CN108805253A - A kind of PM2.5 concentration predictions method - Google Patents
A kind of PM2.5 concentration predictions method Download PDFInfo
- Publication number
- CN108805253A CN108805253A CN201710297281.5A CN201710297281A CN108805253A CN 108805253 A CN108805253 A CN 108805253A CN 201710297281 A CN201710297281 A CN 201710297281A CN 108805253 A CN108805253 A CN 108805253A
- Authority
- CN
- China
- Prior art keywords
- neural network
- wolf
- concentration
- wolves
- optimized
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Abstract
The present invention provides a kind of PM2.5 concentration predictions method, by combining grey wolf optimization algorithm and BP neural network, using the weights and threshold value of grey wolf optimization algorithm Optimized BP Neural Network, using the concentration of the model prediction PM2.5 after optimization.The present invention has the advantageous effect for improving PM2.5 concentration prediction accuracys rate.
Description
Technical field
The present invention relates to Air Quality Forecast technical fields, more particularly, to a kind of PM2.5 concentration predictions method.
Background technology
PM2.5 refers to the particulate matter that dynamics equivalent diameter is less than or equal to 2.5 microns in air.Its, face small with grain size
Product is big, activity is strong, easy the characteristics of being attached to poisonous and harmful substances.And atmospheric residence time is long, fed distance is remote, it can be direct
Into human lung, influence health.When PM2.5 concentration is higher in air, different degrees of haze weather can be formed, is dropped
Low latitude gas visibility.Influence the main dirty in by meteorologic factor (such as weather, temperature, wind speed, wind direction) and air of PM2.5 concentration
Contaminate object concentration (such as NOx、SO2、O3Deng) influence.PM2.5 brings to daily life and seriously affects.Pass through prediction
PM2.5 concentration takes measures to reduce bringing convenience to people's lives and trip.
At present neural network is mostly used to predict PM2.5 concentration.BP neural network is current most widely used god
Through one of network model.When PM2.5 is predicted, meteorological data (such as weather, temperature, wind speed, wind direction) and pollutants in air
Concentration (such as NOx、SO2、O3Deng) input as neural network, PM2.5 concentration exports as network.
However, when using Neural Network model predictive PM2.5 concentration, the weights and threshold value of neural network are not easy to train, net
Network is easily absorbed in local optimum, causes neural network model low to PM2.5 concentration prediction accuracys rate.
Invention content
The present invention solves the above problems in order to overcome the problems referred above or at least partly, provides a kind of PM2.5 concentration predictions
Method is selected a good training starting point to network, is avoided by the weights and threshold value of grey wolf optimization algorithm optimization neural network
Network is absorbed in local optimum.PM2.5 concentration prediction accuracys rate can be improved in neural network after this method optimizes.
According to an aspect of the present invention, a kind of PM2.5 concentration predictions method is provided, is included the following steps:
Step 1, wolf pack corresponding with BP neural network to be optimized is generated at random using grey wolf optimization algorithm;
Step 2, the training BP neural network to be optimized, calculates the fitness value corresponding to every wolf in the wolf pack;
Three wolves for selecting fitness value best are labeled as α, β and δ successively, record α, β and δ respective positions information and fitness value;
Step 3, it is based on α, β and δ respective positions information and fitness value, is optimized using the calculating of grey wolf optimization algorithm
BP neural network model afterwards;
Step 4, using PM2.5 concentration predictions data as input information, the BP neural network model after the optimization is utilized
It calculates and obtains PM2.5 concentration prediction results.
The application proposes a kind of PM2.5 concentration predictions method, by the way that grey wolf optimization algorithm and BP neural network are combined, profit
With the weights and threshold value of grey wolf optimization algorithm Optimized BP Neural Network, using the concentration of the model prediction PM2.5 after optimization.This hair
It is bright that there is the advantageous effect for improving PM2.5 concentration prediction accuracys rate.
