CN109118013A - A kind of management data prediction technique, readable storage medium storing program for executing and forecasting system neural network based - Google Patents
A kind of management data prediction technique, readable storage medium storing program for executing and forecasting system neural network based Download PDFInfo
- Publication number
- CN109118013A CN109118013A CN201810993073.3A CN201810993073A CN109118013A CN 109118013 A CN109118013 A CN 109118013A CN 201810993073 A CN201810993073 A CN 201810993073A CN 109118013 A CN109118013 A CN 109118013A
- Authority
- CN
- China
- Prior art keywords
- value
- neural network
- parameter
- management data
- prediction technique
- 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
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0637—Strategic management or analysis, e.g. setting a goal or target of an organisation; Planning actions based on goals; Analysis or evaluation of effectiveness of goals
- G06Q10/06375—Prediction of business process outcome or impact based on a proposed change
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Strategic Management (AREA)
- General Physics & Mathematics (AREA)
- Economics (AREA)
- Entrepreneurship & Innovation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- Development Economics (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- Educational Administration (AREA)
- Health & Medical Sciences (AREA)
- Game Theory and Decision Science (AREA)
- Artificial Intelligence (AREA)
- Operations Research (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of management data prediction techniques neural network based, comprising: obtains the historical data of target object;It is calculated according to the historical data, obtains optimization input parameter value;The parameter of neural network is initialized, and model training is carried out according to the optimization input parameter value, the model after being trained;In model after inputting parameter to training to be measured, prediction output valve is obtained.It solves the defect for being easily trapped into local minimum in neural network algorithm, and by optimization input parameter value, reduces the training calculation amount of model, improve the speed of operation, improve the precision of prediction.
Description
Technical field
The invention belongs to data prediction and calculating fields, more particularly to a kind of management data neural network based
Prediction technique, readable storage medium storing program for executing and forecasting system.
Background technique
Indices Analysis refers to summary and evaluates the analysis indexes of financial position of the enterprise and management performance, including payment of debts
Capacity index, operation ability index, Profitability Index and developing ability index.In the management data or finance of analysis enterprise
When data, it will usually which analysis has debt paying ability, profitability, contribution ability etc..Specifically, including net assets income ratio, net profit
Moisten the data such as growth rate, main business income growth rate, rate of gross profit, debt ratio, cash flow.In addition, analyzing national related data
When, the growth rate of GDP embodies a national macroeconomy situation, and CPI then directly affects the purchasing power of resident, further investigation point
These macro-performance indicators are analysed, for studying China's process of economic development, economic growth rule is probed into, carries out marcoeconomic regulation and control
And it formulates economic policy and has very important significance.GDP and CPI is as measurement individual, family, enterprise and national economy portion
Door principal economic indicators, the tendency and fluctuating range of these indexs of Accurate Prediction, then for enterprise's reasonable arrangement production and operation with
The macro economic policy that government department formulates science provides effective foundation.
Inventor has found that traditional economic forecasting is mainly with certainty Time series analysis method, correlation analysis
Main, gray prediction method and other combination forecasting methods are also often studied personnel and use.Such method be mainly to data because
Relationship between fruit relationship and time series is analyzed.In actual prediction analytic process, multicollinearity, error sequence are handled
When the problems such as related, information content can be inevitably lost, simulation is ineffective, and precision of prediction is unsatisfactory.Actually
Management data and index or country GDP and CPI of company etc. are influenced by multiple factors, and various factors asks relationship complexity, are in
Reveal complicated time series and non-linear property, predicts extremely difficult.In addition, much being predicted using neural network model
When, it is easily ensnared into local minimum, is unfavorable for the prediction to data.How the pre- of history big data progress index is utilized
It surveys, and improving precision of prediction is problem that at present urgently can not be to be resolved.
Summary of the invention
In view of this, solving BP mind present invention contemplates that providing a kind of management data prediction technique neural network based
Through the defect for being easily trapped into local minimum in network algorithm, and by optimization input parameter value, reduce the instruction of model
Practice calculation amount, improves the speed of operation, improve the precision of prediction.Can preferably carry out through the invention company and other
Number it was predicted that better help can be provided for user, also improve user experience.
In order to achieve the above objectives, the technical scheme of the present invention is realized as follows:
First aspect present invention provides a kind of management data prediction technique neural network based, comprising the following steps:
Obtain the historical data of target object;
It is calculated according to the historical data, obtains optimization input parameter value;
The parameter of neural network is initialized, and model training is carried out according to the optimization input parameter value, after being trained
Model;
In model after inputting parameter to training to be measured, prediction output valve is obtained.
It is described to be calculated according to the historical data in above scheme, obtain optimization input parameter value step specifically:
Obtain historical data values and n parameter;
Calculate the variance contribution ratio of the n parameter;
N variance contribution ratio of the calculating is arranged from big to small;
M variance contribution ratio in the top is subjected to summation addition, obtain variance contribution ratio and judges the variance tribute
It offers rate and whether is greater than optimization threshold value W;
If it is greater than or equal to the optimization threshold value W, then choosing the m variance is input parameter value;
Wherein n is the natural number more than or equal to 2, and m is the natural number less than or equal to n.
In above scheme, the optimization threshold value W is 85-100%.
