CN110827148A - Stock market data analysis method of recurrent neural network based on dimension reduction technology optimization - Google Patents

Stock market data analysis method of recurrent neural network based on dimension reduction technology optimization Download PDF

Info

Publication number
CN110827148A
CN110827148A CN201911057108.3A CN201911057108A CN110827148A CN 110827148 A CN110827148 A CN 110827148A CN 201911057108 A CN201911057108 A CN 201911057108A CN 110827148 A CN110827148 A CN 110827148A
Authority
CN
China
Prior art keywords
data
stock
analysis
information
time
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.)
Granted
Application number
CN201911057108.3A
Other languages
Chinese (zh)
Other versions
CN110827148B (en
Inventor
宋亚童
胡俊丰
于润祥
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenzhen Chengqi Asset Management Co.,Ltd.
Original Assignee
Individual
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Individual filed Critical Individual
Priority to CN201911057108.3A priority Critical patent/CN110827148B/en
Publication of CN110827148A publication Critical patent/CN110827148A/en
Application granted granted Critical
Publication of CN110827148B publication Critical patent/CN110827148B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

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
    • G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/04—Trading; Exchange, e.g. stocks, commodities, derivatives or currency exchange
    • 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
    • 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
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computational Linguistics (AREA)
  • Mathematical Physics (AREA)
  • Biophysics (AREA)
  • Artificial Intelligence (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biomedical Technology (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Business, Economics & Management (AREA)
  • Accounting & Taxation (AREA)
  • Finance (AREA)
  • Development Economics (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • Strategic Management (AREA)
  • Technology Law (AREA)
  • General Business, Economics & Management (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention belongs to the technical field of financial information data processing, and discloses a stock market data analysis method of a recurrent neural network based on dimension reduction technology optimization, wherein a data set is subjected to factor analysis respectively, and a first class and a second class of quality factors of the data set are respectively taken out; finding out the parameter and analyzing the relation of the parameter changing along with time to stock rise and fall; and substituting the parameters into the LSTM model for prediction, adding a forgetting gate to each excitation source of the optimized LSTM, and screening the previous information. The invention expands the LSTM neural network in the financial field, successfully applies the concept of forgetting to stock market analysis and improves the accuracy; and introducing a dimensionality reduction algorithm in data and processing and comparing. The invention highlights the advantages of the dimensionality reduction technology and the accuracy of the LSTM network, so that the stock market prediction is more credible than the traditional analysis method; prediction can be applied in practice.

Description

Stock market data analysis method of recurrent neural network based on dimension reduction technology optimization
Technical Field
The invention belongs to the technical field of financial information data processing, and particularly relates to a circulating neural network stock market data analysis method based on dimension reduction technology optimization.
Background
Currently, the closest prior art: in many fields of research and application, principal component analysis generally requires observation of data containing a plurality of variables, and analysis and rule search are performed after a large amount of data is collected. Multivariate large data sets undoubtedly provide rich information for research and application, but also increase the workload of data acquisition to some extent. More importantly, in many cases, there may be correlations between many variables, thereby increasing the complexity of problem analysis. If each index is analyzed separately, the analysis is often isolated and cannot fully utilize the information in the data, so blindly reducing the index will lose much useful information, thereby leading to erroneous conclusions. Therefore, a reasonable method is needed to be found, so that the loss of information contained in the original index is reduced as much as possible while the index required to be analyzed is reduced, and the purpose of comprehensively analyzing the collected data is achieved. Because a certain correlation exists among the variables, the variables with close relation can be changed into new variables as few as possible, the new variables are unrelated in pairs, and then the various information existing in the variables can be represented by fewer comprehensive indexes. Principal component analysis belongs to such dimension reduction algorithms. The PCA, a principal component analysis method, is a most widely used data dimension reduction algorithm. The main idea of PCA is to map n-dimensional features onto k-dimensions, which are completely new orthogonal features, also called principal components, and k-dimensional features reconstructed on the basis of the original n-dimensional features. The task of PCA is to sequentially find a set of mutually orthogonal axes from the original space, the selection of new axes being strongly dependent on the data itself. The first new coordinate axis is selected to be the direction with the largest square difference in the original data, the second new coordinate axis is selected to be the plane which is orthogonal to the first coordinate axis and enables the square difference to be the largest, and the third axis is the plane which is orthogonal to the 1 st axis and the 2 nd axis and enables the square difference to be the largest. By analogy, n such coordinate axes can be obtained. In the new coordinate axes obtained in this way, most of the variances are contained in the preceding k coordinate axes, and the variance contained in the following coordinate axes is almost 0.