Description of the drawings
Fig. 1 is prior art BP neural network structural schematic diagram;
Fig. 2 is the PM2.5 concentration prediction overall plan schematic diagrames that the present invention utilizes grey wolf algorithm optimization BP neural network;
Fig. 3 is the position that the position vector of wolf and next step may move on under 2 dimension spaces in prior art grey wolf optimization algorithm
Set schematic diagram;
Fig. 4 is the location updating schematic diagram of wolf pack in prior art grey wolf optimization algorithm;
Fig. 5 is a kind of PM2.5 concentration predictions method overall flow schematic diagram in the specific embodiment of the invention;
Fig. 6 is a kind of PM2.5 concentration predictions method flow schematic diagram in the specific embodiment of the invention;
Fig. 7 is a kind of verification signal of PM2.5 concentration predictions method prediction result accuracy rate in the specific embodiment of the invention
Figure.
Specific implementation mode
With reference to the accompanying drawings and examples, the specific implementation mode of the present invention is described in further detail.Implement below
Example is not limited to the scope of the present invention for illustrating the present invention.
PM2.5 concentration is predicted currently, the prior art mostly uses neural network.BP neural network is current using most
One of extensive neural network model.When PM2.5 is predicted, meteorological data (such as weather, temperature, wind speed, wind direction) and air
Middle pollutant concentration (such as NOx、SO2、O3Deng) input as neural network, PM2.5 concentration exports as network.
BP neural network is corresponded to first below carries out preliminary introduction.
BP (Back Propagation) neural network is a kind of Multi-layered Feedforward Networks trained by Back Propagation Algorithm,
Typical BP neural network structure has j input, i hidden node and m a defeated as shown in Figure 1, generally 3-tier architecture j-i-m
Go out.Input node is x=[x1,x2,…,xj]T, the network weight between input node and hidden node is W1, and hidden node is o=[o1,
o2,…,oi]T, the weights between hidden node and output node are W2, and output node is y=[y1,y2,…,ym]T, the expectation of network
Output is yp。φi() be i-th of hidden node activation primitive, activation primitive of the generally use S types logarithmic function as network,
The function of tanh can also be used.
The operation principle of BP neural network is made of two parts:The backpropagation of forward-propagating and error.Input layer is each
Neuron is responsible for receiving from extraneous input information, and passes to each neuron of middle layer;Middle layer is internal information processing
Layer is responsible for information transformation;Hidden layer is transmitted to the information of each neuron of output layer, after further treatment after, complete primary study
Forward-propagating processing procedure, by output layer outwardly output information handling result.When reality output and desired output are not inconsistent,
Into the back-propagation phase of error.Error corrects each layer weights by output layer in such a way that error gradient declines, to hidden
Layer, input layer successively anti-pass.Information forward-propagating in cycles and error back propagation process are that each layer weights constantly adjust
Process and neural network learning training process, this process be performed until network output error be reduced to and can connect
Until the degree received or preset study number.
However, when using Neural Network model predictive PM2.5 concentration, the weights and threshold value of neural network are not easy to train, net
Network is easily absorbed in local optimum, causes neural network model low to PM2.5 concentration prediction accuracys rate.
The present invention proposes a kind of air pollutants PM2.5 concentration predictions based on grey wolf optimization algorithm Optimized BP Neural Network
Method.By the weights and threshold value of grey wolf optimization algorithm optimization neural network, a good training starting point is selected to network, is avoided
Network is absorbed in local optimum.PM2.5 concentration prediction accuracys rate can be improved in neural network after this method optimizes.It is calculated based on grey wolf
The air pollutants PM2.5 concentration prediction overall plans of method Optimized BP Neural Network are as shown in Figure 2.
It is briefly described below for grey wolf optimization algorithm.
Grey wolf optimization algorithm (GWO) is to propose that a kind of new member is inspired in 2014 by Seyedali Mirjalili et al.
Formula biological intelligence algorithm.Wolf pack algorithm mainly according to leader's social estate system of lobo in nature and hunts what mechanism proposed.
