CN108665322A - The construction method of grain ration Consumption forecast model, Consumption forecast method and device - Google Patents
The construction method of grain ration Consumption forecast model, Consumption forecast method and device Download PDFInfo
- Publication number
- CN108665322A CN108665322A CN201810481553.1A CN201810481553A CN108665322A CN 108665322 A CN108665322 A CN 108665322A CN 201810481553 A CN201810481553 A CN 201810481553A CN 108665322 A CN108665322 A CN 108665322A
- Authority
- CN
- China
- Prior art keywords
- particle
- grain ration
- ration consumption
- grain
- networks
- 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
- 238000010276 construction Methods 0.000 title claims abstract description 24
- 238000000034 method Methods 0.000 title claims abstract description 23
- 239000002245 particle Substances 0.000 claims abstract description 109
- 238000005457 optimization Methods 0.000 claims abstract description 28
- 230000007423 decrease Effects 0.000 claims abstract description 23
- 230000006870 function Effects 0.000 claims description 20
- 238000004364 calculation method Methods 0.000 claims description 13
- 230000000670 limiting effect Effects 0.000 claims description 4
- 241000208340 Araliaceae Species 0.000 claims description 3
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 claims description 3
- 235000003140 Panax quinquefolius Nutrition 0.000 claims description 3
- 235000008434 ginseng Nutrition 0.000 claims description 3
- 230000006872 improvement Effects 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 description 12
- 230000008859 change Effects 0.000 description 8
- 238000012549 training Methods 0.000 description 7
- 230000007547 defect Effects 0.000 description 5
- 230000003247 decreasing effect Effects 0.000 description 4
- 238000013507 mapping Methods 0.000 description 3
- 210000002569 neuron Anatomy 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 101001095088 Homo sapiens Melanoma antigen preferentially expressed in tumors Proteins 0.000 description 2
- 102100037020 Melanoma antigen preferentially expressed in tumors Human genes 0.000 description 2
- 101100458287 Saguinus oedipus MSPE gene Proteins 0.000 description 2
- 238000013459 approach Methods 0.000 description 2
- 230000006399 behavior Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 235000013399 edible fruits Nutrition 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000013213 extrapolation Methods 0.000 description 2
- 238000012886 linear function Methods 0.000 description 2
- 230000036961 partial effect Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 208000030507 AIDS Diseases 0.000 description 1
- 241001269238 Data Species 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 238000013499 data model Methods 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 235000006694 eating habits Nutrition 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 235000013305 food Nutrition 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 238000012804 iterative process Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 230000002829 reductive effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000000452 restraining effect Effects 0.000 description 1
- 238000012546 transfer Methods 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
- 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
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0201—Market modelling; Market analysis; Collecting market data
- G06Q30/0202—Market predictions or forecasting for commercial activities
-
- 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
- G06Q50/02—Agriculture; Fishing; Forestry; Mining
Landscapes
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- Development Economics (AREA)
- Entrepreneurship & Innovation (AREA)
- Accounting & Taxation (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Finance (AREA)
- Tourism & Hospitality (AREA)
- Game Theory and Decision Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mining & Mineral Resources (AREA)
- Data Mining & Analysis (AREA)
- Marine Sciences & Fisheries (AREA)
- Animal Husbandry (AREA)
- Agronomy & Crop Science (AREA)
- Primary Health Care (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention relates to a kind of construction method of grain ration Consumption forecast model, Consumption forecast method and device, which includes:It obtains the grain ration consumption figure in each time and influences the impact factor of grain ration consumption figure, and pre-processed;The topological structure of BP networks is built, weights and threshold value in BP networks is initialized, is encoded to particle, the parameter in particle cluster algorithm is set, and parameter includes the inertia weight of decreases in non-linear;The fitness value for calculating each particle determines the individual optimal and global optimum of particle position;Constantly update particle speed and position until meet end condition, to the BP networks initial weight and threshold value after being optimized;According to the initial weight and threshold value after optimization, BP networks are trained to obtain grain ration Consumption forecast model.The present invention has been well adapted for the optimization problem with complex nonlinear variation characteristic, has effectively increased the accuracy of grain ration Consumption forecast by the inertia weight of setting decreases in non-linear.
Description
Technical field
The present invention relates to a kind of construction method of grain ration Consumption forecast model, Consumption forecast method and devices, belong to
Grain ration Consumption forecast technical field.
Background technology
Food problem is the grand strategy goods and materials to involve the interests of the state and the people, is the basis of social stability and economic development.China
It is not only populous nation, large agricultural country, even more a grain big country.And existing grain structure has been unable to meet people
Consumption demand.Therefore, it is vital to the research of China's grain change conditions.