And (3) realizing a PCA algorithm based on the SVD decomposition covariance matrix:
inputting: data set X ═ X1，x2，…xNNeeds to be reduced to the k dimension.
1) De-averaging, i.e., each bit feature minus its respective average.
2) Calculating the covariance matrix XXT。
3) And calculating an eigenvalue and an eigenvector of the covariance matrix through SVD.
4) Sorting the eigenvalues from large to small, and selecting the largest k of the eigenvalues. Then, the corresponding k eigenvectors are respectively used as column vectors to form an eigenvector matrix.
5) The data is transformed into a new space constructed by k feature vectors.
In a traditional recurrent neural network, RNN is a special neural network structure, which is proposed from the viewpoint of "human cognition is based on past experience and memory"; it differs from DNN, CNN by: it not only takes into account the input from the previous moment, but also gives the network a kind of "memory" function for the previous content. The RNN is called a recurrent neural network, i.e., the current output of a sequence is also related to the previous output. The concrete expression is that the network memorizes the previous information and applies the previous information to the calculation of the current output, namely, the nodes between the hidden layers are not connected any more but connected, and the input of the hidden layer comprises not only the output of the input layer but also the output of the hidden layer at the last moment.
In summary, the problems of the prior art are as follows:
(1) the existing principal component analysis method and the traditional recurrent neural network only use a few two or three of the principal component analysis method and the traditional recurrent neural network for prediction, and the rest indexes are ignored, so that the data result is inaccurate.
(2) If the traditional RNN algorithm is optimized by using a gradient descent method, the serious problem of 'gradient disappearance' or 'gradient explosion' can occur, and the accuracy rate of data is low.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides a circulating neural network stock market data analysis method based on dimension reduction technology optimization.
The invention is realized in such a way that a stock market data analysis method of a recurrent neural network based on dimension reduction technology optimization comprises the following steps:
firstly, taking a factor analysis method for a data set to carry out factor analysis on an R-type factor model, and respectively taking out a first class and a second class of quality factors of the data set; the data in the data set are arranged according to time, the shortest 15min is a unit, 5 basic variables are open, high, low, close and volume respectively, and the five basic variables are independent from each other and are marked as X ═ X1，x2，...xN}; several derived quantities derived from 5 basic variables, denoted as Y ═ Y1，y2，...yN}；
The R-type factor model is as follows:
X＝AF+ε；
Figure BSA0000193651000000031
in the formula, A is a factor load matrix, F is a common factor, epsilon is a special factor, the factor load matrix is calculated by using a main factor estimation method, and the factor estimation method is described as follows:
the covariance matrix of the random vector X is sigma, lambda1≥λ2≥...≥λpCharacteristic root, u > 0 ∑1，u2，...，upFor the corresponding orthonormal feature vector, the spectral decomposition of Σ is:
Figure BSA0000193651000000032
Figure BSA0000193651000000033
Figure BSA0000193651000000034
factor load aijDenotes xiDependent on FjThe greater the value of (a), the greater the degree of dependence; respectively carrying out factor analysis on the X and Y data sets, and respectively taking a first class and a second class of the factors; respectively finding out two parameters, and analyzing the relation of the two parameters to stock rise and fall along with the change of time;
secondly, finding out the parameters and analyzing the relation of the parameters to the stock rise and fall along with the change of time;
thirdly, substituting the parameters into the LSTM model for prediction, adding a forgetting gate to each excitation source of the optimized LSTM, and screening previous information;
the LSTM model comprises:
(1) forget gate: choose to forget some information in the past:
Figure BSA0000193651000000041
(2) input gate: memorize some information now:
Figure BSA0000193651000000042
Figure BSA0000193651000000043
(3) merge past and present memory:
Figure BSA0000193651000000044
(4) output gate: and (3) outputting:
Figure BSA0000193651000000045
ht＝ot*tanh(Ct)；
the LSTM model is calculated as follows:
g(t)＝Φ(Wgxx(t)+Wghh(t-1)+bg)
i(t)＝σ(Wixx(t)+Wihh(t-1)+bi)
f(t)＝σ(Wfxx(t)+Wfhh(t-1)+bf)
o(t)＝σ(Woxx(t)+Wohh(t-1)+bo)
s(t)＝g(t)e i(i)+s(t-1)e f(t)
h(t)＝s(t)e o(t)。
further, the first step is preceded by:
source data over a period of time is obtained from the RESSET financial research database, the Wind information database, and from the stock market dataset, and converted into a normalized dataset arranged in time.