The wolf of four types is divided into simulate leader's social estate system inside wolf pack by entire wolf pack:Alpha, beta, delta and
omega.Hunting process is divided into three phases according to the hunting mechanism of wolf pack:Prey is searched for, prey, attack prey are surrounded.
In four groups of wolves, α, β, δ are counted as first three wolf behaved oneself best in wolf pack, they guide other wolves (W) to be intended to search for
Best region in space.With tri- kinds of wolves of α, β, δ come forecast assessment prey possible position in entire iterative search procedures, excellent
During change, wolf pack updates their position according to following formula:
Wherein, t is this iterations,WithIt is coefficient vector,It is the position vector of prey,For the position of wolf
It sets.
VectorWithExpression formula it is as follows:
Wherein, coefficientWith the increase of algorithm iteration number from 2 to 0 linear decrease.WithIt is random in [0,1]
Vector.
Location update formula (1) and (2) thought and concept illustrate as shown in Figure 3.It can be seen that being located at (X, Y) from diagram
Wolf at position can relocate the position of oneself according to location update formula presented above around prey.Although in figure
7 positions that wolf can move into are shown only, wolf can be allowed to be moved to around prey by adjusting random parameter A and C and connected
Any one position in continuous space.
In wolf pack algorithm, we always assume that the location of α, β and δ wolf is likely to the position of prey (optimal solution)
It sets.In searching process, first three the best solution obtained at present is assumed to be α, β and δ respectively, and then others are considered as W
Wolf the position of oneself is repositioned according to the position of tri- head wolves of α, β and δ.It is adjusted again using following mathematics mould preparation formula
The location updating schematic diagram of the position of whole W classes wolf, wolf is as shown in Figure 4.
WhereinIt is the position of alpha wolves,It is the position of beta wolves,It is the position of delta wolves.Be at random to
Amount indicates the position currently solved.Formula (5), (6) and (7) calculates separately between current solution position and alpha, beta and delta
Approximate distance.After having defined the distance between they, the rearmost position currently solved is calculated by following formula:
In formula,Indicate the position of alpha wolves,Indicate the position of beta wolves,Indicate the position of delta wolves.It is random vector, t indicates iterations.
From above-mentioned formula as can be seen that formula (5), (6), the step-length that (7) respectively define when W is intended to α, β, δ wolf is big
It is small.Formula (8), (9), (10) and (11) define the final position of W wolves.
It such as Fig. 5, shows in a specific embodiment of the invention, a kind of overall flow signal of PM2.5 concentration predictions method
Figure.Generally, include the following steps:
Step 1, wolf pack corresponding with BP neural network to be optimized is generated at random using grey wolf optimization algorithm;
Step 2, the training BP neural network to be optimized, calculates the fitness value corresponding to every wolf in the wolf pack;
Three wolves for selecting fitness value best are labeled as α, β and δ successively, record α, β and δ respective positions information and fitness value;
Step 3, it is based on α, β and δ respective positions information and fitness value, is optimized using the calculating of grey wolf optimization algorithm
BP neural network model afterwards;
Step 4, using PM2.5 concentration predictions data as input information, the BP neural network model after the optimization is utilized
It calculates and obtains PM2.5 concentration prediction results.
In another of the invention specific embodiment, a kind of PM2.5 concentration predictions method further includes before the step 1:If
The topological structure of fixed BP neural network to be optimized, sets the size of BP neural network initial weight w and threshold value b to be optimized;Setting
Wolf pack scale and grey wolf optimization algorithm maximum iteration and/or iteration precision.
In another of the invention specific embodiment, a kind of PM2.5 concentration predictions method utilizes grey wolf in the step 1
Optimization algorithm generates wolf pack corresponding with BP neural network to be optimized at random:Utilize opening up for BP neural network to be optimized
Flutter the dimension of structure determination wolf position.