In above scheme, the parameter of the initialization neural network, and model is carried out according to the optimization input parameter value
Training, the model after being trained specifically:
Initialize training parameter;
The position of each particle is mapped in BP neural network;
The training BP neural network, calculates the fitness function value of each particle;
It is iterated calculating, more new particle individual optimal value;
Population Regeneration global optimum updates particle position and speed, according to individual optimal and global optimum to particle
Position and speed is adjusted;
Judge whether to meet termination condition, be returned if being unsatisfactory for, continues to train;Repetitive exercise is saved if meeting
As a result;
Every dimension data of training result is mapped in BP neural network;
BP neural network model after being trained.
In above scheme, the training parameter includes population invariable number, the number of iterations, population particle scale and its position and speed
One or more of degree, speed value range, weight, the weight are w, c1、c2。
In above scheme, the fitness function value for calculating each particle specifically:
It is calculated by the functional value that following formula carries out particle fitness:
Wherein, wnjkIt is n-th of output valve of network, tnIt is the actual value of n training sample, s2 is output layer neuron
Number.
It is described to be iterated calculating, more new particle individual optimal value specifically: more each particle is worked as in above scheme
Preceding calculating fitness function value will be worked as if the fitness value of current iteration is better than the fitness function value of preceding an iteration
Preceding individual values are as new optimal value.
In above scheme, the Population Regeneration global optimum specifically: the fitness value of population is calculated, if current kind
The fitness value of group global optimum is better than the fitness value of last iteration, selects the global optimum of current iteration as newly
Global optimum.
Second aspect of the present invention provides a kind of computer readable storage medium, deposits on the computer readable storage medium
Management data prediction technique program neural network based is contained, when the management data prediction technique program is executed by processor
It realizes such as the step of above-mentioned management data prediction technique.
Third aspect present invention additionally provides a kind of management data forecasting system, the system comprises: memory, processor
And it is stored in the program for the management data prediction technique that can be run on the memory and on the processor, it is described to manage number
It is predicted that realizing when the program of method is executed by the processor such as the step of above-mentioned management data prediction technique.
Present invention contemplates that providing a kind of management data prediction technique neural network based, solves BP neural network algorithm
In the defect for being easily trapped into local minimum, and by optimization input parameter value, reduce the training calculation amount of model, mention
The high speed of operation, improves the precision of prediction.Company and other number can preferably be carried out through the invention it was predicted that
Better help can be provided for user, also improve user experience.
Detailed description of the invention
Fig. 1 shows a kind of management data prediction technique flow chart neural network based of the present invention;
Fig. 2 shows the method flow diagrams of model training of the present invention;
Fig. 3 shows the data prognostic chart of the embodiment of the present invention.
Specific embodiment
In order to more fully hereinafter understand the features of the present invention and technology contents, with reference to the accompanying drawing to reality of the invention
It is now described in detail, appended attached drawing purposes of discussion only for reference, is not used to limit the present invention.
BP (back propagation) neural network is 1986 by the science headed by Rumelhart and McClelland
The concept that family proposes is a kind of multilayer feedforward neural network according to the training of error backpropagation algorithm, is most widely used at present
General neural network.
Basic BP algorithm includes two processes of backpropagation of the propagated forward and error of signal.When calculating error output
It is carried out by from the direction for being input to output, and adjusts weight and threshold value and then carried out from the direction for being output to input.When forward-propagating,
Input signal acts on output node by hidden layer, by nonlinear transformation, generates output signal, if reality output and expectation
Output is not consistent, then is transferred to the back-propagation process of error.Error-duration model be by output error by hidden layer to input layer by
Layer anti-pass, and give error distribution to all units of each layer, from the error signal that each layer obtains as adjustment each unit weight
Foundation.By adjusting the linking intensity and hidden node of input node and hidden node and the linking intensity and threshold of output node
Value declines error along gradient direction, by repetition learning training, determines network parameter (weight corresponding with minimal error
And threshold value), training stops stopping.Trained neural network can voluntarily be handled the input information of similar sample at this time
The smallest information by non-linear conversion of output error.BP algorithm is exactly to be easily trapped into local minimum there are a defect.
BP network is that increase several layers (one or more layers) neuron, these neurons claim between input layer and output layer
For hidden unit, they are not contacted directly with the external world, but the change of its state, then can influence the pass between inputting and exporting
System, each layer can have several nodes.The calculating process of BP neural network is by positive calculating process and retrospectively calculate process group
At.Forward-propagating process, input pattern is successively handled from input layer through hidden unit layer, and turns to output layer, per~layer neuron
State only under the influence of one layer of neuron state.If desired output cannot be obtained in output layer, it is transferred to backpropagation,
Error signal is returned along original connecting path, by modifying the weight of each neuron, so that error signal is minimum.
Fig. 1 shows a kind of management data prediction technique flow chart neural network based of the present invention.
As shown in Figure 1, the invention discloses a kind of management data prediction technique neural network based, including following step
It is rapid:
Obtain the historical data of target object;
It is calculated according to the historical data, obtains optimization input parameter value;
The parameter of neural network is initialized, and model training is carried out according to the optimization input parameter value, after being trained
Model;
In model after inputting parameter to training to be measured, prediction output valve is obtained.
It should be noted that the historical data for obtaining target object can be to be inputted manually by user, Huo Zhetong
The cloud computing for crossing background server obtains, or carries out automatic search by network and obtain.Obtain historical data purpose be into
The training of row model.Target object therein is the various management datas needed, can be with the exchange rate, staple commodities price expectation, GDP
The data such as prediction.Historical data amount therein is The more the better, and the quantity of sample is more, and obtained training pattern will be more quasi-
Really, the variance of the prediction data obtained by model can be closer to actual value.It is searched for automatically according to cloud computing or network
It obtains, then eliminates the manual operation of user, improve the experience sense of user.
