CN110322072A - A kind of economic forecasting method neural network based - Google Patents
A kind of economic forecasting method neural network based Download PDFInfo
- Publication number
- CN110322072A CN110322072A CN201910614831.0A CN201910614831A CN110322072A CN 110322072 A CN110322072 A CN 110322072A CN 201910614831 A CN201910614831 A CN 201910614831A CN 110322072 A CN110322072 A CN 110322072A
- Authority
- CN
- China
- Prior art keywords
- neural network
- economic
- network
- model
- group
- 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
- 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
- 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)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Computational Linguistics (AREA)
- Development Economics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Game Theory and Decision Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
A kind of economic forecasting method neural network based is claimed in the present invention; determine the temporal correlation of each sub-regions in multiple subregions; Space-time Neural Network is constructed based on the temporal correlation of each sub-regions; weight after optimization is assigned to Huffman network; to model and predict; it is output with the predicted value of economical operation maintenance expense, establishes economical operation maintenance expense prediction model, accurate reasonable estimation is carried out to economical operation maintenance expense.Model prediction economic behaviour of the present invention resettles disaggregated model prediction economic situation, few to economic situation labeled data demand in depth prediction, reduces mark work, improves forecasting efficiency.Model of the invention can use initial data, reduce the workload of Feature Engineering.Inventive algorithm overcomes the defect of conventional method very well, can comprehensively, accurate description economic time series changing rule, improve economic forecasting precision.
Description
Technical field
This disclosure relates to which nerual network technique, relates more particularly to a kind of economic forecasting method neural network based.
Background technique
GDP (Gross Domestic Product, abbreviation GDP) is to measure national economic development situation most
The concentrated expression of important an index and Economic Status.It accurately predicts GDP, provides policy ginseng for economic development
It examines, has important practical significance.GDP prediction is the prediction to GDP time series, and GDP time series is special shape
One group of data, the data of front can have an impact subsequent data in this group of data, and this influence relationship is shown as centainly
Long-term change trend or mechanical periodicity etc..And the influence relationship is usually non-linear, is difficult to establish quantitative, fixed mathematical relationship
Formula.Current research method mainly linear prediction technique and Non-linear, such as exponential smoothing, Application of Delphi Method, investment produce
Out, econometric model, Markov forecast techniques, gray prediction, neural network method etc..Economic system is one extremely complex
System, wherein being widely present non-linear, time variation and uncertain interactively.That establishes on the basis of econometrics is each
Kind economic model, most of is all linear model.Linear model also gradually manifests lacking for it while playing a great role
It falls into, that is, is difficult to hold the non-linear phenomena in economic system, thus the error of economic forecasting is necessarily caused to increase.And artificial neuron
Network can then efficiently solve problems, and theoretically artificial neural network can be with Approximation of Arbitrary Nonlinear Function and can be with
Machine adjustment.
Artificial neural network (Artificial neural network, ANN) is simulated human brain working mechanism and is established
Computational theory and technology.Neurobiologist is carried out from single nerve to neural network to stimuli responsive with Self-absorption Correction Factor
Research, and neurophysiologist is that mode studies the understanding of brain with behavioral function from top to bottom, in the part function to brain
On the basis of capable of having certain understanding, mathematician, computer scientist and Engineering Technician attempt to show brain with the method for mathematics
The course of work.Since nineteen forty-three McCulloch and Pitts propose M-P neural network model, ANN has been achieved for resonable
By, the effect of technology and application aspect, application range be related to nervous physiology science, cognitive science, mathematical and physical science, psychic science,
The subjects such as information science, computer science, management science, environmental science.Due to ANN high abstraction simulation human brain thinking process
Store parallel between neuron, transmit information and the information processing function, concurrency, fault-tolerance, stability and self study,
The features such as the non-linear of height, self-organizing and associative memory, makes ANN model have very strong adaptability and application value.
However, space-time data and not always homogeneity, but be usually present heterogeneity, i.e., different classes of space-time data
Temporal correlation and/or spatial coherence be differentiated.And above-mentioned Space-time Neural Network construction method is for all numbers
According to be all made of it is identical handle, therefore for having heterogeneous space-time data, which is simultaneously not suitable for.Tradition
BP algorithm although can guarantee the final convergence of network learning procedure, but there are significant limitations: (1) this network easily falls into office
Portion is minimum, may cannot get total optimization solution;(2) the learning training time is long.Although previous research improves BP algorithm,
Various innovatory algorithms are not compared, and occur new innovatory algorithm, such as Levenberg- in recent years
Therefore Marquardt optimization method is predicted GDP with the artificial neural network of different improved Back Propagations, to disclose it
Changing rule and development trend provide scientific basis to formulate the macro adjustment and control policy of science.