In another of the invention specific embodiment, a kind of PM2.5 concentration predictions method calculates every in the step 2
Fitness value corresponding to wolf further includes:Using the PM2.5 values of the previous day, meteorological data and pollutant concentration data as nerve net
The input of network is exported PM2.5 values one day after as neural network, using PM2.5 in neural network output error as grey wolf
The fitness function of optimization algorithm calculates the fitness value corresponding to every wolf using the comfort level function.
In another of the invention specific embodiment, a kind of PM2.5 concentration predictions method, the step 3 further includes:
S31 updates the location information of other wolf W except α, β and δ using grey wolf optimization algorithm;Grey wolf optimization is updated to calculate
Method parameter;
S32, repetitive cycling step 2 and step 3, until confirm meet stopping criterion for iteration, using α wolves location information more
The weights and threshold value of new BP neural network;
S33 is trained the BP data networks using PM2.5 concentration training data, the BP nerves after being optimized
Network model.
In aforementioned present invention specific embodiment, the network inputs include:First day highest temperature, the lowest temperature,
Daytime weather, night weather, wind-force, wind direction, SO2, CO, NO2, O3, PM10, PM2.5.The network exports:Second day
PM2.5 values;Above-mentioned input and output are when second step into entrance neural network.
In another of the invention specific embodiment, a kind of PM2.5 concentration predictions method, iteration ends in the step 4
Condition includes:Meet grey wolf optimization algorithm maximum iteration or stops GWO algorithms to nerve after reaching the iteration precision of setting
The optimization of network.
In aforementioned present invention specific embodiment, after stopping GWO algorithms to the optimization of neural network, whole process is not
Terminate, is followed by the training optimization for carrying out neural network itself.
In another of the invention specific embodiment, a kind of PM2.5 concentration predictions method is excellent using waiting in the step 1
The topological structure for changing BP neural network determines that the dimension of wolf position further includes:
Wherein, lthIt is th layers of neuron number of BP neural network, Th is the total number of plies of BP neural network, and the BP nerves are total
The number of plies includes the input layer and output layer of BP nerves.
In another of the invention specific embodiment, a kind of PM2.5 concentration predictions method, grey wolf optimizes in the step 2
The fitness function of algorithm further includes:
Wherein, youtiFor i-th of sample PM2.5 concentration output valve of BP neural network, aiIt is real for i-th of sample PM2.5 concentration
Actual value, n are test sample quantity.
In another of the invention specific embodiment, a kind of PM2.5 concentration predictions method, by W wolf roots in the step S31
Repositioning the position of oneself according to the position of tri- head wolves of α, β and δ further includes:
Wherein,For the position of α wolves,For the position of β wolves,It is random vector for the position of δ wolves, indicates current
The position of solution;
Wherein,For the position of α wolves,For the position of β wolves,For the position of δ wolves, WithFor at random to
Amount, t is iterations.
In another of the invention specific embodiment, a kind of PM2.5 concentration predictions method utilizes α wolves in the step 4
The weights and threshold value of updating location information BP neural network further include:
Wi=X (K+1:K+li×li+1),
Bi=X (K+li×li+1+1:K+li×li+1+li+1),
Wherein:
WiThe weights for being i-th layer of BP neural network between i+1 layer, BiFor i+1 layer threshold value, X is the position of α wolves, li
For the number of i-th layer of neuron of BP neural network.
Such as Fig. 6, show in another specific embodiment of the invention, a kind of PM2.5 concentration predictions method operational flow diagram, institute
Method is stated by by grey wolf optimization algorithm and BP neural network combination, utilizing the weights of grey wolf optimization algorithm Optimized BP Neural Network
And threshold value, using the concentration of the model prediction PM2.5 after optimization.The weights and threshold value of network in BP neural network learning process
Directly influence the accuracy rate of PM2.5 predictions.So we use the weights and threshold value of GWO algorithm optimization networks.The embodiment
Method generally comprises following steps:
Step A, parameter initialization.Determine the topological structure of BP neural network, the size of initial network weight w and threshold value b,
Obtain the training sample of BP neural network.Set wolf pack scale n, maximum iteration itermax.