Historical data therein can be, for example, obtaining the various data of company management, net assets income ratio, net profit
The data such as growth rate, main business income growth rate, rate of gross profit, debt ratio, cash flow, and obtain related to historical data
Parameter.The parameters such as net assets income ratio, net profit growth rate, main business income growth rate as escribed above.
According to embodiments of the present invention, described to be calculated according to the historical data, obtain optimization input parameter value step
Specifically:
Obtain historical data values and n parameter;
Calculate the variance contribution ratio of the n parameter;
N variance contribution ratio of the calculating is arranged from big to small;
M variance contribution ratio in the top is subjected to summation addition, obtain variance contribution ratio and judges the variance tribute
It offers rate and whether is greater than optimization threshold value W;
If it is greater than or equal to the optimization threshold value W, then choosing the m variance is input parameter value;
Wherein n is the natural number more than or equal to 2, and m is the natural number less than or equal to n.
It should be noted that the optimization threshold value W is 85-100%.Generally take the contribution rate of m principal component not less than certain
A optimization threshold value W, when this optimization threshold value W is set greater than equal to 85% or more, after taking these m principal component to be used as optimization
Parameter, can so that calculate data volume become smaller, to increase computational efficiency.Preferably, the optimization threshold value W is
95%.
According to embodiments of the present invention, it is described initialization neural network parameter, and according to the optimization input parameter value into
Row model training, the model after being trained specifically:
Initialize training parameter;
The position of each particle is mapped in BP neural network;
The training BP neural network, calculates the fitness function value of each particle;
It is iterated calculating, more new particle individual optimal value;
Population Regeneration global optimum updates particle position and speed, according to individual optimal and global optimum to particle
Position and speed is adjusted;
Judge whether to meet termination condition, be returned if being unsatisfactory for, continues to train;Repetitive exercise is saved if meeting
As a result;
Every dimension data of training result is mapped in BP neural network;
BP neural network model after being trained.
Wherein, when training pattern, particle more new formula are as follows:
Wherein, w, c1、c2For weight, particle search can be accelerated to individual optimum position information and global optimum position
Information, rand are the random number of (0,1).
It should be noted that the training parameter include population invariable number, the number of iterations, population particle scale and its position and
One or more of speed, speed value range, weight, the weight are w, c1、c2.The training parameter is user
Oneself is set.Training parameter can also be set by network.It is set by network specifically: obtain other use
Family setting or prediction model, compare the similarity of the model used with the data to be predicted, by the model of similarity in front
Training parameter be weighted and averaged operation, obtain new training parameter value, the new training parameter value is pre- as this
The numerical value of survey.
According to embodiments of the present invention, the fitness function value for calculating each particle specifically:
It is calculated by the functional value that following formula carries out particle fitness:
Wherein, wnjkIt is n-th of output valve of network, tnIt is the actual value of n training sample, s2 is output layer neuron
Number.
According to embodiments of the present invention, described to be iterated calculating, more new particle individual optimal value specifically: more each grain
The current calculating fitness function value of son, if the fitness value of current iteration is better than the fitness function value of preceding an iteration,
Then using current individual values as new optimal value.
According to embodiments of the present invention, the Population Regeneration global optimum specifically: the fitness value of population is calculated, if
The fitness value of current population global optimum is better than the fitness value of last iteration, and the global optimum of current iteration is selected to make
For new global optimum.
Technical solution of the present invention is further illustrated below by several specific embodiments.
Embodiment one
The principle and step of optimization of parameter choice method are illustrated below by the present embodiment.
If X is a data matrix, every a line represents a data value, and each column represent a target variable, wherein becoming
Measure X1,X2,…XpThere are p, then the linear combination of variable can indicate are as follows:
It can then write a Chinese character in simplified form are as follows:
Yi=a1iX1+a2iX2+…+apiXp, i=1,2 ... p
Wherein, it is necessary to meet following condition:
(1)YiAnd YjFor nonlinear correlation, wherein i ≠ j;I, j=1,2 ... p
(2)YiAnd Yi+1Variance be greater than Yi+1With Yi+2Variance, wherein i=1,2 ... p-2
(3)
The population variance of main variables is equal to the population variance of original variable, i-th of principal component Y in population varianceiVariance institute
The ratio accounted for is principal component YiContribution rate.The sum of contribution rate of m principal component is the accumulation contribution rate of m principal component, invention
People's discovery generally takes the contribution rate of m principal component not less than some optimization threshold value W, this optimization threshold value W, which is set greater than, to be equal to
When 85% or more, take these m principal component as optimization after parameter, can so that calculate data volume become smaller, from
And increase computational efficiency.
The step of finding parameter optimization is as follows:
Obtain historical data values and n parameter;
Calculate the variance contribution ratio of the n parameter;
N variance contribution ratio of the calculating is arranged from big to small;
M variance contribution ratio in the top is subjected to summation addition, obtain variance contribution ratio and judges the variance tribute
It offers rate and whether is greater than optimization threshold value W;
If it is greater than or equal to the optimization threshold value W, then choosing the m variance is input parameter value.
Wherein m is less than or equal to n.The optimization threshold value W is 85-100%, it is preferred that the optimization threshold value is 90%.