Summary of the invention
In order to solve Current Situation of Neural Network for prediction aspect, especially economic forecasting accuracy, the present invention requests to protect
Protect a kind of economic forecasting method neural network based characterized by comprising
Step 1: the temporal correlation of each sub-regions in multiple subregions, the space-time based on each sub-regions are determined
Correlation constructs Space-time Neural Network;
Step 2: the artificial neural network economical operation maintenance expense prediction based on grey correlation analysis, using certain coding
Scheme encodes weight, and clock synchronization null sequence data collection is clustered, as soon as a group distribution is randomly generated, it corresponds to one group of mind
Connection weight through network extracts the major influence factors of economical operation maintenance expense using grey correlation analysis from because of word bank, makees
For the input variable of artificial neural network, input training sample, calculate its error function value, using error sum of squares inverse as
Fitness, if error is smaller, fitness is bigger, otherwise fitness is big, the superiority and inferiority of connection weight is evaluated with this;
Step 3: Time-space serial data set is divided into multiple sub-districts by the big individual of selection fitness on area of space
Domain is directly hereditary to the next generation, recycles and intersects, and the operation such as variation evolves to current group, next-generation group is generated, according to ash
The color degree of association filters out the highest influence factor of change degree synchronous with economical operation maintenance expense, as the defeated of artificial neural network
Enter, one group of originally determined weight is constantly evolved, until training objective meets condition;
Step 4: the power of the selection initialization Huffman network of artificial neural network economical operation maintenance expense prediction model
The random number being worth between section [0,1], GA encode it;Initial population is randomly generated, Population Size N encodes ANN
All weights and threshold value determine type of coding, length, population scale, crossing-over rate, aberration rate and termination condition, evaluate population, such as
Fruit meets stopping criterion, just decodes, and generates all weights of ANN and threshold value, using BP algorithm training network, generate best initial weights and
Threshold value;
Step 5: being decoded each individual in population, and each individual represents a Huffman network structure, warp
The grey relational grade analysis for operation and maintenance expense influence factor of helping, the N group weight decoded correspond to N number of mutually isostructural network;
Determine the training sample and test sample of network;The corresponding network output of input sample collection is calculated by LM algorithm, determines fitness
Function is chosen to be the inverse of the error performance function of network;
Step 6: calculating the fitness of each chromosome, and error amount is bigger, and corresponding fitness is with regard to smaller;Selection adapts to
Big individual is spent as new parent, eliminates the small individual of fitness;New parent is intersected, mutation operation;In grey
On the basis of association analysis, there is the property for infinitely approaching non-linear continuous function relationship using artificial neural network, made
For the prediction model of economical operation maintenance expense, the iteration of a new round is carried out to new population, until training objective meets termination condition
Until, to obtain one group of optimization weight, the weight after optimization is assigned to Huffman network, to model and predict, with economy
The predicted value of operation and maintenance expense is output, establishes economical operation maintenance expense prediction model, and it is accurate to carry out to economical operation maintenance expense
Reasonable estimation.
Chaos time sequence has a regularity that is inherent, determining, and this regularity results from non-linear, shows time sequence
It is listed in correlation in phase space, this characteristic makes system seem there is certain memory capability, but is difficult to analytic method this
Kind rule is expressed, and this information processing manner is exactly what neural network had, in practical applications, is had for one
Which kind of network structure is the forecasting problem of body should use actually, be to be difficult the problem of being determined in advance.Whether level is more complicated,
The problem of prediction effect is better and one can not clearly answer, but a large amount of numerical experimentation shows forecast result to input
Value and hidden layer value number and insensitive, this be artificial neural network in time series forecasting using providing advantage.
Artificial neural network has many advantages, such as that approximation accuracy is high, pace of learning is fast, not high to data length requirement, thus when non-linear
Between sequence prediction in show its unique superiority.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention without any creative labor, may be used also for those of ordinary skill in the art
To obtain other drawings based on these drawings.