In grain structure, traditional model about grain consumption prediction mainly has:ELS models, ELES moulds
Type, trend extrapolation, grey forecasting model, AIDS models, time series models, Regression Model, partial equilibrium mould
Type, Panel Data models etc..However these methods mostly all there are one or multiple defects, when only considering such as time series models
Between factor, if extraneous factor changes larger, predicted value and actual value have very large deviation.Grey forecasting model is to historical data
There is very strong dependence, does not account for the contact between each factor.Regression Model using which kind of factor and
It is a kind of supposition when the expression way of the factor, its availability is subject to certain restrictions.ELS models and ELES models are one
Kind linear model, and in actual life, most of problem all has complexity, non-linear behavior.Partial equilibrium model is isolated
Ground considers the relationship between the supply and demand and price of market segment, without considering the contact between them.Trend extrapolation is basis
Certain development trend is not high to predict the change conditions of the coming years, the method accuracy.
Therefore, a kind of method improving grain consumption precision of prediction of exploration is very necessary.And BP neural network has
There is the features such as very strong non-linear mapping capability, self-learning capability, is widely used in forecasting problem.BP neural network generally has
Input layer, hidden layer, output layer composition, Fig. 1 is typical 3 layers of BP neural network structure chart.In Fig. 1, x is neuron
Input, z is the output of hidden layer, and y is the output of output layer, and each layer has its weights and threshold value.Input layer is responsible for connecing
Extraneous input information is received, and passes to middle layer, middle layer is responsible for internal information processing and transformation, finally outside by output layer
Boundary's output information handling result.When reality output and desired output are not inconsistent, into the back-propagation phase of error.BP algorithm
Learning process is made of forward-propagating and backpropagation, makes error by the connection weight and threshold value of repeatedly changing each layer neuron
It reduces, iterates, until error is less than preset value.
However traditional BP neural network is not a very perfect network, it is primarily present study convergence rate
Slowly, it cannot be guaranteed that the defects of error convergence is to global minima and unstable network structure.In addition when network training,
The selection of initial weight and threshold value is typically random, often passes test of many times but can not find optimal structure and power
Value.
The defects of being absorbed in local optimum slowly and easily there are convergence rate due to traditional BP networks, needs on this basis
It is optimized, and then improves precision of prediction and accelerates convergence rate.The Chinese patent document that publication No. is CN107316099A is public
A kind of Ammunition Storage Reliability prediction technique based on particle group optimizing BP neural network is opened, this method is using population come excellent
Change BP neural network.Wherein, particle swarm optimization algorithm is a kind of intelligent algorithm based on population behavior.Its basic thought be from
RANDOM SOLUTION is set out, and each particle is to track best particle to find best solution in solution space.Its feature is principle letter
It is single, need the parameter changed few, easy to use, convergence rate is very fast, but there are problems that easily being absorbed in local optimum.Using particle
Group's Optimized BP Neural Network, compensates for that original BP neural network convergence rate is slow, the defects of being easily absorbed in local minimum, reaches bullet
The precision of medicine storage reliability prediction.But this method uses linear decrease Weight Algorithm, being not suitable with, there is complex nonlinear to become
Change the optimization problem of feature, and then causes prediction result inaccurate.
Invention content
The object of the present invention is to provide a kind of construction method of grain ration Consumption forecast model, Consumption forecast method and dresses
It sets, for solving the problems, such as that BP neural network can cause prediction result inaccurate using linear decrease Weight Algorithm.
In order to solve the above technical problems, the present invention provides a kind of construction method of grain ration Consumption forecast model, step
It is as follows:
It obtains the grain ration consumption figure in each time and influences the impact factor of grain ration consumption figure, and pre-processed;
According to pretreated grain ration consumption figure and the impact factor of influence grain ration consumption figure, the topology of BP networks is built
Structure;
The weights and threshold value in BP networks are initialized, are encoded to particle, and the parameter in particle cluster algorithm is set, institute
State the inertia weight that parameter includes decreases in non-linear;
It determines the fitness function in particle cluster algorithm, and calculates the fitness value of each particle, determine particle position
Individual optimal and global optimum;
According to the individual optimal and global optimum of particle position, speed and the position of particle are constantly updated, until meeting eventually
Only condition, to the initial weight and threshold value in the BP networks after being optimized;
According to the initial weight and threshold value in the BP networks after optimization, BP networks are trained, to obtain being based on changing
Into the grain ration Consumption forecast model of particle swarm optimization BP network.
The beneficial effects of the invention are as follows:By the way that the inertia weight of decreases in non-linear is arranged, can be well adapted for having multiple
The optimization problem of miscellaneous nonlinear change feature, in algorithm early period, Inertia Weight setting it is bigger, it is possible to prevente effectively from being absorbed in office
Portion is optimal, and in the algorithm later stage, Inertia Weight is arranged smaller, precision and the accuracy of optimal solution is effectively increased, to have
Effect improves the accuracy of grain ration Consumption forecast.