For example, when carrying out the prediction of GDP and CPI.According to previous algorithm, it may be necessary to excessive input parameter value ginseng
With calculating, design parameter value are as follows: industrial added value, M2 index, raw material, foreign exchange reserve, total import and export value, financial revenue and expenditure etc. ten
Several variable parameters, and these variable parameters are also the output variable parameter after prediction.But inventor passes through parameter optimization point
It is found after analysis, input parameter value was industrial added value, M2 index, raw material, foreign exchange reserve, total import and export value this 5 at that time
When, contribution rate to 91.3%.The 85% of optimization threshold value W is had been above, then illustrating can be complete using this 5
Meet the standard of prediction model.
It should be noted that obtaining historical data is the related data before obtaining, for example, obtaining the various of company management
Data, the data such as net assets income ratio, net profit growth rate, main business income growth rate, rate of gross profit, debt ratio, cash flow,
And obtain the relevant parameter with historical data.Net assets income ratio, net profit growth rate, main business are received as described above
Enter the parameters such as growth rate.Wherein, the historical data can for user input, or by network carry out crawl or
Calculating acquisition is carried out by network cloud.
According to embodiments of the present invention, after obtaining above-mentioned data, then the variance contribution of the n parameter is calculated
Rate.Variance is the measurement of the dispersion degree when probability theory and statistical variance measure stochastic variable or one group of data.Side in probability theory
Difference is used to measure the departure degree between stochastic variable and its mathematic expectaion (i.e. mean value).Variance (sample variance) in statistics is
The average of the square value of the difference of the average of each sample value and all sample values.Carry out the calculating of variance and variance contribution ratio
It is the general calculation method of this field, the present invention is not repeated one by one again.
According to embodiments of the present invention, after the variance contribution ratio for calculating the n parameter, by n side of the calculating
Poor contribution rate is arranged from big to small.For example, n=7, that is, have 7 parameters, its contribution degree obtained after calculating contribution rate
Are as follows: parameter 1:20%;Parameter 2:5%;Parameter 3:10%;Parameter 4:3%;Parameter 5:15%;Parameter 6:17%;Parameter 7:30.
It can be obtained after then being arranged from big to small: parameter 7, parameter 1, parameter 6, parameter 5, parameter 3, parameter 2, parameter 4.
According to embodiments of the present invention, m variance contribution ratio in the top is subjected to summation addition, obtains variance contribution ratio
With, judge the variance contribution ratio and whether be greater than optimization threshold value W.For example, taking optimization threshold value W is 90%, referring to above-mentioned steps
As a result, after the contribution rates of 5 parameters in the top is added 92%, be more than optimize threshold value 90%.
M variance contribution ratio in the top is subjected to summation addition, obtain variance contribution ratio and judges the variance tribute
It offers rate and whether is greater than optimization threshold value W;If it is greater than or equal to the optimization threshold value W, then the m variance is chosen as input parameter
Value.Since the first 5 parameter contribution rates that above-mentioned steps are chosen just have reached 92%, 90% optimization threshold value W is had been over,
So the m chosen is 5, specific Optimal Parameters are as follows: parameter 7, parameter 1, parameter 6, parameter 5, parameter 3.By the parameter after optimization
As the input variable parameter of training pattern, the training of model is carried out.
Embodiment two
For the present embodiment by one group of initial solution in solution space parallel search, solution concentrates competition and association between individual and individual
Make the optimizing of realization disaggregation.
Lower mask body carries out the explanation of optimal resolving Algorithm.
In this algorithm, as soon as each particle i is a potential vector solution, several particles constitute a solution set.
Each particle finds optimal solution by fitness function.When particle scans in an a dimension space, each particle
State is all defined as:
Velocity vector Vi,
Position vector Xi,
Particle i is by individual optimum position information PbestIt is stored inBy global optimum position
Information GbestIt is stored inIn.
Particle more new formula are as follows:
Wherein, w, c1、c2For weight, particle search can be accelerated to individual optimum position information and global optimum position
Information, rand are the random number of (0,1).
However, the particle position in optimal resolving Algorithm is an a dimensional vector, by the parameter group to be optimized in BP neural network
At.The value of a is equal to the number of parameters of optimization.It is obtained most in fact, process of the process of optimal resolving Algorithm through network is exactly one
The process of excellent parameter.In each iterative calculation of optimal resolving Algorithm, the position of each particle is mapped to BP neural network
In, obtained training error is used to calculate the fitness value of the particle.The calculation formula of fitness value is as follows:
Wherein, wnjkIt is n-th of output valve of network, tnIt is the actual value of n training sample, s2 is output layer neuron
Number.BP neural network has input layer, hidden layer, output layer, and it is in BP neural network that s2, which is then output layer neuron,
Output layer neuron number.
Fig. 2 shows the method flow diagrams of model training of the present invention.
As shown in Fig. 2, the specific steps of which are as follows:
Initialize training parameter.Wherein the training parameter include population invariable number, the number of iterations, population particle position and
Speed, speed value range.
The position of each particle is mapped in BP neural network.
The training BP neural network, calculates the fitness function value of each particle.The fitness function value calculates such as
Under:
Wherein, wnjkIt is n-th of output valve of network, tnIt is the actual value of n training sample, s2 is output layer neuron
Number.
It is iterated calculating, more new particle individual optimal value.It is described to be iterated calculating, more new particle individual optimal value tool
Body are as follows: to each particle, if the fitness value of current iteration is better than the fitness value of preceding an iteration, by current individual
It is worth as newly optimal.
Population Regeneration global optimum.The Population Regeneration global optimum specifically: if current global optimum is suitable
Angle value is answered to be better than the fitness value of last iteration, population selects the global optimum of current iteration as new global optimum
Value.
