CN108537663A - One B shareB trend forecasting method - Google Patents
One B shareB trend forecasting method Download PDFInfo
- Publication number
- CN108537663A CN108537663A CN201810233670.6A CN201810233670A CN108537663A CN 108537663 A CN108537663 A CN 108537663A CN 201810233670 A CN201810233670 A CN 201810233670A CN 108537663 A CN108537663 A CN 108537663A
- Authority
- CN
- China
- Prior art keywords
- weights
- chromosome
- hidden
- output
- value
- 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
- G06N3/12—Computing arrangements based on biological models using genetic models
- G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
- G06Q40/04—Trading; Exchange, e.g. stocks, commodities, derivatives or currency exchange
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Strategic Management (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biophysics (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- Finance (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- Accounting & Taxation (AREA)
- Human Resources & Organizations (AREA)
- Evolutionary Biology (AREA)
- Development Economics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Genetics & Genomics (AREA)
- Data Mining & Analysis (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Game Theory and Decision Science (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Entrepreneurship & Innovation (AREA)
- Computational Linguistics (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Physiology (AREA)
- Technology Law (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a B shareB trend forecasting methods, including step:It is first the gene in genetic algorithm in population on every chromosome by the threshold map of the input weights of extreme learning machine and hidden node, using the ability of searching optimum of genetic algorithm, selects optimal chromosome and constitute elite group；The local search ability of particle cluster algorithm is recycled to select optimal chromosome, as an optimization the input weights and threshold value of rear extreme learning machine；The output weights for calculating extreme learning machine hidden neuron with least square method again, to calculate predicted value, and are traded according to predicted value.Present invention is generally directed to conventional limit learning machines when handling stock trend prediction problem, and the not high problem of precision of prediction, prediction result Billy are more accurate with conventional method.
Description
Technical field
The present invention relates to a B shareB trend forecasting methods, and in particular to improves pole using genetic algorithm and particle cluster algorithm
Limit learning machine applies to stock trend prediction, belongs to finance data mining field.
Background technology
Stock trend prediction is always the emphasis of financial field research, and a large amount of experimental study shows that stock market has
The characteristics of non-linear, non-stationary, strong noise, stock market are a nonlinear systems, and traditional linear model has been unsatisfactory for
Market survey.
Currently, there are many achievement in research of stock trend prediction, prediction technique mainly has neural network, support vector machines, mixes
Ignorant theory, fractal theory etc., but neural network prediction training speed is slow, are easily trapped into local optimum, and the number of hidden nodes by
Artificial setting leverages precision of prediction；Extreme learning machine is a kind of new neural network algorithm, is single hidden layer feed forward neural
Network.The fields such as extreme learning machine has been widely used for fault diagnosis at present, image segmentation, data mining automatically control；
In the algorithm, input weight randomly selects, and output weight is determined by least square method, greatly improves network training speed
And generalization ability, still, when solving the problems, such as that gradient declines, due to the randomness of hidden neuron parameter, extreme learning machine has
Easily occur that convergence rate is slow, is absorbed in the problems such as local optimum.
In view of the above-mentioned problems, being improved herein using genetic algorithm and particle cluster algorithm on the basis of extreme learning machine
Thus into stock trend prediction model is established；It is genetic algorithm by the weights of extreme learning machine and threshold map in the model
Chromosome vector, optimal individual is selected by the ability of searching optimum of genetic algorithm, constitutes upper layer elite group, then sharp again
Elite group is scanned for the local search ability of particle cluster algorithm, selects optimal chromosome, that is, selecting makes extreme learning machine
The input weights and threshold value of network error minimum improve precision of prediction, and subtract so as to improve the Generalization Capability of extreme learning machine
Small prediction result is influenced by the number of hidden nodes.
Invention content
In view of this, the present invention's learns the not high problem of precision in stock trend prediction mainly for conventional limit,
Purpose is to improve limit study prediction using the ability of searching optimum of genetic algorithm and the local search ability of particle cluster algorithm
Model improves the accuracy of stock trend prediction.
