CN110263991A - A kind of stock index price expectation method of quantum nerve network - Google Patents
A kind of stock index price expectation method of quantum nerve network Download PDFInfo
- Publication number
- CN110263991A CN110263991A CN201910499094.4A CN201910499094A CN110263991A CN 110263991 A CN110263991 A CN 110263991A CN 201910499094 A CN201910499094 A CN 201910499094A CN 110263991 A CN110263991 A CN 110263991A
- Authority
- CN
- China
- Prior art keywords
- data
- module
- quantum
- nerve network
- prediction
- 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
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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)
- Economics (AREA)
- Strategic Management (AREA)
- General Physics & Mathematics (AREA)
- Finance (AREA)
- Development Economics (AREA)
- Marketing (AREA)
- Accounting & Taxation (AREA)
- Human Resources & Organizations (AREA)
- General Business, Economics & Management (AREA)
- Operations Research (AREA)
- Computational Linguistics (AREA)
- Entrepreneurship & Innovation (AREA)
- Technology Law (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Game Theory and Decision Science (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)
- Financial Or Insurance-Related Operations Such As Payment And Settlement (AREA)
Abstract
A kind of stock index price expectation method of quantum nerve network, based on " primary step empirical mode decomposition algorithm ", that is PEEMD algorithm, including the training of data input module, data preprocessing module, data conversion module, data and prediction module, data reconstruction module.Data input module is used to obtain the newest transaction data of stock index, data preprocessing module is for decomposing data, data conversion module is used to convert " quantum state " data for initial data, data training and prediction module are used to for " quantum state " data being trained prediction, and data reconstruction module is used to reconstruct the prediction result of the data.The present invention pre-processes initial data first with PEEMD algorithm, the time series data of non-stationary is decomposed into the approximate stable data of multiple and different frequencies and rejects high fdrequency component therein, simulation and prediction only is carried out to low frequency components throughput sub-neural network, finally each simulation result is reconstructed to obtain final prediction result, to effectively improve the estimated performance of model.
Description
Technical field
The present invention relates to stock index price expectation method and technology field and algorithm application field, in particular to a kind of quantum
The stock index price expectation method of neural network.
Background technique
Stock market unlike general linear system have very strong regularity, it often have non-linear, fluctuation and
It is difficult to predict the features of property, and moreover, the tendency of stock market also suffers from locating economic bad border, the people factors such as expected at heart
Influence.
Non-linear, fluctuation complication system this for stock market establishes model and predict substantially to share price
Exactly establish the non-linear relation that a mathematical model is come between simulation input and output parameter.
Economic growth and development of financial market are there are positive correlation, and stock market has pushed economic growth, it is promoted
The circulation and transaction of other idle and more dispersed capitals in economy.At the same time, when macroeconomic increases, economy starts
When recovery, bring additive effect can promote the development of stock market, this positive correlation mutually promoted in turn again
Relationship can be different because of country, and different because of level of economic development difference.
Quantum nerve network model is one of current more novel neural network prediction model, it is in traditional neural network
Framework on combine " quantum calculation " and form a kind of new neural network structure, input layer and hidden layer neuron are no longer
It is neural unit but quantum neuron, due to introducing " quantum calculation ", so that the forecasting efficiency of the model has obtained pole
It is big to be promoted, and solve some of complex, random Prediction of Nonlinear Dynamical Systems precision problem and traditional neural network and be easy " mistake
The problem of fitting " and " falling into local minimum ".
Due to the fluctuation and uncertainty of stock market, stock price is not only influenced by history in addition, is also wanted
Consider political factor and the psychological factor of people, it is a highly difficult thing that prediction is carried out to share price, for investor,
Its trend direction for being most concerned with stock prices, although estimated performance of the quantum nerve network in linear system is fine,
It is the time series data with nonlinear characteristic this for stock prices, the precision of prediction of quantum nerve network especially trend side
Promotion to the upper not substance of judgement, and none better algorithm incorporating quantum neural network pair currently on the market
Stock index price is predicted.