Further, in the second step, the method for analyzing the relationship between the change of the parameter quantity with time and the stock fluctuation comprises the following steps:
step one, establishing a parameter change-over-time stock rise and fall relation information database, permanently storing the parameter change-over-time stock rise and fall relation information, and allowing the accessed by a business library and a mobile phone APP after networking;
step two, establishing dimension setting; the dimension setting comprises but is not limited to design time, region, stock fluctuation, monitoring object, monitoring type key business dimension;
step three, establishing a data structure and key indexes of a fact database; the function of extracting data from the service library is realized;
establishing a multidimensional analysis set model of the stock fluctuation monitoring data, wherein the multidimensional analysis set model comprises business dimensions, a fact library and the forwarding of data from the business library to the fact library;
step five, browsing the data of the multi-dimensional analysis set: the dimensionality is selected independently according to the established multidimensional analysis set, and various specific statistical reports are combined; processing a multidimensional data warehouse according to a data structure of a stock fluctuation monitoring mode, and forming a summary table of monitoring data according to different parameters; organizing data into a data space;
analyzing and displaying the stock fluctuation monitoring data and the stock fluctuation report data information;
and step seven, deploying the software database at the server side, and dividing the software access side into an entry management side and a statistical analysis side according to different use requirements.
Further, in the sixth step, grouping dimensions are selected autonomously for monitoring information and stock fluctuation information, data summarization is carried out, and a specific statistical report is generated; the monitoring results are transversely compared and analyzed with the data in the same period and the previous period according to the year, month, day and hour, and an analysis chart is generated;
the display implementation method for generating the analysis chart comprises the following steps: the method comprises the steps of setting corresponding parameters for stock fluctuation monitoring and data of a stock fluctuation information database, displaying block distribution characteristics of different stock fluctuations in a gradient mode by using different data values and block diagram colors, and finally generating a stock fluctuation monitoring information result statistical distribution diagram.
Further, in the third step, the composition of each cell of the LSTM model is as follows:
(1) input node gc: as in RNN, accepting the output of the hidden node at the previous time point and the current input as inputs, and then passing through an activation function of tanh;
(2) the input gates ic: the function of controlling input information is realized, the input of the gate is the output of the hidden node at the last time point and the current input, and the activation function is sigmoid;
(3) internal state node sc: the input is the current input filtered by the input gate and the internal state node output of the previous time point;
(4) forgetting to record the door fc: the method has the advantages that the method plays a role in controlling internal state information, the input of a gate is the output of a hidden node at the last time point and the current input, and the activation function is sigmoid;
(5) an output gate oc: the gate has the function of controlling output information, the input of the gate is the output of the hidden node at the last time point and the current input, and the activation function is sigmoid.
Another object of the present invention is to provide a system for analyzing stock market data of a recurrent neural network based on dimensionality reduction technology optimization, the system comprising:
a data acquisition module: the system comprises a data processing module, a data processing module and a data processing module, wherein the data processing module is used for acquiring source data of a past period from a RESSET financial research database, a Wind information database and a stock market data set and processing the source data into a standardized data set arranged according to time;
a data preprocessing module: the method is used for carrying out development analysis in the tune on the obtained data set to obtain more than 30 indexes in the financial field; the source data is a two-dimensional matrix between transaction time and transaction data formed by parameters related to opening price, closing price, highest price, lowest price and transaction amount;
the factor analysis module is used for carrying out orthogonal transformation on the two-dimensional matrix obtained by the data preprocessing module and converting the two-dimensional matrix into a group of new characteristics which have different contribution degrees but are linearly uncorrelated, namely principal component analysis results;
the parameter variable acquisition module is used for establishing a database or a variable-capacity multidimensional array, downloading financial data of a new time node from a financial data website in real time in a networking mode, and grouping training sets of the data according to the principle of meeting the data set of long-line short-line data;
the prediction module is used for analyzing and predicting by adopting an LSTM long-term and short-term neural network;
and the result output module is used for displaying the related data, information and the prediction result by utilizing the display equipment.