In order to achieve the above object, technical solution proposed by the present invention is：
One B shareB trend forecasting method, the appraisal procedure include the following steps:
Step 1, using source data constructing technology index as the input of model, stock price trend state is determined as output
Training set and test set；
Step 2 dyes the threshold map of the input weights of extreme learning machine and hidden node for every in Population in Genetic Algorithms
Gene on body takes the optimal chromosome in each subgroup to constitute elite group using the ability of searching optimum of genetic algorithm；
Step 3 selects optimal chromosome in elite group using the local search ability of particle cluster algorithm, as an optimization the rear limit
The input weights and threshold value of learning machine；
Step 4, according to the input weights and threshold value after optimization, utilize least square method to calculate extreme learning machine hidden neuron
Weights are exported, predicted value is then calculated, is traded according to the predicted value of model.
In conclusion B shareB trend forecasting method of the present invention, genetic algorithm and particle cluster algorithm are improved
Extreme learning machine algorithm apply to stock trend prediction, solve the problems, such as that conventional limit learning machine precision of prediction is low.First will
The input weights of extreme learning machine training data and the threshold map of hidden node is on every chromosomes in Population in Genetic Algorithms
Gene select optimal chromosome and constitute elite group using the ability of searching optimum of genetic algorithm；Then population is recycled
The local search ability of algorithm selects optimal chromosome, as an optimization the input weights and threshold value of rear extreme learning machine；Finally
The output weights for calculating extreme learning machine hidden neuron with least square method again, to calculate prediction output.The algorithm will be asked
Weights and Threshold are converted into the problem of finding optimal chromosome, take full advantage of the ability of searching optimum and grain of genetic algorithm
The local search ability of swarm optimization, and the strong learning ability of extreme learning machine is combined, good prediction effect can be reached.
Description of the drawings
Fig. 1 is the overall procedure schematic diagram of B shareB trend forecasting method of the present invention；
Fig. 2 is the flow diagram present invention determine that mode input output；
Fig. 3 is the flow diagram that the present invention chooses that the optimal chromosome in each subgroup constitutes elite group；
Fig. 4 is that the present invention chooses flow diagram of the optimal chromosome as input weights and threshold value；
Fig. 5 is the flow diagram that the present invention calculates predicted value；
Fig. 6 is model prediction accuracy rate comparison diagram in present example.
Below in conjunction with the attached drawing of the present invention, technical scheme of the present invention is clearly and completely described, it is clear that institute
It gives an actual example for illustrating, and non-limiting embodiments of the present invention, the present invention can also pass through other different specific realities
The mode of applying is implemented.The every other implementation that those of ordinary skill in the art are obtained without creative efforts
Example, shall fall within the protection scope of the present invention.
Fig. 1 is the overall procedure schematic diagram of B shareB trend forecasting method of the present invention, as shown in Figure 1, of the invention
The one B shareB trend forecasting method, includes the following steps:
Step 1, using source data constructing technology index as the input of model, stock price trend state is determined as output
Training set and test set；
Step 2 dyes the threshold map of the input weights of extreme learning machine and hidden node for every in Population in Genetic Algorithms
Gene on body takes the optimal chromosome in each subgroup to constitute elite group using the ability of searching optimum of genetic algorithm；
Step 3 selects optimal chromosome in elite group using the local search ability of particle cluster algorithm, as an optimization the rear limit
The input weights and threshold value of learning machine；
Step 4, according to the input weights and threshold value after optimization, utilize least square method to calculate extreme learning machine hidden neuron
Weights are exported, predicted value is then calculated, is traded according to the predicted value of model.
In short, B shareB trend forecasting method of the present invention is to weigh the input of extreme learning machine training data first
Value and the threshold map of hidden node are the gene in Population in Genetic Algorithms on every chromosome, are searched using the overall situation of genetic algorithm
Suo Nengli selects optimal chromosome and constitutes elite group；Then it is selected using the local search ability of particle cluster algorithm optimal
Chromosome, as an optimization after extreme learning machine input weights and threshold value；Its object is to solve the prediction of conventional limit learning machine
The relatively low problem of accuracy rate；The output weights for finally calculating extreme learning machine hidden neuron with least square method again, to count
Calculate prediction output.By regarding accuracy rate as model evaluation standard, made prediction to stock trend using the model.