Summary of the invention
A kind of stock index price expectation method that the present invention provides quantum nerve network is aimed to solve the problem that just with market
Historical transactional information predicts that stock index price, trend direction prediction precision is not high and model training efficiency is too low
Problem.
A kind of stock index price expectation method of quantum nerve network is mainly located including data input module, data in advance
Module, data conversion module, data training and prediction module, data reconstruction module are managed, the data input module is for obtaining
The newest transaction data of stock index, the data preprocessing module are used for the time series data inputted to the data input module
It is pre-processed by PEEMD algorithm (primary step empirical mode decomposition algorithm), the data conversion module is used for will be described
The data that data preprocessing module is handled are converted into " quantum state " data, and the data training and prediction module are used for institute
It states " quantum state " data that data conversion module obtains and is trained prediction, the data reconstruction module is for reconstructing the data
Trained and prediction module prediction result, prediction steps are: the input data of the data input module is sent to the number
Data preprocess module is pre-processed, and is then sent through the data conversion module and is converted, then by data training and
Prediction module is predicted, prediction result is reconstructed finally by the data reconstruction module to obtain final prediction knot
Fruit.
In the above method, the data preprocessing module mainly includes that initial data is decomposed and high frequency clutter picks
It removes.
In the above method, data training and prediction module mainly include the building module and model of quantum nerve network
Prediction module.
In the above method, the building module of the quantum nerve network mainly includes Three Tiered Network Architecture, wherein input layer
It is quantum neuron with hiding node layer, output node layer is neural unit, and wherein input layer number of nodes is 4, hidden
Hiding node layer number is 3, and output layer number of nodes is 1, and the activation primitive of hidden layer is Sigmoid function, the activation primitive of output layer
It is ReLU function, input layer is connected with hidden layer by rotation parameter and overturning parameter, and hidden layer and output layer pass through commonly
Weight is connected.
In the above method, the model prediction module is to pass through the quantum mind using the data of the data conversion module
Building module through network predicts the daily closing price of stock index.
In the above method, a kind of stock index price expectation method of quantum nerve network, which is characterized in that described
PEEMD algorithm steps are as follows:
1) normal distribution white noise is added in signal to be analyzed;
2) it is mixed with the original signal of normal distribution white noise described in EMD (empirical mode decomposition algorithm) decomposition, obtains one
The intrinsic modal components of series;
3) final decomposition result will be used as after the modal components average treatment obtained every time;
4) it repeats that new normal distribution white noise is added in remaining signal every time 1) to 3);
5) it checks whether last residual error item meets suspension condition, if meeting suspension condition, stops iteration;
6) its medium-high frequency aliazing components is rejected, using low frequency in residue and residual error item component as final output.
In the above method, the initial data, which is decomposed, mainly utilizes EEMD (set empirical mode decomposition algorithm) algorithm,
The algorithm is white noise signal to be added in original signal, and as a whole by the combination of acquisition, noise is repeatedly being asked
Be averaged after cancel out each other, the signal component of different scale be distributed to automatically it is appropriate with reference to scale, finally to being mixed with Gauss
The data of white noise are successively decomposed, and a series of approximate smoothly intrinsic modal components of different frequencies are obtained, different frequency
Component contains the characteristic information of initial data respective frequencies.
In the above method, the rejecting of the high frequency clutter, essence rejects the initial data and decomposes to obtain
The intrinsic modal components of high frequency are only trained prediction to low frequency signal in residue.
In the above method, the EEMD algorithm is by being introduced into white noise in the signal to be analyzed, to being mixed with white noise
Original signal carries out EMD decomposition, then is averaged to obtain final result to each component.