Fig. 1 is a kind of work flow diagram of economic forecasting method neural network based of the invention.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
Referring to attached drawing 1, a kind of work flow diagram of economic forecasting method neural network based of the invention;
A kind of economic forecasting method neural network based is claimed in the present invention characterized by comprising
Step 1: the temporal correlation of each sub-regions in multiple subregions, the space-time based on each sub-regions are determined
Correlation constructs Space-time Neural Network;
Step 2: the artificial neural network economical operation maintenance expense prediction based on grey correlation analysis, using certain coding
Scheme encodes weight, and clock synchronization null sequence data collection is clustered, as soon as a group distribution is randomly generated, it corresponds to one group of mind
Connection weight through network extracts the major influence factors of economical operation maintenance expense using grey correlation analysis from because of word bank, makees
For the input variable of artificial neural network, input training sample, calculate its error function value, using error sum of squares inverse as
Fitness, if error is smaller, fitness is bigger, otherwise fitness is big, the superiority and inferiority of connection weight is evaluated with this;
Step 3: Time-space serial data set is divided into multiple sub-districts by the big individual of selection fitness on area of space
Domain is directly hereditary to the next generation, recycles and intersects, and the operation such as variation evolves to current group, next-generation group is generated, according to ash
The color degree of association filters out the highest influence factor of change degree synchronous with economical operation maintenance expense, as the defeated of artificial neural network
Enter, one group of originally determined weight is constantly evolved, until training objective meets condition;
Step 4: the power of the selection initialization Huffman network of artificial neural network economical operation maintenance expense prediction model
The random number being worth between section [0,1], GA encode it;Initial population is randomly generated, Population Size N encodes ANN
All weights and threshold value determine type of coding, length, population scale, crossing-over rate, aberration rate and termination condition, evaluate population, such as
Fruit meets stopping criterion, just decodes, and generates all weights of ANN and threshold value, using BP algorithm training network, generate best initial weights and
Threshold value;
Step 5: being decoded each individual in population, and each individual represents a Huffman network structure, warp
The grey relational grade analysis for operation and maintenance expense influence factor of helping, the N group weight decoded correspond to N number of mutually isostructural network;
Determine the training sample and test sample of network;The corresponding network output of input sample collection is calculated by LM algorithm, determines fitness
Function is chosen to be the inverse of the error performance function of network;
Step 6: calculating the fitness of each chromosome, and error amount is bigger, and corresponding fitness is with regard to smaller;Selection adapts to
Big individual is spent as new parent, eliminates the small individual of fitness;New parent is intersected, mutation operation;In grey
On the basis of association analysis, there is the property for infinitely approaching non-linear continuous function relationship using artificial neural network, made
For the prediction model of economical operation maintenance expense, the iteration of a new round is carried out to new population, until training objective meets termination condition
Until, to obtain one group of optimization weight, the weight after optimization is assigned to Huffman network, to model and predict, with economy
The predicted value of operation and maintenance expense is output, establishes economical operation maintenance expense prediction model, and it is accurate to carry out to economical operation maintenance expense
Reasonable estimation.
Preferably, step 1: determining the temporal correlation of each sub-regions in multiple subregions, is based on each sub-regions
Temporal correlation construct Space-time Neural Network, specifically include:
Based on the sequencing selection of standard geometry distribution, training sample is input to BP neural network and is learnt, by system
Initialization, sets ant number, the number of iterations, using genetic algorithm to the parameter connection weight w and threshold θ of BP neural network
It optimizes, the starting point of ant is set, construction ant returns to the path of starting point, that is, solves, and updates table;
Judge whether ant all finds food, if so, pheromones are reconfigured, if not construction ant returns to starting point
Path, that is, solve, update table from new, establish the optimal Nonlinear Prediction Models of optimal economic time series;
Economic time series test sample is predicted using the optimal nonlinear Economic Forecasting Mathematical Model of foundation, is examined
Model is effective, finally predicts the future time level of economic development
Further, step 2: the artificial neural network economical operation maintenance expense prediction based on grey correlation analysis uses
Certain encoding scheme encodes weight, and clock synchronization null sequence data collection is clustered, as soon as a group distribution is randomly generated, it is corresponded to
The connection weight of one group of neural network, the main influence of economical operation maintenance expense is extracted from because of word bank using grey correlation analysis
Factor inputs training sample, its error function value is calculated, with error sum of squares as the input variable of artificial neural network
Inverse is used as fitness, if error is smaller, fitness is bigger, otherwise fitness is big, the superiority and inferiority of connection weight is evaluated with this, specifically
Include:
It determines the temporal correlation of each sub-regions in multiple subregions, is made with probability value assigned by each individual
It can be genetic to follow-on probability for it, generate next-generation group with the method that ratio selects based on these probability values,
It determines the delay time T and Embedded dimensions m of economic time data sequence, and weight is carried out to economic time series sample according to τ, m
Structure;
Calculate the auto-correlation coefficient of the Time-space serial data in the subregion;Then based on the auto-correlation coefficient of calculating, really
The partial correlation coefficient of Time-space serial data in the fixed subregion;Then it according to partial correlation coefficient, determines and characterizes in the subregion
Time-space serial data temporal correlation space-time length of delay;
Phase space reconfiguration is carried out to preprocessed data, selects coding strategy, the economic time series after reconstruct are drawn
It is divided into training sample and test sample two parts, training sample is for establishing Economic Forecasting Mathematical Model, to all individuals in group
Descending sort is carried out by its fitness size, test sample tests the Economic Forecasting Mathematical Model of foundation, will be in solution space
Solution data are expressed as the genotype string structure data in hereditary space, and the various combination of these string structure data just constitutes different
Coding;
Definition adapts to value function f (x), determines Genetic Strategies, chaotic economy time series is restored to regular warp
The temporal correlation for the time series data each sub-regions of helping can be calculated by following operation, according to specific Solve problems,
A probability assignments table is designed, each probability value is distributed into each individual, including selection group size by above-mentioned arrangement this journey
N, selection intersect, variation method, and determine the genetic parameters such as crossover probability, mutation probability;
Random initializtion generates group P, the decoded adaptive value f (x) of individual in population bit string is calculated, according to hereditary plan
Slightly, group is acted on selection, intersection and mutation operator, forms next-generation group.