Fig. 2 is the flow diagram present invention determine that mode input output, as shown in Fig. 2, step 1 determines the input of model
Output, includes the following steps：
Step 11 passes through feature letter according to the opening price of initial data, closing price, highest price, lowest price, exchange hand essential information
Number constructing technology index, including：Pass throughIt calculates discrete indicator, pass throughCalculating take advantage of a situation index, pass throughIt calculates momentum index, pass throughIt calculates Relative Strength Index, pass throughMeter
It calculates random index, pass throughIt calculates departure rate, pass throughIt calculates energy indexes, pass through
It calculates Larry William's indexs, pass throughIt calculates n days weighting sliding averages and fluctuates index,
In whereinT days highest prices, lowest price and closing price are indicated respectively；The first n days of t days is indicated respectively
Interior highest price and lowest price；,,；The t days amounts of increase for comparing proxima luce (prox. luc),The t days drop ranges for comparing proxima luce (prox. luc)；For index rolling average index,,, using the index of construction as the input of model；
Stock price trend state is divided into rise, drop two states by step 12, and same day closing price is higher than closing quotation in second day
Valence is labeled as+1 for rise state, and same day closing price is less than or equal to second day closing price, is pulldown conditions, labeled as-
1, using mark value as the output of model；
Established sample set is divided into training set and test set by step 13, and preceding the 80% of sample set is used as training set, sample
Rear the 20% of collection is used as test set.
Fig. 3 is the flow diagram that the present invention chooses that the optimal chromosome in each subgroup constitutes elite group, as shown in figure 3, choosing
It takes the optimal chromosome in each subgroup to constitute elite group, includes the following steps：
Step 21, training dataForDimension data, the input neuron of extreme learning machineIt is a, hidden neuronIt is a,
It will be denoted as initial population labeled as a subgroup per data line, each subgroup includesA dyeing
Body, wherein each chromosomeAll includeA input weights andA threshold value, and using initial population as first generation population：
Wherein,It is input weights,It is hidden neuron threshold value, the weights and threshold value in initial population obtain at random.
Wherein；
Step 22 randomly selects hidden node parameterWith, calculate the fitness value of the chromosome in each subgroup,
Equation is as follows：
WhereinIt isThe output actual value of group data,Weights are exported for hidden neuron,For hidden node
Export square；
Step 23, using algorithm of tournament selection method, select optimal chromosome, equation is as follows：
Wherein, optimal chromosome is the highest chromosome of fitness value,For chromosomeFitness value,,For next-generation population, highest one of fitness in two chromosomes is chosen every time and is entered in next-generation population；
Whether step 24 intersects by crossover probability the individual selectedIt determines, if the random number generated is more than, then
Intersected, is less than or equal toWhen, without intersecting, cross-equation is as follows：
WhereinIt is that equally distributed random number is obeyed in [0,1] range；
Step 25, variation use nonuniform meshes, according to mutation probabilityIt makes a variation per one-dimensional to all individuals, formula
It is as follows：
Wherein,For individualThe upper bound,For individualLower bound,For current iterations,It changes for maximum
Generation number,To determine the parameter of nonuniform meshes degree, value 2；
Step 26 calculates ideal adaptation angle value after compiling, finds out the highest individual of fitness and is used as current optimum individual, takes each
The optimum individual of subgroup forms elite group.
In the present invention, in step 22, hidden node parameter is randomly selectedWithAfterwards, the dye in each subgroup is calculated
The fitness value of colour solid, include the following steps：
Step S1, hidden node parameter is randomly selectedWith, calculate hidden node output matrix, equation
It is as follows：
WhereinIt is vectorial for input weight,It is hidden node excitation function for Sigmoid functions；
Step S2, according to hidden layer output matrixIt calculates hidden layer and exports weights to output layer, formula is as follows：
,,
WhereinIt is hidden layer output matrixLeft pseudo inverse matrix,It is exported for target, i.e.,；
Step S3, according to hidden layer output matrixWeights are exported with hidden layer to output layerCalculate the fitness value of chromosome,
Equation is as follows：
WhereinIt isThe output actual value of group data,Weights are exported for hidden neuron,For hidden node
Output matrix.