The stock index price expectation method of a kind of quantum nerve network, which is characterized in that quantum nerve network
The main learning algorithm for constructing module is as follows:
Real number state is converted into quantum state:
By the real vector of n-dimensional space by certain mapping relations, it is converted into the number for meeting quantum nerve network input
According to structure, conversion formula is as follows:
| X >=[| x1>,|x2>,…|xn>]T,
Error reverse conduction adjusting parameter:
In quantum nerve network, to learning error by quantum nerve network carry out reverse conduction, with this adjust with
Lower four parameters: twiddle factor, the overturning factor, output layer threshold value, connection weight, and global optimum is found by repetition training
Solution, wherein error function is as follows:
According to gradient descent method, the direction of parameter adjustment can be derived and as a result, the relational expression of its adjustment is as follows:
For threshold value, it is adjusted using momentum gradient descent method and factor adaptive learning method, obtains final adjustment
Parameter afterwards.
A kind of stock index price expectation method of quantum nerve network proposed by the present invention, advantages and beneficial effects exist
In using PEEMD algorithm, by non-stationary random fluctuation signal decomposition, for a series of approximations, smoothly time series datas can be improved mould
Stability of the type under fluctuation interference.By rejecting the aliazing components of high frequency mode, mould only is carried out to remaining low frequency components
Type prediction is able to solve influence of the dither signal of high frequency mixing for model training, meanwhile, by the low frequency components amount of being brought into
The trend walking direction and precision of prediction of model can be obviously improved in sub-neural network model, operational efficiency also more traditional mind
There is biggish promotion through Network Prediction Model.
Detailed description of the invention
It, below will be to embodiment reference in order to illustrate more clearly of purpose, performance characteristics and advantage that the present invention realizes
Attached drawing is briefly described.
Fig. 1 is stock index price expectation method schematic diagram of the present invention.
Fig. 2 is that the present invention is based on the flow diagrams of quantum nerve network embodiment.
Fig. 3 is quantum calculation process schematic of the present invention.
Fig. 4 is quantum nerve network of the present invention training flow chart.
In Fig. 1: 1- data input module, 2- data preprocessing module, 3- data conversion module, the training of 4- data and prediction
Module, 5- data reconstruction module, the building module of 6- quantum nerve network model, 7- model prediction module.
In Fig. 2: 8- obtains stock index futures price history transaction data, and 9- cleans initial data, rejects strong
Noise, default value, 10-PEEMD algorithm decompose, and 11- divides training set and test set, 12- training set and test set, 13- model
Trained and model prediction.
Specific embodiment
The specific embodiment of the invention is further explained below in conjunction with embodiment and attached drawing, it should be understood that the solution
It releases and is used only for explaining the present invention, and not intended to limit the protection scope of the present invention.
As shown in Figure 1, a kind of stock index price expectation method of quantum nerve network, mainly includes data processing section
With model training and predicted portions, wherein data processing section includes data input module 1, data preprocessing module 2 and data
Conversion module 3, model training and predicted portions include that data training and prediction module 4 and data reconstructed module 5, the data are defeated
Enter module and is used to input the data for obtaining the newest transaction data of stock index futures, the data preprocessing module
The time series data of module input is pre-processed by PEEMD algorithm, and the data conversion module for locating the data in advance
The data that reason resume module obtains are converted into " quantum state " data, and the data training and prediction module are used to turn the data
Change " quantum state " data that module obtains and be trained prediction, the data reconstruction module is trained and pre- for reconstructing the data
The prediction result of module is surveyed, prediction steps are: the input data of the data input module is sent to the data prediction
Module is pre-processed, and is then sent through the data conversion module and is converted, and the data training and prediction module are then passed through
It is predicted, prediction result is reconstructed to obtain final prediction result finally by the data reconstruction module.
As shown in Figure 1, mainly the building module 6 including quantum nerve network and model are pre- for data training and prediction module 4
Survey module 7.
The building module 6 of the quantum nerve network mainly includes Three Tiered Network Architecture, wherein input layer and hidden layer section
Point is quantum neuron, and output node layer is neural unit, and wherein input layer number of nodes is 4, hidden layer number of nodes
It is 3, output layer number of nodes is 1, and the activation primitive of hidden layer is Sigmoid function, and the activation primitive of output layer is ReLU function,
Input layer is connected with hidden layer by rotation parameter and overturning parameter, and hidden layer is connect with output layer by equity stock heavy phase.
