equipment for automatically regulating and controlling water quantity and water quality of basin scale | |
| CN112836885B (en) | Combined load prediction method, combined load prediction device, electronic equipment and storage medium | |
| CN107156020B (en) | A kind of Intelligent fish tank water quality adjustment method based on intensified learning | |
| Subbiah et al. | A review of bio-inspired computational intelligence algorithms in electricity load forecasting | |
| CN117114915A (en) | Aquaculture PH value prediction method based on improved particle swarm optimization | |
| CN115150284B (en) | Communication network topology optimization method and system based on improved sparrow algorithm | |
| CN109343479A (en) | Flowshop production scheduling method and system based on the adaptive harmonic search algorithm of multiple target |
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 |