Fig. 4 is that the present invention chooses optimal chromosome as input weights and threshold value flow diagram, as shown in figure 4, choosing
Optimal chromosome includes the following steps as input weights and threshold value：
Step 31, every one-dimensional speed of random initializtion population existIt is interior,It is set as 1200, stopping criterion setting
It is more than 10 to reach maximum iteration 50 or fitness；
Step 32, mine massively to the elite after initialization carries out local search with standard particle group algorithm, if meeting stopping criterion,
Select optimal chromosome；If not satisfied, the chromosome randomly selected in 10 chromosome limiting values and the subgroup in elite group is mutual
It is exchanged, is recycled for step 21, until reaching stopping criterion, select optimal chromosome as input weights and threshold value.
Fig. 5 is that the flow diagram that the present invention calculates predicted value includes the following steps as shown in figure 5, calculating predicted value：
Step 41 utilizes the input weights and threshold calculations hidden node output matrix after optimization, equation is such as
Under：
WhereinIt is vectorial for input weight,It is hidden node excitation function for Sigmoid functions；
Step 42, according to hidden layer output matrixIt calculates hidden layer and exports weights to output layer, formula is as follows：
,,
WhereinIt is hidden layer output matrixLeft pseudo inverse matrix,It is exported for target, i.e.,；
Step 43, according to hidden layer output matrixWeights are exported with hidden layer to output layerCalculate output valve, formula is as follows：
,
WhereinIt isThe output actual value of group data,It isGroup data export predicted value,For hidden layer section
Point output matrix；
Step 44 is traded according to the predicted value of model, and prediction result is+1, increases target stock position in storehouse；Prediction result be-
1, reduce target stock position in storehouse.
Embodiment
The present embodiment selects specific stock --- and Shanghai security composite index is changed using genetic algorithm and particle cluster algorithm
Into extreme learning machine model ups and downs trend is predicted, and be traded according to prediction result, specific steps are as follows institute
Show.
Test data is obtained from 365 database of great wisdom by network, by 2005.01.04 to 2017.08.01's
Shanghai security composite index is as experiment initial data, wherein including opening price, highest price, lowest price, the receipts of each day of trade
Disk valence, trading volume.
By initial data constructing technology index, including pass throughIt calculates discrete indicator, pass throughIt calculates
Take advantage of a situation index, pass throughIt calculates momentum index, pass throughIt calculates Relative Strength Index, pass throughIt calculates random index, pass throughIt calculates departure rate, pass throughCalculate energy
Index passes throughIt calculates Larry William's indexs, pass throughWeighting in n days is calculated to slide
It is dynamic averagely to fluctuate index, wherein whereinT days highest prices, lowest price and closing price are indicated respectively；
Highest price and lowest price before indicating t days respectively in n days；,,；The t days amounts of increase for comparing proxima luce (prox. luc),The t days drop ranges for comparing proxima luce (prox. luc)；,,For index rolling average index.
Be used as training set by the 80% of data set, i.e., using the data set of 2005.01.04 to 2015.01.27 as training set,
It is used as test set by the 20% of data set, i.e., using the data set of 2015.01.27 to 2017.08.01 as test set.
Classification prediction is carried out using genetic algorithm and the improved extreme learning machine prediction model of particle cluster algorithm, what is obtained is pre-
It is 69.81% to survey accuracy rate, and is compared with incremental extreme learning machine model, online sequential extreme learning machine model, tradition
The predictablity rate of extreme learning machine model be 60.19%, the predictablity rate of incremental extreme learning machine model is
53.42%, the predictablity rate of online sequential extreme learning machine model is the accuracy rate ratio tradition of 56.88%, this paper prediction models
Extreme learning machine, incremental extreme learning machine, online sequential extreme learning machine difference it is high by 9.62%, 16.39%, 12.93%.