It is described the embodiment of the invention provides a kind of stock index price expectation method of quantum nerve network referring to Fig. 2
Method includes:
Step 8, target stock day history trading situation is obtained.
Target stock be main six deep bids and stock index index, day history transaction value include " opening price ", " closing price ",
The common Day Trading price data of " highest price ", " lowest price ".
Step 9, " cleaning " is carried out to the common Day Trading price initial data, eliminated some endless due to recording
Whole, individual data is omitted and the influence with obvious errors noise spot.
After the characteristic informations such as " turnover rate " in each data are rejected, only retain four price features such as " closing prices "
Index alternately, is analyzed finally by the correlation to each sample data, finds " closing price " for stock valence
Lattice wave moves and the correlation of trend trend is most strong, while during practical stock exchange, investor is more keen to understand every
" closing price " of one stock is to guarantee oneself to make correct selection before stock market closes, therefore this paper final choice " is received
This feature of disk valence " carries out subsequent experimental data processing.
It should be noted that will appear oscillatory occurences during real number state is converted to quantum state, therefore initial
Each parameter can be normalized adjustment when changing parameter, to prevent since some slight disturbances are so that model
There is huge shake.
Here the most commonly used is utilizations its " 2 norm " to be normalized, such as: for single row (row) vector just with biography
" unitization " method of system carries out " normalization " processing, for matrix, then carries out at " 2 norm " normalization to its every a line vector
Reason.A normalized just is carried out to parameter after the completion of iteration each time, so that trained result is lesser at one every time
Fluctuation, the precision of prediction of model are greatly improved in range, and model also can be more stable.
Step 10, the data after cleaning are decomposed by PEEMD algorithm, obtains a series of eigen mode of different frequencies
Then state function removes the high fdrequency component wherein lain in initial data again, remaining low frequency components are passed through the number
It is predicted according to trained and prediction module 4.
It should be noted that for high-frequency oscillation signal, can be seen as influencing stock price trend it is short-term because
Element, such as provisional hand-off transaction, of short duration market concussion, intermediate frequency component can regard a kind of mid-term factor, such as season as
Property influence, cyclic effects, last low frequency component and residual component show a kind of long-term trend, it represents stock price
Whole tendency, that is, the rough trend of stock price is considered in macroscopic aspect, the ups and downs without focusing simply on some point
Situation, the influence factor of this long-term trend may be exactly that the either entire stock market of entire stock market's policy changes changes.
Step 11, training set and test set are divided in the initial data, since the size of training set sample size can shadow
The training effect of model and the predictive ability of model are rung, the number of training set sample should be made to be more than test set when dividing,
It should additionally carry out randomly selecting data composition training set in different data sets to guarantee that model with universality, reaches friendship
Pitch the effect of verifying.
Step 12, it is to be ensured that training set and test set separate completely, since stock index price is time series data,
Therefore it should be divided according to the sequencing of time when dividing training set, otherwise prediction result will lose meaning.
Step 13, model is trained on training set, one group of parameter for adapting to the data is obtained, then in test set
On predicted, obtain prediction result.
It should be noted that regarding the result of each resulting separation as a new original training data, then
Predictive simulation test is carried out to these new initial data, it, can be by it since component its fluctuation after separation substantially reduces
It regards the stable time series of approximation as, after being trained prediction by neural network to each stable timing, then weighs
New consolidated forecast obtains final prediction result, is made in this way by the thought decomposed, predict, reconstructed again as a result, joint account
Prediction result has biggish promotion than direct prediction result on precision of prediction.
Quantum nerve network in a kind of stock index price expectation method of quantum nerve network of the present invention is most
It include mainly " quantum bit " and " quantum calculation ", wherein the essence of " quantum calculation " is exactly that throughput cervical orifice of uterus acts on quantum state
On make it from transformation of quantum states another quantum state, this point and classical field two-dimensional matrix operation are similar.