Step B:It is random to generate population.Random N wolf of generationThe position pair of every wolf
Answer the weights and threshold value of one group of BP neural network.
The specific method for being combined grey wolf position with neural network weight and threshold value in the present invention:According to network topology structure
The dimension for determining wolf position, with the weights and threshold value of the location updating neural network of wolf pack.
Wolf location dimension size determines method:
Wherein, lthIt is th layers of neuron number of network, the Th networks number of plies (including input layer and output layer)
Step C:Training BP neural network, calculates the fitness value corresponding to every wolf, selects three that fitness value is best
Wolf is labeled as α, β and δ successively, retains their position and fitness value fitness.
Wherein, it is by PM2.5 in neural network by the specific link method of PM2.5 concentration and neural network and GWO algorithms
Fitness function of the output error as grey wolf optimization algorithm:
youtiIt is i-th of sample PM2.5 concentration output valve of network pair, aiIt is i-th of sample PM2.5 concentration actual value, n is
Test sample quantity.
Step D:Location updating.The position of other wolf W is updated using formula in this specification (5)-(11), that is, updates W classes
The position of wolf;
Step E:Parameter updates.Utilize formula in this specification (3)-(4) update grey wolf optimization algorithm parameter a, A and C;
Step F:Judge operation.Judge whether to meet stopping criterion for iteration, if it is satisfied, the position of α wolves and fitness value
It is exported as optimal solution, if conditions are not met, return to step C.
Step G:Model is verified.By the position of α wolvesWeights and threshold value of the corresponding parameter as BP neural network,
To training data re -training, BP neural network model is established, test data is used in combination to verify model.
Further include using the weights and threshold value of the updating location information BP neural network of α wolves in step G:
Wi=X (K+1:K+li×li+1),
Bi=X (K+li×li+1+1:K+li×li+1+li+1),
Wherein:
WiThe weights for being i-th layer of BP neural network between i+1 layer, BiFor i+1 layer threshold value, X is the position of α wolves, li
For the number of i-th layer of neuron of BP neural network.
In another specific embodiment of the invention, a kind of PM2.5 concentration predictions method, this example uses Nanjing 2015
The training and prediction data of 500 groups of air pollution concentrations in January in May, 2016 in year and meteorological object data as network, wherein
400 groups are training data, and 100 groups are used as prediction data.Initial data is as shown in table 1 below.
1 Nanjing meteorology of table and pollutant concentration initial data
Rejecting abnormal data is carried out to initial data first, weather, wind-force, wind direction feature are carried out classification and data format
Conversion, is converted to the information that neural network can identify.Weather characteristics are divided into 14 classes, wind direction feature is divided into 8 classes, wind-force is divided into
5 classes.Classification is detailed as shown in the table.
Weather data processing, is divided into 14 classes by weather
It is fine | 1 | Mist | 4 | Shower | 7 | Moderate rain or heavy rain | 10 | Heavy rain | 12 | Light to moderate snow | 14 |
It is cloudy | 2 | Light rain | 5 | Thunder shower | 8 | Heavy rain | 11 | Rain and snow mixed | 13 | Moderate snow | 14 |
It is cloudy | 3 | Drizzle or moderate rain | 6 | Moderate rain | 9 | Heavy or torrential rain | 12 | Slight snow | 14 | Heavy snow | 14 |
Wind data 5 ranks of processing point
Gentle breeze | 2 |
3-4 grades | 3 |
4-5 grades | 4 |
5-6 grades | 5 |
6-7 grades | 6 |
Wind direction number data processing, point 8 wind directions
East wind | 1 | Northwest | 5 |
Southeaster | 2 | West | 6 |
Northeast | 3 | Southwest | 7 |
North | 4 | South | 8 |
Data are as shown in table 2 after processing:
Data after 2 Nanjing meteorology of table and pollutant concentration processing
The PM2.5 values, meteorological data and pollutant concentration data of the previous day and PM2.5 values one day after are corresponded.