The selection of the number of hidden nodes has a certain impact to the accuracy rate of model prediction, by the hidden node of all models
Number is initially set to 1, every time plus 1, until 50, obtains model prediction accuracy rate comparison diagram as shown in fig. 6, wherein horizontal axis
Indicate that the number of hidden nodes, the longitudinal axis represent the accuracy rate of prediction, GA-PSO-ELM indicates the utilization genetic algorithm and population of this paper
The extreme learning machine prediction model of algorithm improvement, ELM indicate that traditional extreme learning machine model, I-ELM indicate the incremental limit
Learning machine model, OS-ELM indicate online sequential extreme learning machine model.
Claims (6)
1. a B shareB trend forecasting method, for conventional limit learning machine, when handling stock trend prediction, precision is not high asks
Topic is studied, which is characterized in that the prediction technique includes the following steps:
Step 1, using source data constructing technology index as the input of model, stock price trend state is determined as output
Training set and test set；
Step 2 dyes the threshold map of the input weights of extreme learning machine and hidden node for every in Population in Genetic Algorithms
Gene on body takes the optimal chromosome in each subgroup to constitute elite group using the ability of searching optimum of genetic algorithm；
Step 3 selects optimal chromosome in elite group using the local search ability of particle cluster algorithm, as an optimization the rear limit
The input weights and threshold value of learning machine；
Step 4, according to the input weights and threshold value after optimization, utilize least square method to calculate extreme learning machine hidden neuron
Weights are exported, predicted value is then calculated, is traded according to the predicted value of model.
2. B shareB trend forecasting method according to claim 1, which is characterized in that in step 1, utilize source data structure
Technical indicator is made as mode input, stock price trend state determines that training set and test set include following step as output
Suddenly：
Step 11 passes through feature letter according to the opening price of initial data, closing price, highest price, lowest price, exchange hand essential information
Number constructing technology index, including：Discrete indicator, index of taking advantage of a situation, momentum index, Relative Strength Index, random index, departure rate,
Energy indexes, Larry William's indexs, n days weighting sliding averages fluctuate index, using the index of construction as the defeated of model
Enter；
Stock price trend state is divided into rise, drop two states by step 12, and same day closing price is higher than closing quotation in second day
Valence is labeled as+1 for rise state, and same day closing price is less than or equal to second day closing price, is pulldown conditions, labeled as-
1, using mark value as the output of model；
Established sample set is divided into training set and test set by step 13, and preceding the 80% of sample set is used as training set, sample
Rear the 20% of collection is used as test set.
3. B shareB trend forecasting method according to claim 1, which is characterized in that in step 2, by extreme learning machine
Input weights and the threshold map of hidden node be gene in Population in Genetic Algorithms on every chromosome, utilize genetic algorithm
Ability of searching optimum, take the optimal chromosome in each subgroup to constitute elite group, include the following steps：
Step 21, training dataForDimension data, the input neuron of extreme learning machineIt is a, hidden neuronIt is a,
It will be denoted as initial population labeled as a subgroup per data line, each subgroup includesA dyeing
Body, wherein each chromosomeAll includeA input weights andA threshold value, and using initial population as first generation population：
Wherein,It is input weights,It is hidden neuron threshold value, the weights and threshold value in initial population obtain at random；
Wherein；
Step 22 randomly selects hidden node parameterWith, calculate the fitness value of the chromosome in each subgroup,
Equation is as follows：
WhereinIt isThe output actual value of group data,Weights are exported for hidden neuron,For hidden node
Export square；
Step 23, using algorithm of tournament selection method, select optimal chromosome, equation is as follows：
Wherein, optimal chromosome is the highest chromosome of fitness value,For chromosomeFitness value,,
For next-generation population, highest one of fitness in two chromosomes is chosen every time and is entered in next-generation population；
Whether step 24 intersects by crossover probability the individual selectedIt determines, if the random number generated is more than, then into
Row intersects, and is less than or equal toWhen, without intersecting, cross-equation is as follows：
WhereinIt is that equally distributed random number is obeyed in [0,1] range；
Step 25, variation use nonuniform meshes, according to mutation probabilityIt makes a variation per one-dimensional to all individuals, formula is such as
Under：
Wherein,For individualThe upper bound,For individualLower bound,For current iterations,For greatest iteration
Number,To determine the parameter of nonuniform meshes degree, value 2；
Step 26 calculates ideal adaptation angle value after compiling, finds out the highest individual of fitness and is used as current optimum individual, takes each
The optimum individual of subgroup forms elite group.