Further, in order to the inertia weight of decreases in non-linear, the calculation formula of the inertia weight of decreases in non-linear is arranged
For:
Wherein, wmax、wminIndicate that the upper and lower bound of inertia weight w, iter are current iteration number, Max respectivelyiterFor
Maximum iteration, a are constant coefficient, control the decline curve shape of nonlinear function, take a=10.
Further, wmax=0.9, wmin=0.4.
Further, in order to allow algorithm to carry out extensive search in early period, and search can be reduced in the algorithm later stage
Range is defined speed edges, and the parameter in particle cluster algorithm further includes particle rapidity, wherein the meter on particle rapidity boundary
Calculating formula is:
Vmax=v1-(v1-v2)(iter/Maxiter)
Wherein, VmaxFor particle rapidity boundary, v1For the upper limit value of speed, v2For the lower limiting value of speed, iter is current changes
Generation number, MaxiterFor maximum iteration
Further, v1=1.5, v2=0.5.
Further, the calculation formula of the fitness function in particle cluster algorithm is:
Wherein, MSE indicates the fitness function in particle cluster algorithm, fiIndicate the network output node of i-th of particle
Predicted value, yiIndicate that the actual value of the network output node of i-th of particle, N indicate number of particles.
The present invention also provides a kind of construction device of grain ration Consumption forecast model, including processor and memory, institutes
Processor is stated for handling the instruction being stored in the memory to realize following method:
It obtains the grain ration consumption figure in each time and influences the impact factor of grain ration consumption figure, and pre-processed;
According to pretreated grain ration consumption figure and the impact factor of influence grain ration consumption figure, the topology of BP networks is built
Structure;
The weights and threshold value in BP networks are initialized, are encoded to particle, and the parameter in particle cluster algorithm is set, institute
State the inertia weight that parameter includes decreases in non-linear;
It determines the fitness function in particle cluster algorithm, and calculates the fitness value of each particle, determine particle position
Individual optimal and global optimum;
According to the individual optimal and global optimum of particle position, speed and the position of particle are constantly updated, until meeting eventually
Only condition, to the initial weight and threshold value in the BP networks after being optimized;
According to the initial weight and threshold value in the BP networks after optimization, BP networks are trained, to obtain being based on changing
Into the grain ration Consumption forecast model of particle swarm optimization BP network.
Further, the calculation formula of the inertia weight of decreases in non-linear is:
Wherein, wmax、wminIndicate that the upper and lower bound of inertia weight w, iter are current iteration number, Max respectivelyiterFor
Maximum iteration, a are constant coefficient, control the decline curve shape of nonlinear function, take a=10.
Further, wmax=0.9, wmin=0.4.
Further, the parameter in particle cluster algorithm further includes particle rapidity, wherein the calculation formula on particle rapidity boundary
For:
Vmax=v1-(v1-v2)(iter/Maxiter)
Wherein, VmaxFor particle rapidity boundary, v1For the upper limit value of speed, v2For the lower limiting value of speed, iter is current changes
Generation number, MaxiterFor maximum iteration.
Further, v1=1.5, v2=0.5.
Further, the calculation formula of the fitness function in particle cluster algorithm is:
Wherein, MSE indicates the fitness function in particle cluster algorithm, fiIndicate the network output node of i-th of particle
Predicted value, yiIndicate that the actual value of the network output node of i-th of particle, N indicate number of particles.
The present invention also provides a kind of grain ration consumption figures of the construction method using above-mentioned grain ration Consumption forecast model
Prediction technique, steps are as follows:
Obtain influence grain ration consumption figure impact factor data, and by get influence grain ration consumption figure influence because
The data of son are brought into the grain ration Consumption forecast model based on Modified particle swarm optimization BP networks, to obtain grain ration consumption figure
Predicted value.
The present invention also provides a kind of prediction meanss of grain ration consumption figure, including processor and memory, the processors
For handling the instruction being stored in the memory to realize following method:
Obtain influence grain ration consumption figure impact factor data, and by get influence grain ration consumption figure influence because
The data of son are brought into the grain ration Consumption forecast model based on Modified particle swarm optimization BP networks, to obtain grain ration consumption figure
Predicted value.
Description of the drawings
Fig. 1 is three layers of BP neural network structure chart;
Fig. 2 is the flow chart of the construction method of grain ration Consumption forecast model of the present invention;
Fig. 3 is the comparison diagram of Inertia Weight decreases in non-linear curve and Inertia Weight linear decrease curve;
Fig. 4 is the maximum speed of particle with iterations change curve;
Fig. 5 is the training sample result of model after optimization;
Fig. 6 is that three kinds of model predication values are compared with desired value;
Fig. 7 is three kinds of model absolute error curves;
Fig. 8 is three kinds of model relative error curves.