Preferably, step 3: Time-space serial data set is divided into more by the big individual of selection fitness on area of space
Sub-regions are directly hereditary to the next generation, recycle and intersect, and the operation such as variation evolves to current group, generate next-generation group,
The highest influence factor of change degree synchronous with economical operation maintenance expense is filtered out according to grey relational grade, as artificial neural network
The input of network, one group of originally determined weight are constantly evolved, until training objective meets condition, are specifically included:
RBF neural network structure figure is constructed according to the data of selection, determines the number of nodes of input quantity, hidden layer, output quantity,
And determine activation primitive and Gaussian function;
The set of an ant sample data is generated at random;
According to desired precision, calculate each ant just when;
With the increase of time t, and using the formula of Chaos Ant Colony Optimization, optimal value is obtained by Chaos Search;
By test of many times, the sample after training is obtained, and observe training curve.
Preferably, step 4: step 4: the selection initialization of artificial neural network economical operation maintenance expense prediction model
Random number of the weight of Huffman network between section [0,1], GA encode it;Initial population, population is randomly generated
Size is N, and all weights of coding ANN and threshold value determine type of coding, length, population scale, crossing-over rate, aberration rate and termination
Condition evaluates population, if meeting stopping criterion, just decodes, and generates all weights of ANN and threshold value, utilizes BP algorithm training net
Network generates best initial weights and threshold value, specifically includes:
The data set W that sample size is n will have been obtained and fifty-fifty, be randomly divided into K parts, and n1∪n2∪...nk-1∪nk=
W,Then n is randomly choosedi(1=1,2 ... it k) is used as training set, remaining (K-1) part is made
For test set;
So in turn, circulation will obtain K error result, this K error result can be used as using least square method
Last evaluating basis can also be used for multiple times K folding cross validation, measure the accurate of model with final average accuracy
Degree, and then the superiority and inferiority of evaluation model.
Least square method specific steps include:
The first step calculates the weight between the hidden layer of continuous Fourier neural network and characterization layer according to the following formula:
Wherein, T "t+1Indicate that weight when the t+1 times recurrence between hidden layer and characterization layer, t indicate continuous Fourier's nerve
The recursive number of network weight training, Tt" indicate that weight when the t times recurrence between hidden layer and characterization layer, μ indicate hidden layer and table
The learning rate of weight between layer is levied, general value range is 0 < μ < 1,Indicate the absolute mistake of sample when the t times recurrence
Difference operates the partial derivative of weight between hidden layer and characterization layer, and α is momentum variable, and general value range is 0.9 < α < 1, Δ
TtIt " indicates hidden layer when the t times recurrence and characterizes the weight school deviator between layer.
Second step calculates the weight between the calling layer and hidden layer of continuous Fourier neural network according to the following formula:
Wherein, T 't+1Indicate that weight when the t+1 times recurrence between calling layer and hidden layer, t indicate continuous Fourier's nerve
The recursive number of network weight training, Tt' indicate the t time recurrence when calling layer and hidden layer between weight, μ indicate calling layer with
The learning rate of weight between hidden layer, general value range are 0 < μ < 1,Indicate the absolute error of sample when the t times recurrence
Partial derivative operation to weight between calling layer and hidden layer, α is momentum variable, and general value range is 0.9 < α < 1, Δ Tt′
Weight school deviator between calling layer and hidden layer when indicating the t times recurrence.
Third step calculates the scaling variable of continuous Fourier neural network hidden layer Fourier's activation primitive according to the following formula:
Wherein, mt+1Indicate that the scaling variable of hidden layer Fourier activation primitive when the t+1 times recurrence, t indicate continuous Fourier
The recursive number of neural network weight training, mtIndicate the scaling variable of hidden layer Fourier activation primitive when the t times recurrence, μ table
Show the learning rate of hidden layer Fourier's activation primitive scaling variable, general value range is 0 < μ < 1,When indicating the t times recurrence
The absolute error of sample operates the partial derivative of hidden layer Fourier's activation primitive scaling variable, and α is momentum variable, general value model
It encloses for 0.9 < α < 1, Δ mtIndicate the school deviator of hidden layer Fourier activation primitive scaling variable when the t times recurrence.
4th step calculates the offset variable of continuous Fourier neural network hidden layer Fourier's activation primitive according to the following formula:
Wherein, nt+1Indicate the offset variable of hidden layer Fourier activation primitive when the t+1 times recurrence, t indicates continuous Fourier
The recursive number of neural network weight training, ntIndicate the offset variable of hidden layer Fourier activation primitive when the t times recurrence, μ table
Show the learning rate of hidden layer Fourier's activation primitive offset variable, general value range is 0 < μ < 1,When indicating the t times recurrence
The absolute error of sample operates the partial derivative of hidden layer Fourier's activation primitive offset variable, and α is momentum variable, general value model
It encloses for 0.9 < α < 1, Δ ntIndicate the school deviator of hidden layer Fourier activation primitive offset variable when the t times recurrence.