4. B shareB trend forecasting method according to claim 1, which is characterized in that in step 22, randomly select hidden
Node layer parameterWithAfterwards, the fitness value of the chromosome in each subgroup is calculated, include the following steps：
Step S1, hidden node parameter is randomly selectedWith, calculate hidden node output matrix, equation is such as
Under：
WhereinIt is vectorial for input weight,It is hidden node excitation function for Sigmoid functions；
Step S2, according to hidden layer output matrixIt calculates hidden layer and exports weights to output layer, formula is as follows：
,,
WhereinIt is hidden layer output matrixLeft pseudo inverse matrix,It is exported for target, i.e.,；
Step S3, according to hidden layer output matrixWeights are exported with hidden layer to output layerCalculate the fitness value of chromosome,
Equation is as follows：
WhereinIt isThe output actual value of group data,Weights are exported for hidden neuron,For hidden node
Output matrix.
5. B shareB trend forecasting method according to claim 1, which is characterized in that in step 3, calculated using population
The local search ability of method selects optimal chromosome, as an optimization the input weights and threshold value of rear extreme learning machine, including such as
Lower step：
Step 31, every one-dimensional speed of random initializtion population existIt is interior,It is set as 1200, stopping criterion setting
It is less than 1 to reach maximum iteration 50 or fitness；
Step 32, mine massively to the elite after initialization carries out local search with standard particle group algorithm, if meeting stopping criterion,
Select optimal chromosome；If not satisfied, the chromosome randomly selected in 10 chromosome limiting values and the subgroup in elite group is mutual
It is exchanged, is recycled for step 21, until reaching stopping criterion, select optimal chromosome as input weights and threshold value.
6. B shareB trend forecasting method according to claim 1, which is characterized in that in step 4, after optimization
Weights and threshold value are inputted, the output weights of extreme learning machine hidden neuron are calculated using least square method, then calculates prediction
Value, is traded according to the predicted value of model, includes the following steps：
Step 41 utilizes the input weights and threshold calculations hidden node output matrix after optimization, equation is such as
Under：
WhereinIt is vectorial for input weight,It is hidden node excitation function for Sigmoid functions；
Step 42, according to hidden layer output matrixIt calculates hidden layer and exports weights to output layer, formula is as follows：
,,
WhereinIt is hidden layer output matrixLeft pseudo inverse matrix,It is exported for target, i.e.,；
Step 43, according to hidden layer output matrixWeights are exported with hidden layer to output layerCalculate output valve, formula is as follows：
,
WhereinIt isThe output actual value of group data,It isGroup data export predicted value,For hidden layer section
Point output matrix；
Step 44 is traded according to the predicted value of model, and prediction result is+1, increases target stock position in storehouse；Prediction result be-
1, reduce target stock position in storehouse.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810233670.6A CN108537663A (en) | 2018-03-21 | 2018-03-21 | One B shareB trend forecasting method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810233670.6A CN108537663A (en) | 2018-03-21 | 2018-03-21 | One B shareB trend forecasting method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108537663A true CN108537663A (en) | 2018-09-14 |
Family
ID=63484508
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810233670.6A Pending CN108537663A (en) | 2018-03-21 | 2018-03-21 | One B shareB trend forecasting method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108537663A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109635914A (en) * | 2018-12-17 | 2019-04-16 | 杭州电子科技大学 | Optimization extreme learning machine trajectory predictions method based on hybrid intelligent Genetic Particle Swarm |