As shown in figure 3, quantum calculation process schematic, wherein X, Z, H respectively for quantum non-gate, Z and Hadamard
Door, concrete form are as follows:
Quantum state α | 0 >+β | 1 > indicated with vector symbol are as follows:
Wherein α is | 0 > probability amplitude, β is | 1 > probability amplitude, then the quantum state passes through the output of quantum non-gate
The result is that:
For quantum state α | 0 >+β | 1 >, it is desirable that α and β meets | α |2+|β2=1, therefore obtained newly by quantum non-gate operation
Quantum state α ' | 1 >+β ' | 0 > also to meet | α ' |2+|β′|2=1, it can thus be appreciated that matrix U meetsThat is U is
Unitary matrice, whereinMatrix is the associate matrix of U matrix, and I is unit matrix, and this unitary matrice constraint is quantum door
Unique constraint, any one unitary matrice all specify an effective quantum door.
As shown in figure 4, the node of input layer is to convert the amount to be formed by data in quantum nerve network training flow chart
Sub- neuron, connecting between input layer and hidden layer is rotation parameter and overturning parameter, and it is common for connecting hidden layer with output layer
Connection weight, key step are as follows: input sample first pass around revolving door transformation, secondly using controlled not-gate construct hide
Layer, then hidden layer is mapped to network and exported by connection weight, determines final output finally by stop condition is judged
As a result.
The most basic element of quantum calculation is quantum bit, compared with " position " basic element in traditional calculations, in quantum
In calculating, the different conditions of quantum are described usually using quantum bit, quantum bit state | 1 > and | 0 > respectively indicate excitability
And inhibitory neuron.
In the quantized system containing n quantum bit, quantum operation is usually that throughput cervical orifice of uterus executes single operation in fact
Existing, it is stateful that quantum operation can update institute simultaneously according to quantum parallelism, effectively improves computational efficiency.
For ease of calculation with application, quantum state can indicate with following equation, wherein probability amplitude | 1 >, | 0 > it is corresponding
In real and imaginary parts.
| θ >=f (θ)=eiθ=cos θ+isin θ.
Illustrate the processing of a kind of stock index price expectation method of quantum nerve network below with reference to specific example
Process.
" Hu-Shen 300 index " is used as target stock index, initial data is from April 4 days to 2019 year January in 2005
Obtained in day history trading situation (day k line) on the 11st, which includes " opening price ", " closing price ", " highest price ",
Price data referenced by " lowest price " common Day Trading and quantitative analysis.
In data selection, " cleaning " first is carried out to initial data, eliminates some imperfect due to recording, individual data
Omission and the influence with obvious errors noise spot, have finally obtained 3468 valid data, and each of them data represent
The Day Trading information of " Hu-Shen 300 index ".
Due to the requirement to training data, after the characteristic informations such as " turnover rate " in each data are rejected, only retain
" closing price " characteristic index alternately, is analyzed finally by the correlation to each sample data, discovery " closing quotation
The correlation that valence " moves towards Stock Price Fluctuation and trend is most strong, therefore final choice " closing price " this price feature refers to
Number carries out subsequent experimental data processing.
Before predicting " Hu-Shen 300 index ", first passes through the initial data after PEEMD method is cleaned and divided
Solution, obtains 10 intrinsic mode functions and 1 residual error trend term.
After each intrinsic mode function is normalized, can obtain dimension is 11 × 3468 sample numbers
According to matrix, 3468 raw sample datas are then divided into the training set that sample size is 2312, the total number of samples of Zhan
2/3 and sample size be 1156 test sets, the 1/3 of the total number of samples of Zhan.
Node all amounts of being of a total of three-decker of quantum nerve network model of the present invention, input layer and hidden layer
Sub- neuron is attached between quantum neuron by revolving door and turnover door, there is one on each quantum neuron
Threshold value and activation primitive, this point are similar compared with traditional neural network.
Since effective input of training sample set is 2312 samples, and in view of stock market's period etc. is for stock market
The influence of price trend, therefore the data after the standardization of preceding 4 phase are regarded as to 4 dimension input feature values of model herein, it
It is inputted as input value, the closing price that the latter phase to be predicted is exported as output valve, realized to the one of closing price
Step prediction, so the training network has the node of 4 input terminals, there is 1 output end node.Since the data of each phase below are with before
The data connection of face several phases is closer, therefore by means of the thought of sliding average in time series, for each step predicted value
It carries out sliding backward a phase with trained, final Prediction of Stock Price sequence is obtained with this.