Specific implementation mode
In order to make the purpose , technical scheme and advantage of the present invention be clearer, below in conjunction with attached drawing and specific implementation
The present invention will be described in further detail for example.
The present invention provides a kind of construction device of grain ration Consumption forecast model, including processor and memory, at this
Reason device is for handling instruction stored in memory, to realize a kind of construction method of grain ration Consumption forecast model.The mouth
The construction method of grain Consumption forecast model is slow for the convergence rate of BP algorithm, and the defects of being easily trapped into local optimum utilizes
The weights and threshold value of particle cluster algorithm Optimized BP Neural Network are improved, best initial weights and threshold value are found, so as in solution space
In orient some preferable search spaces.Then BP networks are trained according to best initial weights threshold value, in these small solutions
Optimal solution is searched out in space, then constantly iteration continues.It is finally predicted using the BP neural network optimized, prediction is made to tie
Fruit has better non-linear mapping capability, higher precision of prediction, faster convergence rate and is not easy to be absorbed in local optimum,
Its flow chart is as shown in Fig. 2, be as follows:
(1) it obtains the grain ration consumption figure in each time and influences the impact factor of grain ration consumption figure, and pre-processed.
Wherein, the factor for influencing grain ration consumption figure mainly has Macroscopic Factors and microcosmic influence factors.Macroscopic Factors include mainly receiving
Enter level, income gap, the level of urbanization, relative price level, area differentiation etc., microcosmic influence factors mainly have population, population old
Age, eating habit etc..By analysis, in the present embodiment, chooses the size of population, disaster area, Engel coefficient, per capita may be used
Seven income, grain relative price, the level of urbanization, between twenty and fifty population accounting principal elements are dominated as influence grain ration consumption figure
Impact factor.
It is on the grain ration consumption figure in each time and the pretreated method of impact factor progress for influencing grain ration consumption figure:
The value of grain ration consumption figure and impact factor is normalized respectively using linear function transformation approach, calculation formula is:
X '=(x-xmin)/(xmax-xmin)
Wherein, x, x ' are the sample value of pre- and post-conversion, x respectivelymax、xminIt is maximum value in sample and most respectively
Small value.
(2) according to pretreated grain ration consumption figure and the impact factor of influence grain ration consumption figure, BP networks are built
Topological structure.
Wherein, when constructing the topological structure of BP networks, the impact factor of grain ration consumption figure will be influenced as input variable,
Using grain ration consumption figure as output variable.The relevant parameter of BP networks includes mainly network node, the selection of activation primitive and power
It is worth the determination of threshold value, other parameters can obtain based on experience value.BP network nodes include input layer, hidden layer and output layer, root
Determine that the input number of plies, the number of output variable determine that the output number of plies finds network by constantly testing according to the number of input variable
Number of nodes when training error minimum is as the implicit number of plies.In the present embodiment, finally determining BP network structures are:7-8-1,
The determination formula of node in hidden layer n is:
Wherein, m is the node number of input layer, and q is output layer node number, and b is constant, the value range of b be [1,
10]。
In addition, in the present embodiment, hidden layer shifts transfer function and Sigmoid functions, output layer transmission function is used to adopt
Use linear function.To improve the stability of network, learning algorithm is set as LM algorithms.
(3) weights and threshold value in BP networks are initialized, be encoded to particle, and the ginseng in particle cluster algorithm is set
Number.
Wherein, initial weight threshold value generally takes the random value between -0.1~0.1.Parameter in particle cluster algorithm includes kind
Group's scale, maximum iteration, speed, inertia weight etc..Inertia weight w uses nonlinear decreasing strategy, specifically, it is calculated
Formula is:
Wherein, wmax、wminIndicate that the upper and lower bound of inertia weight w, iter are current iteration number, Max respectivelyiterFor
Maximum iteration, a are constant coefficient, control the decline curve shape of nonlinear function, take a=10.
Accelerated factor c1And c2Calculation formula be:
Wherein, cstart、cendAccelerated factor c is indicated respectively1Initial value and end value, and 0 < cend< cstart≤ 4, it enables
cstart=3, cend=1;Iter is current iteration number, MaxiterFor maximum iteration.
The improvement formula of speed edges is:
Vmax=v1-(v1-v2)(iter/Maxiter)
Wherein, VmaxFor particle rapidity boundary, v1For the upper limit value of speed, v2For the lower limiting value of speed, iter is current changes
Generation number, MaxiterFor maximum iteration.In the present embodiment, v is taken1=1.5, v2=0.5.Certainly, meeting application demand
In the case of, it can also be by v1、v2It is set as other numerical value.