5th step judges whether to reach maximum recurrence number, if it is not, returning to the first step, if so, stopping recurrence, obtains net
The best initial weights of network, the optimal scaling variable and optimum displacement variable of Fourier's activation primitive and optimal hidden layer characterization.
Further, step 5: being decoded each individual in population, and each individual represents a Huffman net
Network structure, the grey relational grade analysis of economical operation maintenance expense influence factor, the N group weight decoded correspond to N number of identical knot
The network of structure;Determine the training sample and test sample of network;The corresponding network output of input sample collection is calculated by LM algorithm, really
Determine fitness function, be chosen to be the inverse of the error performance function of network, specifically further include:
The Relative increasing rate for calculating crude oil economic forecasting apparatus system, according to Relative increasing rate to crude oil economic forecasting
The control loop of device is matched;
Test and excitation signal is added in open cycle system, the output of acquisition system;
It is newrb () function in MATLAB to construct RBF neural network model, newrb () function itself can be certainly
Row changes the node number of hidden layer in neural network, until mean square error (RMSE) reaches in accuracy value required by system,
And utilizing simulink software is institute's prediction model compilation run function;
According to the data of input and output, crude oil economic forecasting device is modeled by support vector regression algorithm;
Using obtained model as the prediction model in economic model forecast control method;
Functional form is as follows:
Net=newrb (P, T, goal, spread, MN, DF)
In formula, what spread was indicated is dispersion constant, and value can have direct influence to the predictablity rate of network,
When sperad value is larger, RBF neural can more level off to prediction curve, and the flatness of system is even better, conversely, system
Error can be larger;What goal was indicated is system mean square error, and MN refers to neuronal quantity maximum value;DF is indicated in test twice
The knots modification of neuron number;
The setting value of model predictive controller tracking is provided according to production requirement, and according to energy prices and product price
Economic indicator is provided, in summary the two provides the objective function of model predictive controller;
Model predictive controller is applied in the control of device, when carrying out PREDICTIVE CONTROL using neural network, model is defeated
Prediction error between neural network output out, is used as the training signal of neural network, according to actual requirement to device
Each control variable scan for solving, optimal control sequence is applied in each control variable of device, realize exist
Make the target of maximization of economic benefit under conditions of qualified product.
Upper is only presently preferred embodiments of the present invention, is not intended to limit the invention, all in spirit and original of the invention
Within then, any modification, equivalent substitution, improvement and etc. done be should be included within the scope of the present invention.
Claims (6)
1. a kind of economic forecasting method neural network based characterized by comprising
Step 1: the temporal correlation of each sub-regions in multiple subregions, the temporal and spatial correlations based on each sub-regions are determined
Property constructs Space-time Neural Network;
Step 2: the artificial neural network economical operation maintenance expense prediction based on grey correlation analysis, using certain encoding scheme
Weight is encoded, clock synchronization null sequence data collection is clustered, as soon as a group distribution is randomly generated, it corresponds to one group of nerve net
The connection weight of network extracts the major influence factors of economical operation maintenance expense using grey correlation analysis, as people from because of word bank
The input variable of artificial neural networks inputs training sample, calculates its error function value, using error sum of squares inverse as adaptation
Degree, if error is smaller, fitness is bigger, otherwise fitness is big, the superiority and inferiority of connection weight is evaluated with this;
Step 3: Time-space serial data set is divided into multiple subregions, directly by the big individual of selection fitness on area of space
It connects and is hereditary to the next generation, recycle and intersect, the operation such as variation evolves to current group, generates next-generation group, closes according to grey
Connection degree filters out the highest influence factor of change degree synchronous with economical operation maintenance expense, as the input of artificial neural network,
One group of originally determined weight is constantly evolved, until training objective meets condition;
Step 4: artificial neural network economical operation maintenance expense prediction model selection initialization Huffman network weight be
Random number between section [0,1], GA encode it;Initial population is randomly generated, Population Size N, coding ANN are all
Weight and threshold value determine type of coding, length, population scale, crossing-over rate, aberration rate and termination condition, evaluate population, if full
Sufficient stopping criterion just decodes, and generates all weights of ANN and threshold value, using BP algorithm training network, generates best initial weights and threshold
Value;
Step 5: being decoded each individual in population, and each individual represents a Huffman network structure, economy fortune
The grey relational grade analysis of row maintenance expense influence factor, the N group weight decoded correspond to N number of mutually isostructural network;It determines
The training sample and test sample of network;The corresponding network output of input sample collection is calculated by LM algorithm, determines fitness function,
It is chosen to be the inverse of the error performance function of network;
Step 6: calculating the fitness of each chromosome, and error amount is bigger, and corresponding fitness is with regard to smaller;Select fitness big
Individual as new parent, eliminate the small individual of fitness;New parent is intersected, mutation operation;In grey correlation
On the basis of analysis, there is the property for infinitely approaching non-linear continuous function relationship using artificial neural network, as warp
The prediction model for operation and maintenance expense of helping carries out the iteration of a new round to new population, until training objective meets termination condition,
To obtain one group of optimization weight, the weight after optimization is assigned to Huffman network, to model and predict, with economical operation
The predicted value of maintenance expense is output, establishes economical operation maintenance expense prediction model, and it is accurate reasonable to carry out to economical operation maintenance expense
Estimation.