Update particle position and speed.The position and speed of particle is adjusted with global optimum according to individual is optimal.
Judge whether to meet termination condition.It is returned if being unsatisfactory for, continues to train;Repetitive exercise is saved if meeting
As a result, terminating training.
Every dimension of training result is mapped in BP neural network.Specific training pattern has just been obtained by this step,
The prediction of data can be carried out by this model.
Input test data are predicted.This step is to need data to be tested into trained model by input, is obtained
To the numerical value of prediction.
Embodiment three
Net assets income ratio, net profit growth rate, main business income growth rate, rate of gross profit, debt ratio, cash flow, member
The data such as work average salary are respectively parameter 1-7.N=7 has 7 parameters, its tribute is obtained after calculating contribution rate
Degree of offering are as follows: parameter 1:20%;Parameter 2:5%;Parameter 3:10%;Parameter 4:3%;Parameter 5:15%;Parameter 6:17%;Parameter 7:
30.It can be obtained after then being arranged from big to small: parameter 7, parameter 1, parameter 6, parameter 5, parameter 3, parameter 2, parameter 4.
According to embodiments of the present invention, m variance contribution ratio in the top is subjected to summation addition, obtains variance contribution ratio
With, judge the variance contribution ratio and whether be greater than optimization threshold value W.For example, taking optimization threshold value W is 90%, referring to above-mentioned steps
As a result, after the contribution rates of 5 parameters in the top is added 92%, be more than optimize threshold value 90%.
M variance contribution ratio in the top is subjected to summation addition, obtain variance contribution ratio and judges the variance tribute
It offers rate and whether is greater than optimization threshold value W;If it is greater than or equal to the optimization threshold value W, then the m variance is chosen as input parameter
Value.Since the first 5 parameter contribution rates that above-mentioned steps are chosen just have reached 92%, 90% optimization threshold value W is had been over,
So the m chosen is 5, specific Optimal Parameters are as follows: parameter 7, parameter 1, parameter 6, parameter 5, parameter 3.By the parameter after optimization
As the input variable parameter of training pattern, the training of model is carried out.
The initial value of BP neural network, i.e. training parameter is arranged: initial learning rate is 0.015;The number of iterations is 1000, repeatedly
It is 0.0001 for target;Particle populations scale is 64;Population the number of iterations is 40;Weight w=0.9;c1And c2It is 2.It will be on 37 groups
The historical data values for stating 5 parameters carry out model training, obtain trained model.
By net assets income ratio, net profit growth rate, main business income growth rate, rate of gross profit, debt ratio, cash flow,
This 7 data of employee's average salary are input in trained model, have just obtained the prediction output valve of model.Fig. 3 is shown
The data prognostic chart of the embodiment of the present invention can be seen that by Fig. 3 through the prediction data value and truth after model training
Data value.Curve the upper surface of in figure is the data value of prediction, and curve below is actual data value.As can be seen that real
The data value on border and the data value by predicting are very close, and variance has reached expected standard already close to minimum.It is logical
The forecast analysis for crossing this when using 5 Optimal Parameters, saves 40% when calculating the time than calculating 7 parameters, so
The speed for improving calculating, improves computational efficiency.
Example IV
The training parameter includes population invariable number, the number of iterations, population particle scale and its position and speed, speed value
One or more of range, weight, the weight are w, c1、c2.The training parameter is that user oneself sets.
Training parameter can also be set by network.It is set by network specifically: obtain other users setting or pre-
Survey model, compare the similarity of the model used with the data to be predicted, by the training parameter of the model of similarity in front into
Row weighted mean operation obtains new training parameter value, the numerical value that the new training parameter value is predicted as this.
Wherein server obtains other users setting or prediction model.What server obtained can be other users at it
The preceding data predicted or model are also possible to the optimal prediction model that the operation before system passes through obtains.?
That is server obtain be before used model either the used data of user and model.Then compare with
The similarity for the model that the data to be predicted use, comparing similarity is model similar with the model of this prediction in order to obtain
Group is conducive to the use of this prediction in this way.It is calculated again, compares the similarity of the model used with the data to be predicted,
The training parameter of the model of similarity in front is weighted and averaged operation.This step reduces number using weighted average calculation
According to Statistical Complexity, can also preferably complete the selection of model parameter.Finally, new training parameter value is obtained, it will be described
The numerical value that new training parameter value is predicted as this.
Training parameter value is obtained by network, can preferably complete the training of BP neural network model, and can obtain
To preferable training pattern.
Scheme through the invention applies also for stock price, stock index, the exchange rate, staple commodities price expectation, enterprise
The early warning of industry financial situation, traffic passenger flow estimation etc..
It should be noted that not believing that is be illustrated is the common technology hand of those skilled in the art in the present invention
Section, therefore the present invention is no longer repeated one by one.
The present invention also provides a kind of computer readable storage medium, base is stored on the computer readable storage medium
In the management data prediction technique program of neural network, realized such as when the management data prediction technique program is executed by processor
The step of above-mentioned management data prediction technique.
It should be noted that computer readable storage medium can use in mobile phone terminal, it can also be in the server
It uses.
The present invention also provides a kind of management data forecasting system, the system comprises: it memory, processor and is stored in
On the memory and the program of management data prediction technique that can run on the processor, the management data prediction side
It realizes when the program of method is executed by the processor such as the step of above-mentioned management data prediction technique.