(4) it determines the fitness function in particle cluster algorithm, and calculates the fitness value of each particle, determine particle position
Individual optimal and global optimum.
In the present embodiment, using the mean square error MSE of BP networks as fitness function, calculation formula is:
Wherein, fiIndicate the predicted value of the network output node of i-th of particle, yiIndicate the network output section of i-th of particle
The actual value of point, N indicate number of particles.
In the size by comparing fitness, individual optimal and global optimum position is found out, i.e., by the adaptation of each particle
The particle of angle value and individual desired positions is compared, the smaller conduct individual optimal solution of the two;Similarly, global optimum is obtained
Solution.
(5) according to the individual optimal and global optimum of particle position, speed and the position of particle are constantly updated, until meeting
End condition, to the initial weight and threshold value in the BP networks after being optimized.
Specifically, speed more new formula is:
Wherein,For speed of i-th of particle in the t times iteration,It is i-th of particle in the t-1 times iteration
Speed, w are inertia weight, r1For the random number between [0,1], r2For the random number between [0,1], pbestidMost for individual
It is excellent, gbestdFor global optimum,For position of i-th of particle in the t-1 times iteration.
Particle position more new formula is:
xi(t+1)=xi(t)+vi(t+1)
Wherein, xi(t+1) it is position of i-th of particle in the t+1 times iteration, xi(t) it is that i-th of particle changes at the t times
Position in generation, vi(t+1) it is speed of i-th of particle in the t+1 times iteration.
In the continuous iterative process of step (5), if output layer fails to obtain expected result, according to output error pair
Weights and threshold value are modified, until the prediction result of output constantly approaches predicted value or output error certain acceptable
In range.General allowable error takes 0.001~0.00001, and the present embodiment chooses 0.0001.
(6) according to the initial weight and threshold value in the BP networks after optimization, BP networks are trained, to be based on
The grain ration Consumption forecast model of Modified particle swarm optimization BP networks.
By above-mentioned steps (1)~(6), so that it may with obtain for predict grain ration consumption figure based on Modified particle swarm optimization
The grain ration Consumption forecast model of BP networks.
Based on the above-mentioned grain ration Consumption forecast model based on Modified particle swarm optimization BP networks, the present invention also provides
A kind of prediction meanss of grain ration consumption figure, the prediction meanss include processor and memory, which is stored in for handling
Instruction in memory, to realize a kind of prediction technique of grain ration consumption figure, i.e.,:When need to the grain ration consumption figure in a certain year into
When row prediction, the data for the impact factor for influencing grain ration consumption figure are updated to above-mentioned based on Modified particle swarm optimization BP networks
Grain ration Consumption forecast model, you can obtain the predicted value of grain ration consumption figure.
The construction method and device of above-mentioned grain ration Consumption forecast model use inertia weight nonlinear decreasing strategy, and right
It is improved on the boundary of maximum speed.Fig. 3 gives above-mentioned Inertia Weight decreases in non-linear curve and is linearly passed with Inertia Weight
Subtract the comparison diagram of curve, as seen from Figure 3, with the increase of iterations, exponentially downward trend becomes smaller Inertia Weight, says
Bright inertia weight nonlinear decreasing strategy can be well adapted for the optimization problem with complex nonlinear variation characteristic.In algorithm
Early period, by the bigger of Inertia Weight setting, is thus easier to seek in global scope in order to avoid being absorbed in local optimum
Look for optimal solution;And in the algorithm later stage, in order to improve precision and the accuracy of optimal solution, by the smaller of Inertia Weight setting.Together
When, this method can obtain preferably restraining effect in early period and later stage.In addition, accelerated factor dynamic changes and can enhance early period
Ability of searching optimum and later stage local search ability, speed maximum value dynamic adjustment can increase global search early period area
Domain and later stage local search ability.
Fig. 4 gives the maximum speed of particle with iterations change curve, from fig. 4, it can be seen that in algorithm iteration
The maximum speed of early period, particle are bigger, and search range is larger, and ability of searching optimum is stronger;And in the iteration later stage, particle
Maximum speed is smaller, and search range is smaller, and local search ability is stronger.With the increase of iterations, the maximum of particle
Speed linearity successively decreases, and is first searched on a large scale so meeting, then carries out the rule of local search.Speed edges are changed
Into, enable it to want extensive search in early period, the later stage reduce search range in terms of increase;And unmodified particle is maximum
Speed is always fixed value 1, defines the search range of particle rapidity.