2. a kind of economic forecasting method neural network based as claimed in claim 1, it is characterised in that:
Step 1: the temporal correlation of each sub-regions in multiple subregions, the temporal and spatial correlations based on each sub-regions are determined
Property constructs Space-time Neural Network, specifically includes:
Based on the sequencing selection of standard geometry distribution, training sample is input to BP neural network and is learnt, system is initial
Change, set ant number, the number of iterations, is carried out using parameter connection weight w and threshold θ of the genetic algorithm to BP neural network
The starting point of ant is arranged in optimization, and construction ant returns to the path of starting point, that is, solves, and updates table;
Judge whether ant all finds food, if so, pheromones are reconfigured, if not construction ant returns to the road of starting point
Diameter solves, update table from new, establish the optimal Nonlinear Prediction Models of optimal economic time series;
Economic time series test sample is predicted using the optimal nonlinear Economic Forecasting Mathematical Model of foundation, testing model
Effectively, finally the future time level of economic development is predicted.
3. a kind of economic forecasting method neural network based as claimed in claim 1, it is characterised in that:
Step 2: the artificial neural network economical operation maintenance expense prediction based on grey correlation analysis, using certain encoding scheme
Weight is encoded, clock synchronization null sequence data collection is clustered, as soon as a group distribution is randomly generated, it corresponds to one group of nerve net
The connection weight of network extracts the major influence factors of economical operation maintenance expense using grey correlation analysis, as people from because of word bank
The input variable of artificial neural networks inputs training sample, calculates its error function value, using error sum of squares inverse as adaptation
Degree, if error is smaller, fitness is bigger, otherwise fitness is big, and the superiority and inferiority of connection weight is evaluated with this, is specifically included:
The temporal correlation for determining each sub-regions in multiple subregions, using it is each individual assigned by probability value as its
It can be genetic to follow-on probability, generate next-generation group with the method that ratio selects based on these probability values, determined
The delay time T and Embedded dimensions m of economic time data sequence, and economic time series sample is reconstructed according to τ, m;
Calculate the auto-correlation coefficient of the Time-space serial data in the subregion;Then based on the auto-correlation coefficient of calculating, determining should
The partial correlation coefficient of Time-space serial data in subregion;Then according to partial correlation coefficient, determine characterize in the subregion when
The space-time length of delay of the temporal correlation of null sequence data;
Phase space reconfiguration is carried out to preprocessed data, coding strategy is selected, the economic time series after reconstruct is divided into
Training sample and test sample two parts, training sample press it for establishing Economic Forecasting Mathematical Model, to all individuals in group
Fitness size carries out descending sort, and test sample tests the Economic Forecasting Mathematical Model of foundation, by the skill in solution space
According to the genotype string structure data for being expressed as hereditary space, the various combination of these string structure data just constitutes different volumes
Code;
Definition adapts to value function f (x), determines Genetic Strategies, chaotic economy time series is restored to the regular daylight saving time
Between the temporal correlations of sequence data each sub-regions can be calculated by following operation, according to specific Solve problems, design
Each probability value is distributed to each individual, including selection group size n, choosing by above-mentioned arrangement this journey by one probability assignments table
It selects, intersect, variation method, and determining the genetic parameters such as crossover probability, mutation probability;
Random initializtion generates group P, calculates the decoded adaptive value f (x) of individual in population bit string, according to Genetic Strategies, fortune
Group is acted on selection, intersection and mutation operator, forms next-generation group.
4. a kind of economic forecasting method neural network based as claimed in claim 1, it is characterised in that:
Step 3: Time-space serial data set is divided into multiple subregions, directly by the big individual of selection fitness on area of space
It connects and is hereditary to the next generation, recycle and intersect, the operation such as variation evolves to current group, generates next-generation group, closes according to grey
Connection degree filters out the highest influence factor of change degree synchronous with economical operation maintenance expense, as the input of artificial neural network,
One group of originally determined weight is constantly evolved, and until training objective meets condition, is specifically included:
RBF neural network structure figure is constructed according to the data of selection, determines the number of nodes of input quantity, hidden layer, output quantity, and really
Determine activation primitive and Gaussian function;
The set of an ant sample data is generated at random;
According to desired precision, calculate each ant just when;
With the increase of time t, and using the formula of Chaos Ant Colony Optimization, optimal value is obtained by Chaos Search;
By test of many times, the sample after training is obtained, and observe training curve.