With the PM2.5 values of the previous day, the value of the PM2.5 of meteorology and pollutant concentration data prediction one day after.
Network inputs:First day highest temperature, the lowest temperature, weather on daytime, night weather, wind-force, wind direction, SO2,
CO,NO2,O3,PM10,PM2.5.Network exports:Second day PM2.5 value.
The sample of generation is upset at random, enters training net network after finally carrying out data normalization processing.
This experiment includes input layer, single hidden layer, output layer using 3 layers of BP neural network.Input layer dimension is 12, defeated
Go out the dimension of layer 1, hidden layer neuron number is 5.Grey wolf algorithm scale n=30, grey wolf algorithm iteration number iter=100, nerve
Network training iterations epochs=1000, e-learning rate lr=0.1.
Test verification is carried out to trained network, Fig. 7 is the test result comparison diagram of 100 groups of data.
Finally, the present processes are only preferable embodiment, are not intended to limit the scope of the present invention.It is all
Within the spirit and principles in the present invention, any modification, equivalent replacement, improvement and so on should be included in the protection of the present invention
Within the scope of.
Claims (10)
1. a kind of PM2.5 concentration predictions method, which is characterized in that include the following steps:
Step 1, wolf pack corresponding with BP neural network to be optimized is generated at random using grey wolf optimization algorithm;
Step 2, the training BP neural network to be optimized, calculates the fitness value corresponding to every wolf in the wolf pack;Selection
Three best wolves of fitness value are labeled as α, β and δ successively, record α, β and δ respective positions information and fitness value;
Step 3, it is based on α, β and δ respective positions information and fitness value, after being optimized using the calculating of grey wolf optimization algorithm
BP neural network model;
Step 4, it using PM2.5 concentration predictions data as input information, is calculated using the BP neural network model after the optimization
Obtain PM2.5 concentration prediction results.
2. the method as described in claim 1, which is characterized in that further include before the step 1:Set BP neural network to be optimized
Topological structure, set the size of BP neural network initial weight w and threshold value b to be optimized;Set wolf pack scale and grey wolf optimization
Algorithm maximum iteration and/or iteration precision.
3. the method as described in claim 1, which is characterized in that in the step 1 using grey wolf optimization algorithm at random generate with
The corresponding wolf pack of BP neural network to be optimized further includes:Wolf position is determined using the topological structure of BP neural network to be optimized
Dimension.
4. the method as described in claim 1, which is characterized in that calculate the fitness value corresponding to every wolf in the step 2
Further include:It, will one day after using the PM2.5 values of the previous day, meteorological data and pollutant concentration data as the input of neural network
PM2.5 values as neural network export, using PM2.5 in neural network output error as the fitness letter of grey wolf optimization algorithm
Number, the fitness value corresponding to every wolf is calculated using the fitness function.
5. the method as described in claim 1, which is characterized in that the step 3 further includes:
S31 updates the location information of other wolf W except α, β and δ using grey wolf optimization algorithm;Update grey wolf optimization algorithm ginseng
Number;
S32, repetitive cycling step 2 and step 3 meet stopping criterion for iteration until confirming, utilize the updating location information BP of α wolves
The weights and threshold value of neural network;
S33 is trained the BP data networks using PM2.5 concentration training data, the BP neural network after being optimized
Model.
6. the method as described in claim 1, which is characterized in that stopping criterion for iteration includes in the step 4:It is excellent to meet grey wolf
Change algorithm maximum iteration or stops optimization of the GWO algorithms to neural network after reaching the iteration precision of setting.
7. the method as described in claim 1, which is characterized in that utilize the topology of BP neural network to be optimized in the step 1
The dimension of structure determination wolf position further includes:
Wherein, lthIt is th layers of neuron number of BP neural network, Th is the total number of plies of BP neural network, the total number of plies of the BP nerves
Include the input layer and output layer of BP nerves.