In order to verify the construction method of above-mentioned grain ration Consumption forecast model and the validity of device, three kinds are built respectively in advance
Survey model:BP network models, standard particle group Optimizing BP Network (PSO-BP) model, Modified particle swarm optimization BP networks of the present invention
(improving PSO-BP) model.Sample is 1984-2014 totally 31 years Chinese grain ration consumption figure data and influence grain ration consumption
The data of each factor of amount, and it is normalized.Then 25 years data are randomly selected to be trained network, are remained
Under 6 annual datas as forecast sample.By constantly testing, training error is minimum when hidden layer is 8 nodes, and error is
0.0013, so hidden layer is ultimately determined to 8, the corresponding network error of different neuron numbers is as shown in table 1.
Table 1
Wherein, BP network maximum frequency of training is set as 10000, learning rate 0.05, training precision 0.0001.To pre-
The index for surveying result performance evaluation has:Mean absolute error MAE, average absolute percent error MAPE, root-mean-square error RMSE,
Square percentage error MSPE, absolute error, relative error, corresponding calculation formula are as follows:
Wherein, fiIndicate predicted value, yiIndicate actual value.
Three kinds of BP networks, standard particle group Optimizing BP Network, Modified particle swarm optimization BP networks models are imitated respectively
True experiment, Fig. 5 give the training sample of Modified particle swarm optimization BP network models of the present invention as a result, three kinds of models prediction knot
Fruit is respectively as shown in Fig. 6, Fig. 7 and Fig. 8.The predicted value of three kinds of models and desired value Comparative result are as shown in table 2, three kinds of models
Evaluation results are as shown in table 3.
Table 2
Sample | Actual value | BP predicted values | PSO-BP predicted values | Improve PSO-BP predicted values |
5 | 26656.81 | 25419.61 | 26548.10 | 26413.49 |
11 | 26956.10 | 26691.09 | 27171.87 | 26491.08 |
12 | 26802.57 | 27220.47 | 26831.02 | 26724.92 |
18 | 23900.03 | 23897.77 | 24032.90 | 23728.90 |
23 | 21004.22 | 20252.15 | 20386.37 | 20415.84 |
27 | 19317.68 | 19884.98 | 19906.06 | 19123.72 |
Table 3
Model | MAE | RMSE | MSPE | MAPE |
BP | 540.27 | 666.21 | 0.6124 | 2.25% |
PSO-BP | 282.01 | 265.53 | 0.4676 | 1.31% |
Improve PSO-BP | 289.91 | 138.90 | 0.4549 | 1.24% |
The construction method and device of grain ration Consumption forecast model of the present invention, using inertia weight nonlinear decreasing strategy,
The adjustment of accelerated factor dynamic, maximum speed value dynamic change adjust weights and threshold value, so as to faster searching most
Excellent weights and threshold value, and improve particle cluster algorithm then can be to avoid being absorbed in local optimum in searching process.Experiment shows relatively
In other traditional prediction techniques, have more using the obtained prediction result of Modified particle swarm optimization BP network models of the present invention
Non-linear mapping capability well, higher precision of prediction, faster convergence rate, and be not easy to be absorbed in local optimum, it effectively improves
The precision of prediction of China's grain ration consumption figure.
Claims (10)
1. a kind of construction method of grain ration Consumption forecast model, which is characterized in that steps are as follows:
It obtains the grain ration consumption figure in each time and influences the impact factor of grain ration consumption figure, and pre-processed;
According to pretreated grain ration consumption figure and the impact factor of influence grain ration consumption figure, the topology knot of structure BP networks
Structure;
The weights and threshold value in BP networks are initialized, are encoded to particle, and the parameter in particle cluster algorithm is set, the ginseng
Number includes the inertia weight of decreases in non-linear;
It determines the fitness function in particle cluster algorithm, and calculates the fitness value of each particle, determine the individual of particle position
Optimal and global optimum;
According to the individual optimal and global optimum of particle position, speed and the position of particle are constantly updated, item is terminated until meeting
Part, to the initial weight and threshold value in the BP networks after being optimized;
According to the initial weight and threshold value in the BP networks after optimization, BP networks are trained, to obtain based on improvement grain
The grain ration Consumption forecast model of subgroup Optimizing BP Network.
2. the construction method of grain ration Consumption forecast model according to claim 1, which is characterized in that decreases in non-linear
The calculation formula of inertia weight is:
Wherein, wmax、wminIndicate that the upper and lower bound of inertia weight w, iter are current iteration number, Max respectivelyiterFor maximum
Iterations, a are constant coefficient, control the decline curve shape of nonlinear function, take a=10.
3. the construction method of grain ration Consumption forecast model according to claim 2, which is characterized in that wmax=0.9, wmin
=0.4.