Quantum nerve network model of the present invention training when need to adjust four parameters, be respectively " rotation because
Son ", " the overturning factor ", " articulamentum weight " and " output layer threshold value " can all be set in advance when initializing to each
A fixed random number, random number appears in " 1 " and " -1 " nearby (because near " 1 " and " -1 ", network is extremely easy in order to prevent
Over-fitting), therefore a range is set to it when setting random number, allow it exported in [- 0.5,0.5] to avoid this
Problem.
The input terminal that the Prediction of Stock Price sequence is brought into model is inputted, the ginseng of training pattern is carried out with this
Number.
According to output interpretation of result it is found that on training sample set, preceding 5 components are since its frequency is higher and noise spot ratio
It is more, by preceding 5 component data, that is, imf1~imf5Data are rejected before model training, are only carried out to 6 components below pre-
It surveys.
According to output interpretation of result it is found that in test sample collection, a series of prediction of available intrinsic modal components
Then data set carries out anti-normalization processing to the predictive data set and is reduced into the data under original dimension, it is again whole
It closes in forecasting sequence, joint account, finally reconstructs forecasting sequence and obtain the prediction result of final stock index price.
The present embodiment only predicts to solve low frequency components by rejecting the high fdrequency component in intrinsic mode function
Model easily falls into the excessively slow problem of " local minimum " and training speed, and stock index price expectation and trend direction can be improved
The precision of judgement.
The foregoing is merely a prefered embodiment of the invention, is not intended to limit the scope of the invention, all to utilize this hair
Any improvement, equivalent replacement made by bright specification and accompanying drawing content, optimization either equivalent structure and algorithm flow transformation etc.
And other correlative technology fields are directly or indirectly used in, should similarly it be included in scope of patent protection of the invention.
Claims (10)
1. a kind of stock index price expectation method of quantum nerve network, it is characterised in that mainly include data input module,
Data preprocessing module, data conversion module, data training and prediction module, data reconstruction module, the data input module
For obtaining the newest transaction data of stock index, what the data preprocessing module was used to input the data input module
Time series data is pre-processed by PEEMD algorithm, and the data conversion module is for handling the data preprocessing module
Obtained data are converted into " quantum state " data, and the data training and prediction module are for obtaining the data conversion module
" quantum state " data be trained prediction, the data reconstruction module is used to reconstruct the pre- of the data training and prediction module
It surveys as a result, its prediction steps is: the input data of the data input module being sent to the data preprocessing module and is carried out in advance
Processing, is then sent through the data conversion module and is converted, and is then predicted by the data training and prediction module, most
Afterwards prediction result is reconstructed to obtain final prediction result by the data reconstruction module.
2. a kind of stock index price expectation method of quantum nerve network according to claim 1, which is characterized in that institute
State rejecting of the data preprocessing module (2) mainly including initial data decomposition and high frequency clutter.
3. a kind of stock index price expectation method of quantum nerve network according to claim 1, which is characterized in that institute
Data training and prediction module (4) are stated mainly including the building module (6) and model prediction module (7) of quantum nerve network.
4. a kind of stock index price expectation method of quantum nerve network according to claim 1, which is characterized in that institute
It is quantum that the building module (6) for stating quantum nerve network, which mainly includes Three Tiered Network Architecture, wherein input layer and hiding node layer,
Neuron, output node layer are neural unit, and wherein input layer number of nodes is 4, and hidden layer number of nodes is 3, output
Node layer number is 1, and the activation primitive of hidden layer is Sigmoid function, and the activation primitive of output layer is ReLU function, input layer with
Hidden layer is connected by rotation parameter with overturning parameter, and hidden layer is connect with output layer by equity stock heavy phase.