8. method as claimed in claim 4, which is characterized in that the fitness function of grey wolf optimization algorithm is also in the step 2
Including:
Wherein, youtiFor i-th of sample PM2.5 concentration output valve of BP neural network, aiIt is practical for i-th of sample PM2.5 concentration
Value, n are test sample quantity.
9. method as claimed in claim 5, which is characterized in that in the step S31 using grey wolf optimization algorithm update α, β and
The position of remaining outer wolf W of δ further includes:
Wherein,For the position of α wolves,For the position of β wolves,It is random vector for the position of δ wolves, what expression currently solved
Position;
Wherein,For the position of α wolves,For the position of β wolves,For the position of δ wolves, WithFor random vector, t is
Iterations.
10. method as claimed in claim 6, which is characterized in that utilize the updating location information BP god of α wolves in the step 4
Weights and threshold value through network further include:
Wi=X (K+1:K+li×li+1),
Bi=X (K+li×li+1+1:K+li×li+1+li+1),
Wherein:
WiThe weights for being i-th layer of BP neural network between i+1 layer, BiFor i+1 layer threshold value, X is the position of α wolves, liFor BP
The number of i-th layer of neuron of neural network.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710297281.5A CN108805253B (en) | 2017-04-28 | 2017-04-28 | PM2.5 concentration prediction method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710297281.5A CN108805253B (en) | 2017-04-28 | 2017-04-28 | PM2.5 concentration prediction method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108805253A true CN108805253A (en) | 2018-11-13 |
CN108805253B CN108805253B (en) | 2021-03-02 |
Family
ID=64053868
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710297281.5A Expired - Fee Related CN108805253B (en) | 2017-04-28 | 2017-04-28 | PM2.5 concentration prediction method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108805253B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109696412A (en) * | 2018-12-20 | 2019-04-30 | 南京信息工程大学 | Infrared gas sensor and atmospheric pressure compensating method based on AGNES Optimized BP Neural Network |
CN111371607A (en) * | 2020-02-28 | 2020-07-03 | 大连大学 | Network traffic prediction method for optimizing LSTM based on decision-making graying algorithm |
CN113777000A (en) * | 2021-10-09 | 2021-12-10 | 山东科技大学 | Dust concentration detection method based on neural network |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104092503A (en) * | 2014-07-15 | 2014-10-08 | 哈尔滨工程大学 | Artificial neural network spectrum sensing method based on wolf pack optimization |
CN106022517A (en) * | 2016-05-17 | 2016-10-12 | 温州大学 | Risk prediction method and device based on nucleus limit learning machine |
CN106355192A (en) * | 2016-08-16 | 2017-01-25 | 温州大学 | Support vector machine method based on chaos and grey wolf optimization |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5927988A (en) * | 1997-12-17 | 1999-07-27 | Jenkins; William M. | Method and apparatus for training of sensory and perceptual systems in LLI subjects |
CN106374534B (en) * | 2016-11-17 | 2018-11-02 | 云南电网有限责任公司玉溪供电局 | A kind of extensive home energy management method based on multiple target grey wolf optimization algorithm |
-
2017
- 2017-04-28 CN CN201710297281.5A patent/CN108805253B/en not_active Expired - Fee Related
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104092503A (en) * | 2014-07-15 | 2014-10-08 | 哈尔滨工程大学 | Artificial neural network spectrum sensing method based on wolf pack optimization |
CN106022517A (en) * | 2016-05-17 | 2016-10-12 | 温州大学 | Risk prediction method and device based on nucleus limit learning machine |
CN106355192A (en) * | 2016-08-16 | 2017-01-25 | 温州大学 | Support vector machine method based on chaos and grey wolf optimization |
Non-Patent Citations (3)
Title |
---|
M.R. MOSAVI ET AL: "CLASSIFICATION OF SONAR DATA SET USING NEURAL NETWORK TRAINED BY GRAY WOLF OPTIMIZATION", 《NEURAL NETWORK WORLD》 * |