Further, the parameter acquiring module includes:
the parameter variable acquisition module can select a specific data set from specific stocks to realize training of the specific stocks, and comprises a database, a dimension setting unit, a fact library data structure and key index construction unit, a stock fluctuation monitoring data multidimensional analysis set model construction unit, a multidimensional analysis set data browsing unit, a report information analysis and display unit and a software access end;
the database is used for permanently storing the stock rising and falling relation information of the parameter change along with the time and allowing the accessed by the business library and the mobile phone APP after the networking;
the dimension setting unit is used for designing time, area, stock fluctuation, monitoring object and monitoring type key business dimensions;
the system comprises a fact library data structure and key index construction unit, a business library data extraction unit and a database management unit, wherein the fact library data structure and key index construction unit is used for realizing the function of extracting data from a business library;
the stock fluctuation monitoring data multidimensional analysis set model establishing unit is used for establishing business dimensions, a fact library and forwarding data from the business library to the fact library;
the multidimensional analysis set data browsing unit is used for autonomously selecting dimensions according to the established multidimensional analysis set and combining various specific statistical reports; processing a multidimensional data warehouse according to a data structure of a stock fluctuation monitoring mode, and forming a summary table of monitoring data according to different parameters; organizing data into a data space;
the report information analysis and display unit is used for analyzing and displaying the stock fluctuation monitoring data and the stock fluctuation report data information;
and the software access end is used for deploying the software database at the server end and is divided into an input management end and a statistical analysis end according to different use requirements.
Further, the prediction module specifically includes:
the LSTM long-short term neural network takes the feature vector of each time node as input data, along with the advance of time, a forgetting gate of the LSTM is matched with the screening of an activation function sigmoid, the part with the large influence factor on the result is selectively transmitted into a next-stage input gate, the data of each time node is combined with all the previous input data to influence the total output, along with the advance of the time nodes, the forgetting degree of secondary information and the influence on the result are reduced at a large rate, the primary information is iterated for multiple times to influence the result, the final output is related to the data of all the previous time nodes, and the prediction is realized.
The invention also aims to provide an information data processing terminal for implementing the stock market data analysis method based on the recurrent neural network optimized by the dimension reduction technology.
Another object of the present invention is to provide a computer-readable storage medium, which includes instructions that, when executed on a computer, cause the computer to execute the stock market data analysis method based on a recurrent neural network optimized by a dimension reduction technique.
In summary, the advantages and positive effects of the invention are: the invention expands the LSTM neural network in the financial field, successfully applies the concept of forgetting to stock market analysis and improves the accuracy; and introducing a dimensionality reduction algorithm in data and processing and comparing. The invention highlights the advantages of the dimensionality reduction technology and the accuracy of the LSTM network, so that the stock market prediction is more credible than the traditional analysis method; prediction can be applied in practice.
The invention analyzes the relationship between the change of the parameter variable along with the time and the rise and fall of the stock, establishes a database of the relationship information between the change of the parameter variable along with the time and the rise and fall of the stock, permanently stores the relationship information between the change of the parameter variable along with the time and allows the access by a business base and a mobile phone APP after the networking; establishing dimension setting; establishing a data structure and key indexes of a fact library; the function of extracting data from the service library is realized; establishing a multidimensional analysis set model of stock fluctuation monitoring data, wherein the multidimensional analysis set model comprises business dimensions, a case library and data forwarding from the business library to the case library; browsing the data of the multi-dimensional analysis set; analyzing and displaying stock fluctuation monitoring data and stock fluctuation report data information; the software database is deployed at a server side, and a software access side is divided into an entry management side and a statistical analysis side according to different use requirements. The intelligent display and the data sharing with the APP of the user can be realized.
Drawings
Fig. 1 is a flowchart of a stock market data analysis method based on a recurrent neural network optimized by a dimension reduction technique according to an embodiment of the present invention.
FIG. 2 is a flowchart of a method for analyzing the relationship between the time variation of the parameters and the rise and fall of the stocks according to the embodiment of the present invention.