It should be noted that this system can be the server system of computer, by being stored on the memory simultaneously
The program of the program for the management data prediction technique that can be run on the processor, the management data prediction technique is described
It realizes when processor executes such as the step of above-mentioned management data prediction technique.It can be to mobile phone or calculating by the result of prediction
Machine carries out push and shows.
According to the technical solution of the present invention, it solves in BP neural network algorithm and is easily trapped into lacking for local minimum
It falls into, and by optimization input parameter value, reduces the training calculation amount of model, improve the speed of operation, improve prediction
Precision.Company and other number can be preferably carried out through the invention it was predicted that better help can be provided for user,
Also improve user experience.
In several embodiments provided herein, it should be understood that disclosed device and method can pass through it
Its mode is realized.Apparatus embodiments described above are merely indicative, for example, the division of the unit, only
A kind of logical function partition, there may be another division manner in actual implementation, such as: multiple units or components can combine, or
It is desirably integrated into another system, or some features can be ignored or not executed.In addition, shown or discussed each composition portion
Mutual coupling or direct-coupling or communication connection is divided to can be through some interfaces, the INDIRECT COUPLING of equipment or unit
Or communication connection, it can be electrical, mechanical or other forms.
Above-mentioned unit as illustrated by the separation member, which can be or may not be, to be physically separated, aobvious as unit
The component shown can be or may not be physical unit;Both it can be located in one place, and may be distributed over multiple network lists
In member;Some or all of units can be selected to achieve the purpose of the solution of this embodiment according to the actual needs.
In addition, each functional unit in various embodiments of the present invention can be fully integrated in one processing unit, it can also
To be each unit individually as a unit, can also be integrated in one unit with two or more units;It is above-mentioned
Integrated unit both can take the form of hardware realization, can also realize in the form of hardware adds SFU software functional unit.
Those of ordinary skill in the art will appreciate that: realize that all or part of the steps of above method embodiment can pass through
The relevant hardware of program instruction is completed, and program above-mentioned can store in computer-readable storage medium, which exists
When execution, step including the steps of the foregoing method embodiments is executed;And storage medium above-mentioned includes: movable storage device, read-only deposits
Reservoir (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or
The various media that can store program code such as CD.
If alternatively, the above-mentioned integrated unit of the present invention is realized in the form of software function module and as independent product
When selling or using, it also can store in a computer readable storage medium.Based on this understanding, the present invention is implemented
Substantially the part that contributes to existing technology can be embodied in the form of software products the technical solution of example in other words,
The computer software product is stored in a storage medium, including some instructions are used so that computer equipment (can be with
It is personal computer, server or network equipment etc.) execute all or part of each embodiment the method for the present invention.
And storage medium above-mentioned includes: that movable storage device, ROM, RAM, magnetic or disk etc. are various can store program code
Medium.
The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any
Those familiar with the art in the technical scope disclosed by the present invention, can easily think of the change or the replacement, and should all contain
Lid is within protection scope of the present invention.Therefore, protection scope of the present invention should be based on the protection scope of the described claims.
Claims (10)
1. a kind of management data prediction technique neural network based characterized by comprising
Obtain the historical data of target object;
It is calculated according to the historical data, obtains optimization input parameter value;
The parameter of neural network is initialized, and model training is carried out according to the optimization input parameter value, the mould after being trained
Type;
In model after inputting parameter to training to be measured, prediction output valve is obtained.
2. a kind of management data prediction technique neural network based according to claim 1, which is characterized in that described
It is calculated according to the historical data, obtains optimization input parameter value step specifically:
Obtain historical data values and n parameter;
Calculate the variance contribution ratio of the n parameter;
N variance contribution ratio of the calculating is arranged from big to small;
M variance contribution ratio in the top is subjected to summation addition, obtain variance contribution ratio and judges the variance contribution ratio
Whether optimization threshold value W is greater than;
If it is greater than or equal to the optimization threshold value W, then choosing the m variance is input parameter value;
Wherein n is the natural number more than or equal to 2, and m is the natural number less than or equal to n.
3. a kind of management data prediction technique neural network based according to claim 2, which is characterized in that described excellent
Change threshold value W is 85-100%.
4. a kind of management data prediction technique neural network based according to claim 1, which is characterized in that described first
The parameter of beginningization neural network, and model training is carried out according to the optimization input parameter value, the model after being trained is specific
Are as follows:
Initialize training parameter;
The position of each particle is mapped in BP neural network;
The training BP neural network, calculates the fitness function value of each particle;
It is iterated calculating, more new particle individual optimal value;
Population Regeneration global optimum updates particle position and speed, according to individual optimal and global optimum to the position of particle
It is adjusted with speed;
Judge whether to meet termination condition, be returned if being unsatisfactory for, continues to train;Repetitive exercise knot is saved if meeting
Fruit;
Every dimension data of training result is mapped in BP neural network;
BP neural network model after being trained.
5. a kind of management data prediction technique neural network based according to claim 4, which is characterized in that the instruction
Practicing parameter includes population invariable number, the number of iterations, population particle scale and its position and speed, speed value range, in weight
It is one or more of.
6. a kind of management data prediction technique neural network based according to claim 4, which is characterized in that the meter
Calculate the fitness function value of each particle specifically:
It is calculated by the functional value that following formula carries out particle fitness:
Wherein, wnjkIt is n-th of output valve of network, tnIt is the actual value of n training sample, s2 is of output layer neuron
Number.
7. a kind of management data prediction technique neural network based according to claim 4, which is characterized in that it is described into
Row iteration calculates, more new particle individual optimal value specifically: the current calculating fitness function value of more each particle, if worked as
The fitness value of preceding iteration is better than the fitness function value of preceding an iteration, then using current individual values as new optimal value.