MOHAMED F.HASSANIN ET AL: "Grey Wolf Optimizer-based Back-propagation Neural Network Algorithm", 《2016 12TH INTERNATIONAL COMPUTER ENGINEERING CONFERENCE (ICENCO)》 * |
尹安琪等: "基于 PSO-BP 神经网络预测广州市日均 PM10浓度", 《中国卫生统计》 * |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109696412A (en) * | 2018-12-20 | 2019-04-30 | 南京信息工程大学 | Infrared gas sensor and atmospheric pressure compensating method based on AGNES Optimized BP Neural Network |
CN111371607A (en) * | 2020-02-28 | 2020-07-03 | 大连大学 | Network traffic prediction method for optimizing LSTM based on decision-making graying algorithm |
CN111371607B (en) * | 2020-02-28 | 2022-09-16 | 大连大学 | Network flow prediction method for optimizing LSTM based on decision-making graying algorithm |
CN113777000A (en) * | 2021-10-09 | 2021-12-10 | 山东科技大学 | Dust concentration detection method based on neural network |
CN113777000B (en) * | 2021-10-09 | 2024-04-12 | 山东科技大学 | Dust concentration detection method based on neural network |
Also Published As
Publication number | Publication date |
---|---|
CN108805253B (en) | 2021-03-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108426812B (en) | PM2.5 concentration value prediction method based on memory neural network | |
CN110782093B (en) | PM fusing SSAE deep feature learning and LSTM2.5Hourly concentration prediction method and system | |
CN109886444A (en) | A kind of traffic passenger flow forecasting, device, equipment and storage medium in short-term | |
US11874429B2 (en) | High-temperature disaster forecast method based on directed graph neural network | |
CN111783987A (en) | Farmland reference crop evapotranspiration prediction method based on improved BP neural network | |
CN108805253A (en) | A kind of PM2.5 concentration predictions method | |
CN114170789B (en) | Intelligent network link lane change decision modeling method based on space-time diagram neural network | |
CN109164707A (en) | A kind of indoor environment negative-feedback regu- lation system based on artificial neural network algorithm | |
CN111222706A (en) | Chaos time sequence prediction method based on particle swarm optimization and self-encoder | |
CN108062595A (en) | Complex landform region wind energy Forecasting Methodology in short-term based on WRF/CFD/SAHDE-RVM couplings | |
CN110222411B (en) | Non-measurement-area solar radiation estimation method based on mRMR-DBN algorithm | |
CN113642268B (en) | River bloom prediction method and system | |
CN109143408A (en) | Combine short-term precipitation forecasting procedure in dynamic area based on MLP | |
CN108399470A (en) | A kind of indoor PM2.5 prediction techniques based on more example genetic neural networks | |
CN106991493A (en) | Sewage disposal water outlet parameter prediction method based on Grey production fuction | |
CN111242380A (en) | Lake (reservoir) eutrophication prediction method based on artificial intelligence algorithm | |
CN111292124A (en) | Water demand prediction method based on optimized combined neural network | |
Zhao et al. | Prediction of daily reference crop evapotranspiration in different Chinese climate zones: Combined application of key meteorological factors and Elman algorithm | |
CN106896724A (en) | For the tracking system and tracking of sun tracker | |
CN111815051B (en) | GRNN photovoltaic power generation prediction method considering weather influence factors | |
CN113379148A (en) | Pollutant concentration inversion method based on fusion of multiple machine learning algorithms | |
Gunawan et al. | Long Short-Term Memory Approach for Predicting Air Temperature In Indonesia | |
Raza et al. | Comparative study of powerful predictive modeling techniques for modeling monthly reference evapotranspiration in various climatic regions | |
Long et al. | A short-term climate prediction model based on a modular fuzzy neural network | |
CN113971477A (en) | PM based on time series and deep learning framework2.5Estimation method |
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 | ||
CF01 | Termination of patent right due to non-payment of annual fee | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20210302 |