FIG. 3 is a schematic structural diagram of a stock market data analysis system based on a recurrent neural network optimized by a dimension reduction technique according to an embodiment of the present invention;
in the figure: 1. a data acquisition module; 2. a data preprocessing module; 3. a factor analysis module; 4. a parameter acquiring module; 5. a prediction module; 6. and a result output module.
FIG. 4 is a schematic diagram of a parameter obtaining module according to an embodiment of the present invention;
in the figure: 7. a database; 8. a dimension setting module; 9. a fact library data structure and key index construction module; 10. a multidimensional analysis set model building module of stock fluctuation monitoring data; 11. a multidimensional analysis set data browsing module; 12. a report information analysis and display module; 13. and a software access terminal.
Fig. 5 is a schematic diagram of a prediction result provided by the embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail with reference to the following embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
Aiming at the problems in the prior art, the invention provides a stock market data analysis method of a recurrent neural network based on dimension reduction technology optimization, and the invention is described in detail below with reference to the accompanying drawings.
As shown in fig. 1, the stock market data analysis method based on the recurrent neural network optimized by the dimension reduction technology provided by the embodiment of the present invention includes the following steps:
s101, obtaining source data of a past period of time from a RESSET financial research database, a Wind information database and a stock market data set, and converting the source data into a standardized data set arranged according to time.
S102, respectively obtaining factor analysis method for data setPerforming factor analysis on the R-type factor model, and respectively extracting a first class and a second class of the quality factors of the data set; the data in the data set are arranged according to time, the shortest 15min is a unit, 5 basic variables are open, high, low, close and volume respectively, and the five basic variables are independent from each other and are marked as X ═ X1，x2，...xN}; several derived quantities derived from 5 basic variables, denoted as Y ═ Y1，y2，...yN}。
The R-type factor model is as follows:
X＝AF+ε；
Figure BSA0000193651000000091
in the formula, A is a factor load matrix, F is a common factor, epsilon is a special factor, the factor load matrix is calculated by using a main factor estimation method, and the factor estimation method is described as follows:
the covariance matrix of the random vector X is sigma, lambda1≥λ2≥...≥λpCharacteristic root, u > 0 ∑1，u2，...，upFor the corresponding orthonormal feature vector, the spectral decomposition of Σ is:
Figure BSA0000193651000000101
Figure BSA0000193651000000102
Figure BSA0000193651000000103
factor load aijDenotes xiDependent on FjThe greater the value of (a), the greater the degree of dependence; respectively carrying out factor analysis on the X and Y data sets, and respectively taking a first class and a second class of the factors; and respectively finding two parameters and analyzing the relation of the changes of the two parameters along with time on the stock rise and fall.
S103, finding out the parameter and analyzing the relation of the parameter changing along with time to stock rise and fall.
And S104, substituting the parameters into the LSTM model for prediction, adding a forgetting gate to each excitation source of the optimized LSTM, and screening the previous information.
The LSTM model comprises:
(1) forget gate: choose to forget some information in the past:
Figure BSA0000193651000000104
(2) input gate: memorize some information now:
Figure BSA0000193651000000105
Figure BSA0000193651000000106
(3) merge past and present memory:
Figure BSA0000193651000000107
(4) output gate: and (3) outputting:
ht＝ot*tanh(Ct)。
the LSTM model is calculated as follows:
g(t)＝Φ(Wgxx(t)+Wghh(t-1)+bg)
i(t)＝σ(Wixx(t)+Wihh(t-1)+bi)
f(t)＝σ(Wfxx(t)+Wfhh(t-1)+bf)
o(t)＝σ(Woxx(t)+Wohh(t-1)+bo)
s(t)＝g(t)e i(t)+s(t-1)e f(t)
h(t)＝s(t)e o(t)。
as shown in fig. 2, in the second step, the method for analyzing the relationship between the change of the parameter with time and the stock fluctuation provided by the embodiment of the present invention includes:
s201, establishing a database of the stock rising and falling relation information of the parameter change along with the time, permanently storing the stock rising and falling relation information of the parameter change along with the time, and allowing the accessed by a business base and a mobile phone APP after networking.
S202, establishing dimension setting; the dimension settings include, but are not limited to, design time, area, stock fluctuation, monitoring objects, monitoring type key business dimensions.
S203, establishing a data structure and key indexes of a fact database; and the function of extracting data from the service library is realized.