8. a kind of management data prediction technique neural network based according to claim 4, which is characterized in that it is described more
New population global optimum specifically: the fitness value of population is calculated, if the fitness value of current population global optimum is better than
The fitness value of last iteration selects the global optimum of current iteration as new global optimum.
9. a kind of computer readable storage medium, which is characterized in that be stored on the computer readable storage medium based on mind
Management data prediction technique program through network realizes such as right when the management data prediction technique program is executed by processor
It is required that described in any one of 1 to 8 the step of management data prediction technique.
10. a kind of management data forecasting system, which is characterized in that the system comprises: it memory, processor and is stored in described
On memory and the program of management data prediction technique that can run on the processor, the management data prediction technique
It realizes when program is executed by the processor such as the step of management data prediction technique described in any item of the claim 1 to 8.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810993073.3A CN109118013A (en) | 2018-08-29 | 2018-08-29 | A kind of management data prediction technique, readable storage medium storing program for executing and forecasting system neural network based |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810993073.3A CN109118013A (en) | 2018-08-29 | 2018-08-29 | A kind of management data prediction technique, readable storage medium storing program for executing and forecasting system neural network based |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109118013A true CN109118013A (en) | 2019-01-01 |
Family
ID=64861114
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810993073.3A Pending CN109118013A (en) | 2018-08-29 | 2018-08-29 | A kind of management data prediction technique, readable storage medium storing program for executing and forecasting system neural network based |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109118013A (en) |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110084493A (en) * | 2019-04-11 | 2019-08-02 | 企家有道网络技术(北京)有限公司 | Enterprise diagnosis, prediction technique and device, server based on artificial intelligence |
CN110083750A (en) * | 2019-03-15 | 2019-08-02 | 平安科技(深圳)有限公司 | Blacklist screening method, device, computer equipment and storage medium |
CN110460880A (en) * | 2019-08-09 | 2019-11-15 | 东北大学 | Wireless industrial streaming media self-adapting transmission method based on population and neural network |
CN110782339A (en) * | 2019-10-22 | 2020-02-11 | 黑龙江工业学院 | Default probability prediction method, system and readable storage medium |
CN111126676A (en) * | 2019-12-05 | 2020-05-08 | 北京明略软件系统有限公司 | Method, device and equipment for predicting company operation risk |
CN111240282A (en) * | 2019-12-31 | 2020-06-05 | 联想(北京)有限公司 | Process optimization method, device, equipment and computer readable storage medium |
CN111414922A (en) * | 2019-01-07 | 2020-07-14 | 阿里巴巴集团控股有限公司 | Feature extraction method, image processing method, model training method and device |
CN111417129A (en) * | 2019-01-07 | 2020-07-14 | 中国移动通信有限公司研究院 | Configuration method and equipment of multi-frequency and multi-mode network |
CN111861050A (en) * | 2019-04-25 | 2020-10-30 | 鸿富锦精密电子(天津)有限公司 | Apparatus and method for managing and controlling manufacturing process and computer readable storage medium |
CN112801708A (en) * | 2021-02-05 | 2021-05-14 | 通联数据股份公司 | Business income prediction model determination method and device, and prediction method and device |
CN113658689A (en) * | 2021-08-25 | 2021-11-16 | 深圳前海微众银行股份有限公司 | Multi-agent model training method and device, electronic equipment and storage medium |
CN114792258A (en) * | 2022-06-27 | 2022-07-26 | 北京京东振世信息技术有限公司 | Information generation method and device, electronic equipment and computer readable medium |
CN115526266A (en) * | 2022-10-18 | 2022-12-27 | 支付宝(杭州)信息技术有限公司 | Model training method and device, and business prediction method and device |
CN115880102A (en) * | 2023-03-08 | 2023-03-31 | 国网福建省电力有限公司 | Electric energy metering method, system, equipment and storage medium |
CN116383670A (en) * | 2023-03-18 | 2023-07-04 | 宝钢工程技术集团有限公司 | Operation equipment coding similarity association method based on factory object position |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107818386A (en) * | 2017-11-28 | 2018-03-20 | 国网湖南省电力有限公司 | Power grid enterprises' Operating profit Forecasting Methodology |
-
2018
- 2018-08-29 CN CN201810993073.3A patent/CN109118013A/en active Pending
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107818386A (en) * | 2017-11-28 | 2018-03-20 | 国网湖南省电力有限公司 | Power grid enterprises' Operating profit Forecasting Methodology |
Non-Patent Citations (1)
Title |
---|
王永杰 等: "基于主成分分析的PSO-BP算法在GDP和CPI预测中的应用", 《重庆理工大学学报(自然科学)》 * |
Cited By (20)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111414922B (en) * | 2019-01-07 | 2022-11-15 | 阿里巴巴集团控股有限公司 | Feature extraction method, image processing method, model training method and device |
CN111414922A (en) * | 2019-01-07 | 2020-07-14 | 阿里巴巴集团控股有限公司 | Feature extraction method, image processing method, model training method and device |
CN111417129A (en) * | 2019-01-07 | 2020-07-14 | 中国移动通信有限公司研究院 | Configuration method and equipment of multi-frequency and multi-mode network |
CN110083750A (en) * | 2019-03-15 | 2019-08-02 | 平安科技(深圳)有限公司 | Blacklist screening method, device, computer equipment and storage medium |