5. a kind of stock index price expectation method of quantum nerve network according to claim 1, which is characterized in that institute
Stating model prediction module (7) is the building mould for passing through the quantum nerve network using the data of the data conversion module (3)
Block (6) predicts the daily closing price of stock index.
6. a kind of stock index price expectation method of quantum nerve network according to claim 1, which is characterized in that institute
It is as follows to state PEEMD algorithm steps:
1) normal distribution white noise is added in signal to be analyzed;
2) EMD algorithm is used, the original signal for being mixed with normal distribution white noise is decomposed, obtains a series of intrinsic modal components;
3) final decomposition result will be used as after the modal components average treatment obtained every time;
4) it repeats that new normal distribution white noise is added in remaining signal every time 1) to 3);
5) it checks whether last residual error item meets suspension condition, if meeting suspension condition, stops iteration;
6) its medium-high frequency aliazing components is rejected, using low frequency in residue and residual error item component as final output.
7. a kind of stock index price expectation method of quantum nerve network according to claim 2, which is characterized in that institute
It states initial data to decompose mainly using EEMD algorithm, which is that white noise signal is added in original signal, and will obtain
As a whole, noise is cancelled out each other after multiple summation is averaged, and the signal component of different scale is divided automatically for the combination obtained
Cloth refers to scale to appropriate, is finally successively decomposed to the data for being mixed with white Gaussian noise, obtains a series of different frequencies
Approximate smoothly intrinsic modal components, the component of different frequency contain the characteristic information of initial data respective frequencies.
8. a kind of stock index price expectation method of quantum nerve network according to claim 2, which is characterized in that institute
The rejecting of high frequency clutter is stated, essence is the intrinsic modal components of high frequency rejecting the initial data and decomposing, only right
Low frequency signal is trained prediction in residue.
9. a kind of stock index price expectation method of quantum nerve network according to claim 7, which is characterized in that institute
EEMD algorithm is stated by the way that white noise is introduced into the signal to be analyzed, EMD decomposition is carried out to the original signal for being mixed with white noise, then
Each component is averaged to obtain final result.
10. a kind of stock index price expectation method of quantum nerve network according to any one of claims 1 to 9,
It is characterized in that, the main learning algorithm of the building module (6) of the quantum nerve network is as follows:
Real number state is converted into quantum state:
By the real vector of n-dimensional space by certain mapping relations, it is converted into the data knot for meeting quantum nerve network input
Structure, conversion formula are as follows:
| X >=[| x1>,|x2>,…|xn>]T,
Error reverse conduction adjusting parameter:
In quantum nerve network, reverse conduction carried out by quantum nerve network to learning error, adjust following four with this
A parameter: twiddle factor, the overturning factor, output layer threshold value, connection weight, and globally optimal solution is found by repetition training,
Middle error function is as follows:
According to gradient descent method, the direction of parameter adjustment can be derived and as a result, the relational expression of its adjustment is as follows:
For output layer threshold value, it is adjusted using momentum gradient descent method and factor adaptive learning method, is finally adjusted
Parameter after whole.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910499094.4A CN110263991A (en) | 2019-06-11 | 2019-06-11 | A kind of stock index price expectation method of quantum nerve network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910499094.4A CN110263991A (en) | 2019-06-11 | 2019-06-11 | A kind of stock index price expectation method of quantum nerve network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110263991A true CN110263991A (en) | 2019-09-20 |
Family
ID=67917460