| CN110222606A (en) * | 2019-05-24 | 2019-09-10 | 电子科技大学 | Electronic system fault forecast method based on tree search extreme learning machine |
| CN111413285A (en) * | 2020-05-08 | 2020-07-14 | 中南大学 | Method for correcting oxygen detection error in glass bottle based on environmental compensation model |
-
2018
- 2018-03-21 CN CN201810233670.6A patent/CN108537663A/en active Pending
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109635914A (en) * | 2018-12-17 | 2019-04-16 | 杭州电子科技大学 | Optimization extreme learning machine trajectory predictions method based on hybrid intelligent Genetic Particle Swarm |
| CN110222606A (en) * | 2019-05-24 | 2019-09-10 | 电子科技大学 | Electronic system fault forecast method based on tree search extreme learning machine |
| CN110222606B (en) * | 2019-05-24 | 2022-09-06 | 电子科技大学 | Early failure prediction method of electronic system based on tree search extreme learning machine |
| CN111413285A (en) * | 2020-05-08 | 2020-07-14 | 中南大学 | Method for correcting oxygen detection error in glass bottle based on environmental compensation model |
| CN111413285B (en) * | 2020-05-08 | 2021-04-20 | 中南大学 | Method for correcting oxygen detection error in glass bottle based on environmental compensation model |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106600059A (en) | Intelligent power grid short-term load predication method based on improved RBF neural network | |
| CN109214449A (en) | A kind of electric grid investment needing forecasting method | |
| CN103105246A (en) | Greenhouse environment forecasting feedback method of back propagation (BP) neural network based on improvement of genetic algorithm | |
| CN110298663A (en) | Based on the wide fraudulent trading detection method learnt deeply of sequence | |
| CN108537663A (en) | One B shareB trend forecasting method | |
| CN106845012A (en) | A kind of blast furnace gas system model membership function based on multiple target Density Clustering determines method | |
| CN108229750A (en) | A kind of stock yield Forecasting Methodology | |
| CN108734220A (en) | Adaptive Financial Time Series Forecasting method based on k lines cluster and intensified learning | |
| CN110059867A (en) | A kind of wind speed forecasting method of SWLSTM combination GPR | |
| CN110210973A (en) | Insider trading recognition methods based on random forest and model-naive Bayesian | |
| Zhang et al. | Non-tradable shares pricing and optimal default point based on hybrid KMV models: Evidence from China | |
| CN112884590A (en) | Power grid enterprise financing decision method based on machine learning algorithm | |
| Xu et al. | A novel ensemble credit scoring model based on extreme learning machine and generalized fuzzy soft sets | |
| Liu et al. | Quantified moving average strategy of crude oil futures market based on fuzzy logic rules and genetic algorithms | |
| Wu et al. | Ensemble model of intelligent paradigms for stock market forecasting | |
| Zhu et al. | Exploring a multi-objective cluster-decomposition framework for optimizing flood control operation rules of cascade reservoirs in a river basin | |
| CN109697531A (en) | A kind of logistics park-hinterland Forecast of Logistics Demand method | |
| Tian et al. | A wind speed prediction system based on new data preprocessing strategy and improved multi-objective optimizer | |
| Syu et al. | Anomaly detection networks and fuzzy control modules for energy grid management with q-learning-based decision making | |
| Du | Prediction of consumer price index based on RBF neural network | |
| Ying et al. | A novel optimization algorithm for BP neural network based on RS-MEA | |
| Zhong et al. | Improved BP neural network's application in the bank early warning | |
| Zhao et al. | Construction and simulation of market risk warning model based on deep learning | |
| Banik et al. | Modeling chaotic behavior of Dhaka stock market index values using the neuro-fuzzy model | |
| Subhan et al. | Uncovering Price Puzzle in the Wheat Economy of Pakistan: An Application of Artificial Neural Networks |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20180914 |
|
| WD01 | Invention patent application deemed withdrawn after publication |