CN110084493A (en) * | 2019-04-11 | 2019-08-02 | 企家有道网络技术(北京)有限公司 | Enterprise diagnosis, prediction technique and device, server based on artificial intelligence |
CN111861050A (en) * | 2019-04-25 | 2020-10-30 | 鸿富锦精密电子(天津)有限公司 | Apparatus and method for managing and controlling manufacturing process and computer readable storage medium |
CN111861050B (en) * | 2019-04-25 | 2024-02-20 | 富联精密电子(天津)有限公司 | Production process control device, method and computer readable storage medium |
CN110460880A (en) * | 2019-08-09 | 2019-11-15 | 东北大学 | Wireless industrial streaming media self-adapting transmission method based on population and neural network |
CN110782339A (en) * | 2019-10-22 | 2020-02-11 | 黑龙江工业学院 | Default probability prediction method, system and readable storage medium |
CN111126676A (en) * | 2019-12-05 | 2020-05-08 | 北京明略软件系统有限公司 | Method, device and equipment for predicting company operation risk |
CN111240282A (en) * | 2019-12-31 | 2020-06-05 | 联想(北京)有限公司 | Process optimization method, device, equipment and computer readable storage medium |
CN112801708A (en) * | 2021-02-05 | 2021-05-14 | 通联数据股份公司 | Business income prediction model determination method and device, and prediction method and device |
CN112801708B (en) * | 2021-02-05 | 2024-01-16 | 通联数据股份公司 | Business income prediction model determining method and device, and business income prediction model determining method and device |
CN113658689A (en) * | 2021-08-25 | 2021-11-16 | 深圳前海微众银行股份有限公司 | Multi-agent model training method and device, electronic equipment and storage medium |
CN114792258A (en) * | 2022-06-27 | 2022-07-26 | 北京京东振世信息技术有限公司 | Information generation method and device, electronic equipment and computer readable medium |
CN115526266B (en) * | 2022-10-18 | 2023-08-29 | 支付宝(杭州)信息技术有限公司 | Model Training Method and Device, Service Prediction Method and Device |
CN115526266A (en) * | 2022-10-18 | 2022-12-27 | 支付宝(杭州)信息技术有限公司 | Model training method and device, and business prediction method and device |
CN115880102A (en) * | 2023-03-08 | 2023-03-31 | 国网福建省电力有限公司 | Electric energy metering method, system, equipment and storage medium |
CN116383670A (en) * | 2023-03-18 | 2023-07-04 | 宝钢工程技术集团有限公司 | Operation equipment coding similarity association method based on factory object position |
CN116383670B (en) * | 2023-03-18 | 2024-04-19 | 宝钢工程技术集团有限公司 | Operation equipment coding similarity association method based on factory object position |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109118013A (en) | A kind of management data prediction technique, readable storage medium storing program for executing and forecasting system neural network based | |
Li et al. | Risk prediction in financial management of listed companies based on optimized BP neural network under digital economy | |
Dash | Performance analysis of a higher order neural network with an improved shuffled frog leaping algorithm for currency exchange rate prediction | |
CN106886846A (en) | A kind of bank outlets' excess reserve Forecasting Methodology that Recognition with Recurrent Neural Network is remembered based on shot and long term | |
Fisichella et al. | Can deep learning improve technical analysis of forex data to predict future price movements? | |
Dash | An improved shuffled frog leaping algorithm based evolutionary framework for currency exchange rate prediction | |
CN110059852A (en) | A kind of stock yield prediction technique based on improvement random forests algorithm | |
Stepashko et al. | Construction and research of the generalized iterative GMDH algorithm with active neurons | |
Li et al. | Heterogeneous ensemble learning with feature engineering for default prediction in peer-to-peer lending in China | |
Azadeh et al. | An integrated artificial neural network fuzzy C-means-normalization algorithm for performance assessment of decision-making units: The cases of auto industry and power plant | |
CN109242558A (en) | A kind of management data prediction technique, readable storage medium storing program for executing and forecasting system neural network based | |
Pereira | Genetic algorithm optimisation for finance and investments | |
Yang et al. | Prediction of equipment performance index based on improved chaotic lion swarm optimization–LSTM | |
Dash et al. | Utility of a Shuffled Differential Evolution algorithm in designing of a Pi-Sigma Neural Network based predictor model | |
CN108459997A (en) | High skewness data value probability forecasting method based on deep learning and neural network | |
Sharma et al. | Intelligent portfolio asset prediction enabled by hybrid Jaya-based spotted hyena optimization algorithm | |
Caraka | Prediction of Euro 50 Using Back Propagation Neural Network (BPNN) and Genetic Algorithm (GA) | |
Ly et al. | Applying deep learning to better predict cryptocurrency trends | |
CN110991600B (en) | Drought intelligent prediction method integrating distribution estimation algorithm and extreme learning machine | |
Faghri et al. | Artificial neural network–based approach to modeling trip production | |
Samarawickrama et al. | Multi-step-ahead prediction of exchange rates using artificial neural networks: a study on selected sri lankan foreign exchange rates | |
Chirita et al. | Usefulness of artificial neural networks for predicting financial and economic crisis | |
Li et al. | Prediction and Sensitivity Analysis of Companies’ Return on Equity Based on Deep Neural Network | |
Patel et al. | Introduction to machine learning and its application | |
Baranes et al. | Predicting Earnings Directional Movement Utilizing Recurrent Neural Networks (RNN) |
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: 20190101 |
|
RJ01 | Rejection of invention patent application after publication |