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910499094.4A Pending CN110263991A (en) | 2019-06-11 | 2019-06-11 | A kind of stock index price expectation method of quantum nerve network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110263991A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113077276A (en) * | 2020-01-06 | 2021-07-06 | 阿里巴巴集团控股有限公司 | Behavior data processing method and system, storage medium and processor |
CN113222160A (en) * | 2020-01-21 | 2021-08-06 | 合肥本源量子计算科技有限责任公司 | Quantum state conversion method and device |
CN113409072A (en) * | 2021-05-31 | 2021-09-17 | 河北科技师范学院 | Empirical mode decomposition and distributed GRU neural network and price prediction method |
CN114496227A (en) * | 2022-01-26 | 2022-05-13 | 电子科技大学 | Disease development prediction system and platform based on quantum neural network |
CN115907019A (en) * | 2023-01-09 | 2023-04-04 | 苏州浪潮智能科技有限公司 | Quantum computer, quantum network and time sequence data prediction method |
-
2019
- 2019-06-11 CN CN201910499094.4A patent/CN110263991A/en active Pending
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113077276A (en) * | 2020-01-06 | 2021-07-06 | 阿里巴巴集团控股有限公司 | Behavior data processing method and system, storage medium and processor |
CN113222160A (en) * | 2020-01-21 | 2021-08-06 | 合肥本源量子计算科技有限责任公司 | Quantum state conversion method and device |
CN113222160B (en) * | 2020-01-21 | 2023-08-08 | 本源量子计算科技(合肥)股份有限公司 | Quantum state conversion method and device |
CN113409072A (en) * | 2021-05-31 | 2021-09-17 | 河北科技师范学院 | Empirical mode decomposition and distributed GRU neural network and price prediction method |
CN114496227A (en) * | 2022-01-26 | 2022-05-13 | 电子科技大学 | Disease development prediction system and platform based on quantum neural network |
CN114496227B (en) * | 2022-01-26 | 2023-04-28 | 电子科技大学 | Disease development prediction system and platform based on quantum neural network |
CN115907019A (en) * | 2023-01-09 | 2023-04-04 | 苏州浪潮智能科技有限公司 | Quantum computer, quantum network and time sequence data prediction method |
CN115907019B (en) * | 2023-01-09 | 2023-11-07 | 苏州浪潮智能科技有限公司 | Quantum computer for weather prediction |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110263991A (en) | A kind of stock index price expectation method of quantum nerve network | |
Balcerzak | Multiple-criteria evaluation of quality of human capital in the European Union countries | |
Wei | A GA-weighted ANFIS model based on multiple stock market volatility causality for TAIEX forecasting | |
CN109242207A (en) | A kind of Financial Time Series prediction technique based on deeply study | |
Martínez-Ballesteros et al. | Selecting the best measures to discover quantitative association rules | |
Wang et al. | A new scheme for probabilistic forecasting with an ensemble model based on CEEMDAN and AM-MCMC and its application in precipitation forecasting | |
Long et al. | A combination interval prediction model based on biased convex cost function and auto-encoder in solar power prediction | |
Maiti | Indian stock market prediction using deep learning | |
Ouyang et al. | Are Deep Learning Models Practically Good as Promised? A Strategic Comparison of Deep Learning Models for Time Series Forecasting | |
Goumatianos et al. | Stock selection system: building long/short portfolios using intraday patterns | |
Yuan et al. | A dynamic clustering ensemble learning approach for crude oil price forecasting | |
Zhang | Decision Trees for Objective House Price Prediction | |
Tsyganov | Intelligent technologies for large-scale social system sustainable development | |
Tang et al. | One-shot pruning of gated recurrent unit neural network by sensitivity for time-series prediction | |
Yeh et al. | Fuzzy rule-based stock trading system | |
Huan et al. | A hybrid model of empirical wavelet transform and extreme learning machine for dissolved oxygen forecasting | |
Jiang | Prediction and industrial structure analysis of local GDP economy based on machine learning | |
Howell et al. | Conservation of northern bobwhite on private lands in Georgia, USA under uncertainty about landscape-level habitat effects | |
Sen et al. | Improved classification algorithm by minsup and minconf based on ID3 | |
Reddy et al. | A Beginner's Guide to Federated Learning | |
Plantamura et al. | The holographic fuzzy learning for credit scoring | |
Liu et al. | Efficient electricity sales forecasting based on curve decomposition and factor regression | |
Zhang et al. | Parameter analysis of hybrid intelligent model for the prediction of rare earth stock futures | |
Grinberg | Stock market statistical data analysis for prices forecasting and trading decision support | |
Nartikoev et al. | Endogenous household classification: Russian regions |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190920 |