4. the construction method of grain ration Consumption forecast model according to any one of claim 1-3, which is characterized in that grain
Parameter in swarm optimization further includes particle rapidity, and the calculation formula on wherein particle rapidity boundary is:
Vmax=v1-(v1-v2)(iter/Maxiter)
Wherein, VmaxFor particle rapidity boundary, v1For the upper limit value of speed, v2For the lower limiting value of speed, iter is current iteration time
Number, MaxiterFor maximum iteration.
5. the construction method of grain ration Consumption forecast model according to claim 4, which is characterized in that v1=1.5, v2=
0.5。
6. the construction method of grain ration Consumption forecast model according to any one of claim 1-3, which is characterized in that grain
The calculation formula of fitness function in swarm optimization is:
Wherein, MSE indicates the fitness function in particle cluster algorithm, fiIndicate the prediction of the network output node of i-th of particle
Value, yiIndicate that the actual value of the network output node of i-th of particle, N indicate number of particles.
7. a kind of construction device of grain ration Consumption forecast model, which is characterized in that including processor and memory, the processing
Device is for handling the instruction being stored in the memory to realize following method:
It obtains the grain ration consumption figure in each time and influences the impact factor of grain ration consumption figure, and pre-processed;
According to pretreated grain ration consumption figure and the impact factor of influence grain ration consumption figure, the topology knot of structure BP networks
Structure;
The weights and threshold value in BP networks are initialized, are encoded to particle, and the parameter in particle cluster algorithm is set, the ginseng
Number includes the inertia weight of decreases in non-linear;
It determines the fitness function in particle cluster algorithm, and calculates the fitness value of each particle, determine the individual of particle position
Optimal and global optimum;
According to the individual optimal and global optimum of particle position, speed and the position of particle are constantly updated, item is terminated until meeting
Part, to the initial weight and threshold value in the BP networks after being optimized;
According to the initial weight and threshold value in the BP networks after optimization, BP networks are trained, to obtain based on improvement grain
The grain ration Consumption forecast model of subgroup Optimizing BP Network.
8. the construction device of grain ration Consumption forecast model according to claim 7, which is characterized in that decreases in non-linear
The calculation formula of inertia weight is:
Wherein, wmax、wminIndicate that the upper and lower bound of inertia weight w, iter are current iteration number, Max respectivelyiterFor maximum
Iterations, a are constant coefficient, control the decline curve shape of nonlinear function, take a=10.
9. a kind of prediction side of the grain ration consumption figure of construction method using grain ration Consumption forecast model described in claim 1
Method, which is characterized in that steps are as follows:
Obtain the data for the impact factor for influencing grain ration consumption figure, and by the impact factor of the influence grain ration consumption figure got
Data are brought into the grain ration Consumption forecast model based on Modified particle swarm optimization BP networks, to obtain grain ration Consumption forecast
Value.
10. a kind of prediction meanss of grain ration consumption figure, which is characterized in that including processor and memory, the processor is used for
The instruction being stored in the memory is handled to realize following method:
Obtain the data for the impact factor for influencing grain ration consumption figure, and by the impact factor of the influence grain ration consumption figure got
Data are brought into the grain ration Consumption forecast model based on Modified particle swarm optimization BP networks, to obtain grain ration Consumption forecast
Value.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810481553.1A CN108665322A (en) | 2018-05-18 | 2018-05-18 | The construction method of grain ration Consumption forecast model, Consumption forecast method and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810481553.1A CN108665322A (en) | 2018-05-18 | 2018-05-18 | The construction method of grain ration Consumption forecast model, Consumption forecast method and device |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108665322A true CN108665322A (en) | 2018-10-16 |
Family
ID=63776965
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810481553.1A Pending CN108665322A (en) | 2018-05-18 | 2018-05-18 | The construction method of grain ration Consumption forecast model, Consumption forecast method and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108665322A (en) |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109492816A (en) * | 2018-11-16 | 2019-03-19 | 辽宁工程技术大学 | A kind of coal and gas prominent dynamic prediction method based on hybrid intelligent |
CN110020712A (en) * | 2019-03-26 | 2019-07-16 | 浙江大学 | A kind of optimization population BP neural network forecast method and system based on cluster |
CN110928357A (en) * | 2019-12-16 | 2020-03-27 | 徐州工业职业技术学院 | Maximum power point tracking method of photovoltaic array under time-varying shadow condition |