5. a kind of economic forecasting method neural network based as claimed in claim 1, it is characterised in that:
Step 4: step 4: the selection initialization Huffman network of artificial neural network economical operation maintenance expense prediction model
Random number of the weight between section [0,1], GA encode it;Initial population, Population Size N, coding is randomly generated
All weights of ANN and threshold value determine type of coding, length, population scale, crossing-over rate, aberration rate and termination condition, evaluation kind
Group just decodes if meeting stopping criterion, generates all weights of ANN and threshold value, using BP algorithm training network, generates optimal
Weight and threshold value, specifically include:
The data set W that sample size is n will have been obtained and fifty-fifty, be randomly divided into K parts, and n1∪n2∪...nk-1∪nk=W,Then n is randomly choosedi(i=1,2 ... k) it is used as training set, remaining (K-1) part conduct
Test set;
So in turn, circulation will obtain K error result, this K error result, which is done one, averagely can be used as last comment
Determine foundation, K folding cross validation can also be used for multiple times, the accuracy of model, Jin Erping are measured with final average accuracy
The superiority and inferiority of valence model.
6. a kind of economic forecasting method neural network based as claimed in claim 1, it is characterised in that:
Step 5: being decoded each individual in population, and each individual represents a Huffman network structure, economy fortune
The grey relational grade analysis of row maintenance expense influence factor, the N group weight decoded correspond to N number of mutually isostructural network;It determines
The training sample and test sample of network;The corresponding network output of input sample collection is calculated by LM algorithm, determines fitness function,
It is chosen to be the inverse of the error performance function of network, specifically further include:
The Relative increasing rate for calculating crude oil economic forecasting apparatus system, according to Relative increasing rate to crude oil economic forecasting device
Control loop matched;
Test and excitation signal is added in open cycle system, the output of acquisition system;
It is newrb () function in MATLAB to construct RBF neural network model, newrb () function itself can voluntarily change
The node number of hidden layer in Modified neural network, until mean square error (RMSE) reaches in accuracy value required by system, and benefit
It is institute's prediction model compilation run function with simulink software;
According to the data of input and output, crude oil economic forecasting device is modeled by support vector regression algorithm;
Using obtained model as the prediction model in economic model forecast control method;
Functional form is as follows:
Net=newrb (P, T, goal, spread, MN, DF)
In formula, what spread was indicated is dispersion constant, and value can have direct influence, sperad value to the predictablity rate of network
When larger, RBF neural can more level off to prediction curve, and the flatness of system is even better, conversely, the error of system can be compared with
Greatly;What goal was indicated is system mean square error, and MN refers to neuronal quantity maximum value;DF indicates neuron in test twice
Several knots modifications;
The setting value of model predictive controller tracking is provided according to production requirement, and is provided according to energy prices and product price
Economic indicator, in summary the two provides the objective function of model predictive controller;
By model predictive controller be applied to device control in, using neural network carry out PREDICTIVE CONTROL when, model output with
Prediction error between neural network output, is used as the training signal of neural network, according to actual requirement to each of device
A control variable scans for solving, and optimal control sequence is applied in each control variable of device, realizes in product
Make the target of maximization of economic benefit under conditions of up-to-standard.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910614831.0A CN110322072A (en) | 2019-07-09 | 2019-07-09 | A kind of economic forecasting method neural network based |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910614831.0A CN110322072A (en) | 2019-07-09 | 2019-07-09 | A kind of economic forecasting method neural network based |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110322072A true CN110322072A (en) | 2019-10-11 |
Family
ID=68123234
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910614831.0A Pending CN110322072A (en) | 2019-07-09 | 2019-07-09 | A kind of economic forecasting method neural network based |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110322072A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112270336A (en) * | 2020-09-07 | 2021-01-26 | 河北工业大学 | GA-BP working condition identification method and system |
CN116303094A (en) * | 2023-05-10 | 2023-06-23 | 江西财经大学 | Multipath coverage test method based on RBF neural network and individual migration |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101599138A (en) * | 2009-07-07 | 2009-12-09 | 武汉大学 | Land evaluation method based on artificial neural network |
CN103279813A (en) * | 2013-06-21 | 2013-09-04 | 哈尔滨工业大学(威海) | Steam load prediction method |
CN104636801A (en) * | 2013-11-08 | 2015-05-20 | 国家电网公司 | Transmission line audible noise prediction method based on BP neural network optimization |
CN105184678A (en) * | 2015-09-18 | 2015-12-23 | 齐齐哈尔大学 | Method for constructing photovoltaic power station generation capacity short-term prediction model based on multiple neural network combinational algorithms |
CN106156844A (en) * | 2015-03-23 | 2016-11-23 | 日本电气株式会社 | For building Space-time Neural Network and utilizing its method and apparatus being predicted |
CN107085371A (en) * | 2017-05-03 | 2017-08-22 | 北京国控天成科技有限公司 | Crude(oil)unit economic model forecast control method based on data-driven |
CN107194461A (en) * | 2017-05-27 | 2017-09-22 | 国网浙江省电力公司经济技术研究院 | A kind of artificial neural network distribution operation maintenance expense Forecasting Methodology |
CN108804800A (en) * | 2018-06-04 | 2018-11-13 | 桂林电子科技大学 | Lithium ion battery SOC on-line prediction methods based on echo state network |
-
2019
- 2019-07-09 CN CN201910614831.0A patent/CN110322072A/en active Pending
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101599138A (en) * | 2009-07-07 | 2009-12-09 | 武汉大学 | Land evaluation method based on artificial neural network |
CN103279813A (en) * | 2013-06-21 | 2013-09-04 | 哈尔滨工业大学(威海) | Steam load prediction method |
CN104636801A (en) * | 2013-11-08 | 2015-05-20 | 国家电网公司 | Transmission line audible noise prediction method based on BP neural network optimization |
CN106156844A (en) * | 2015-03-23 | 2016-11-23 | 日本电气株式会社 | For building Space-time Neural Network and utilizing its method and apparatus being predicted |
CN105184678A (en) * | 2015-09-18 | 2015-12-23 | 齐齐哈尔大学 | Method for constructing photovoltaic power station generation capacity short-term prediction model based on multiple neural network combinational algorithms |
CN107085371A (en) * | 2017-05-03 | 2017-08-22 | 北京国控天成科技有限公司 | Crude(oil)unit economic model forecast control method based on data-driven |
CN107194461A (en) * | 2017-05-27 | 2017-09-22 | 国网浙江省电力公司经济技术研究院 | A kind of artificial neural network distribution operation maintenance expense Forecasting Methodology |
CN108804800A (en) * | 2018-06-04 | 2018-11-13 | 桂林电子科技大学 | Lithium ion battery SOC on-line prediction methods based on echo state network |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112270336A (en) * | 2020-09-07 | 2021-01-26 | 河北工业大学 | GA-BP working condition identification method and system |
CN116303094A (en) * | 2023-05-10 | 2023-06-23 | 江西财经大学 | Multipath coverage test method based on RBF neural network and individual migration |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Duong | Bio-inspired computing | |
Ersel Canyurt et al. | Energy demand estimation based on two-different genetic algorithm approaches | |
CN108876001B (en) | Short-term power load prediction method based on twin support vector machine | |
CN102622418B (en) | Prediction device and equipment based on BP (Back Propagation) nerve network | |
CN109659933A (en) | A kind of prediction technique of power quality containing distributed power distribution network based on deep learning model | |
CN108764473A (en) | A kind of BP neural network water demands forecasting method based on correlation analysis | |
CN102520342B (en) | Analog circuit test node selecting method based on dynamic feedback neural network modeling | |
CN110232203A (en) | Knowledge distillation optimization RNN has a power failure prediction technique, storage medium and equipment in short term | |
CN108320053A (en) | A kind of region electricity demand forecasting method, apparatus and system | |
CN104318020A (en) | Multi-objective sensor distributed point optimizing method on basis of self-adaptive differential evolution | |
CN102024179A (en) | Genetic algorithm-self-organization map (GA-SOM) clustering method based on semi-supervised learning | |
CN108694473A (en) | Building energy consumption prediction technique based on RBF neural | |
Oludolapo et al. | Comparing performance of MLP and RBF neural network models for predicting South Africa's energy consumption | |
CN109214503A (en) | Project of transmitting and converting electricity cost forecasting method based on KPCA-LA-RBM | |
CN109919356A (en) | One kind being based on BP neural network section water demand prediction method | |
CN105184415A (en) | Power distribution network reconstruction design method | |
CN109934422A (en) | Neural network wind speed prediction method based on time series data analysis | |
CN110322072A (en) | A kind of economic forecasting method neural network based | |
CN110163444A (en) | A kind of water demand prediction method based on GASA-SVR | |
CN109242558A (en) | A kind of management data prediction technique, readable storage medium storing program for executing and forecasting system neural network based | |
CN105844353A (en) | Aquatic product price prediction method and device | |
CN101739614A (en) | Hierarchy-combined prediction method for communication service | |
CN102456109A (en) | Training method for Trojan horse incident prediction least squares support vector machine, and prediction method | |
CN110298506A (en) | A kind of urban construction horizontal forecast system | |
Vavilova et al. | Application of genetic algorithm for adjusting the structure of multilayered neural network for prediction of investment processes |
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: 20191011 |
|
RJ01 | Rejection of invention patent application after publication |