CN112012875A (en) * | 2020-07-23 | 2020-12-01 | 国网江西省电力有限公司电力科学研究院 | Optimization method of PID control parameters of water turbine regulating system |
CN112183884A (en) * | 2020-10-19 | 2021-01-05 | 河南工业大学 | Grain storage quality prediction method and device |
CN113570120A (en) * | 2021-07-07 | 2021-10-29 | 上海电机学院 | Wind power prediction method and device based on improved particle swarm optimization |
CN114353872A (en) * | 2021-12-30 | 2022-04-15 | 中国电信股份有限公司 | Prediction method and device for machine room temperature |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105260615A (en) * | 2015-10-29 | 2016-01-20 | 河南工业大学 | Grain consumption forecasting method |
-
2018
- 2018-05-18 CN CN201810481553.1A patent/CN108665322A/en active Pending
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105260615A (en) * | 2015-10-29 | 2016-01-20 | 河南工业大学 | Grain consumption forecasting method |
Non-Patent Citations (2)
Title |
---|
李操: "基于IPSO-BP神经网络模型的风电功率预测研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
李雪: "基于粒子群的粮食产量预测神经网络优化研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109492816A (en) * | 2018-11-16 | 2019-03-19 | 辽宁工程技术大学 | A kind of coal and gas prominent dynamic prediction method based on hybrid intelligent |
CN109492816B (en) * | 2018-11-16 | 2021-11-09 | 辽宁工程技术大学 | Coal and gas outburst dynamic prediction method based on hybrid intelligence |
CN110020712A (en) * | 2019-03-26 | 2019-07-16 | 浙江大学 | A kind of optimization population BP neural network forecast method and system based on cluster |
CN110928357A (en) * | 2019-12-16 | 2020-03-27 | 徐州工业职业技术学院 | Maximum power point tracking method of photovoltaic array under time-varying shadow condition |
CN112012875A (en) * | 2020-07-23 | 2020-12-01 | 国网江西省电力有限公司电力科学研究院 | Optimization method of PID control parameters of water turbine regulating system |
CN112012875B (en) * | 2020-07-23 | 2022-03-11 | 国网江西省电力有限公司电力科学研究院 | Optimization method of PID control parameters of water turbine regulating system |
CN112183884A (en) * | 2020-10-19 | 2021-01-05 | 河南工业大学 | Grain storage quality prediction method and device |
CN113570120A (en) * | 2021-07-07 | 2021-10-29 | 上海电机学院 | Wind power prediction method and device based on improved particle swarm optimization |
CN114353872A (en) * | 2021-12-30 | 2022-04-15 | 中国电信股份有限公司 | Prediction method and device for machine room temperature |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108665322A (en) | The construction method of grain ration Consumption forecast model, Consumption forecast method and device | |
CN104361393B (en) | Data predication method is used for based on the improved neural network model of particle swarm optimization algorithm | |
CN108959728B (en) | Radio frequency device parameter optimization method based on deep learning | |
CN111104595B (en) | Deep reinforcement learning interactive recommendation method and system based on text information | |
CN111563706A (en) | Multivariable logistics freight volume prediction method based on LSTM network | |
CN106815782A (en) | A kind of real estate estimation method and system based on neutral net statistical models | |
CN108932671A (en) | A kind of LSTM wind-powered electricity generation load forecasting method joined using depth Q neural network tune | |
CN107729999A (en) | Consider the deep neural network compression method of matrix correlation | |
CN107679617A (en) | The deep neural network compression method of successive ignition | |
CN106600059A (en) | Intelligent power grid short-term load predication method based on improved RBF neural network | |
CN110096634A (en) | A kind of house property data vector alignment schemes based on particle group optimizing | |
CN108764540A (en) | Water supply network pressure prediction method based on parallel LSTM series connection DNN | |
CN104636801A (en) | Transmission line audible noise prediction method based on BP neural network optimization | |
CN108564592A (en) | Based on a variety of image partition methods for being clustered to differential evolution algorithm of dynamic | |
CN106529574A (en) | Image classification method based on sparse automatic encoder and support vector machine | |
CN114511021A (en) | Extreme learning machine classification algorithm based on improved crow search algorithm | |
CN104794367A (en) | Hospitalizing resource scoring and recommending method based on latent factor model | |
CN114943345A (en) | Federal learning global model training method based on active learning and model compression | |
CN108460462A (en) | A kind of Interval neural networks learning method based on interval parameter optimization | |
CN116170328A (en) | Method and device for predicting bandwidth used for graphic coding | |
CN109492816B (en) | Coal and gas outburst dynamic prediction method based on hybrid intelligence | |
CN113987203A (en) | Knowledge graph reasoning method and system based on affine transformation and bias modeling | |
CN109697531A (en) | A kind of logistics park-hinterland Forecast of Logistics Demand method | |
CN108090564A (en) | Based on network weight is initial and the redundant weighting minimizing technology of end-state difference | |
CN117034060A (en) | AE-RCNN-based flood classification intelligent forecasting 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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20181016 |
|
RJ01 | Rejection of invention patent application after publication |