S204, establishing a stock fluctuation monitoring data multidimensional analysis set model, including business dimension, a fact library and data forwarding from the business library to the fact library.
S205, browsing the data of the multi-dimensional analysis set: the dimensionality is selected independently according to the established multidimensional analysis set, and various specific statistical reports are combined; processing a multidimensional data warehouse according to a data structure of a stock fluctuation monitoring mode, and forming a summary table of monitoring data according to different parameters; data is organized into a data space.
And S206, analyzing and displaying the stock fluctuation monitoring data and the stock fluctuation report data information.
And S207, deploying the software database at the server side, and dividing the software access side into an entry management side and a statistical analysis side according to different use requirements.
In step S206, the embodiment of the present invention performs data summarization and generates a specific statistical report by autonomously selecting grouping dimensions of the monitoring information and the stock fluctuation information; and performing transverse comparison analysis on the monitoring results according to year, month, day, time and contemporaneous and previous data and generating an analysis chart.
The display implementation method for generating the analysis chart comprises the following steps: the method comprises the steps of setting corresponding parameters for stock fluctuation monitoring and data of a stock fluctuation information database, displaying block distribution characteristics of different stock fluctuations in a gradient mode by using different data values and block diagram colors, and finally generating a stock fluctuation monitoring information result statistical distribution diagram.
In the third step, the composition of each cell of the LSTM model provided in the embodiment of the present invention is as follows:
(1) input node gc: as in RNN, the output of the hidden node at the last time point and the current input are accepted as inputs, and then passed through an activation function of tanh.
(2) The input gates ic: the gate has the function of controlling input information, the input of the gate is the output of the hidden node at the last time point and the current input, and the activation function is sigmoid.
(3) Internal state node sc: the inputs are the current input filtered by the input gate and the internal state node output at the previous point in time.
(4) Forgetting to record the door fc: the method has the function of controlling internal state information, the input of the gate is the output of the hidden node at the last time point and the current input, and the activation function is sigmoid.
(5) An output gate oc: the gate has the function of controlling output information, the input of the gate is the output of the hidden node at the last time point and the current input, and the activation function is sigmoid.
As shown in fig. 3, the stock market data analysis system based on the recurrent neural network optimized by the dimension reduction technique according to the embodiment of the present invention includes:
the data acquisition module 1: for obtaining source data for a past period of time from a RESSET financial research database, a Wind information database, and from a stock market dataset, and processing the source data into a normalized dataset arranged in time.
The data preprocessing module 2: the method is used for carrying out development analysis in the tune on the obtained data set to obtain more than 30 indexes in the financial field; the source data is a two-dimensional matrix between transaction time and transaction data formed by parameters related to opening price, closing price, highest price, lowest price and transaction amount.
Factor analysis module 3: the method is used for performing orthogonal transformation on the two-dimensional matrix obtained by the data preprocessing module and converting the two-dimensional matrix into a group of new characteristics which have different contribution degrees but are linearly uncorrelated, namely principal component analysis results.
The parameter quantity obtaining module 4: the method comprises the steps of establishing a database or a variable-capacity multidimensional array, downloading financial data of a new time node from a financial data website in real time in a networking mode, and grouping training sets of the data according to the principle of meeting data sets of long lines and short lines.
The prediction module 5: the method is used for analyzing and predicting by adopting the LSTM long-term and short-term neural network.
The result output module 6: and the display device is used for displaying related data, information and prediction results.
As shown in fig. 4, the parameter obtaining module 4 provided in the embodiment of the present invention includes:
the parameter acquiring module 4 can select a specific data set from specific stocks to realize training of the specific stocks, and comprises a database 7, a dimension setting unit 8, a fact library data structure and key index constructing unit 9, a stock fluctuation monitoring data multidimensional analysis set model establishing unit 10, a multidimensional analysis set data browsing unit 11, a report information analyzing and displaying unit 12 and a software access terminal 13.
And the database 7 is used for permanently storing the stock rise and fall relation information of the parameter change along with the time change and allowing the accessed stock rise and fall relation information to be accessed by the service library and the mobile phone APP after networking.
And the dimension setting unit 8 is used for designing time, area, stock fluctuation, monitoring object and monitoring type key business dimensions.
And the fact library data structure and key index construction unit 9 is used for realizing the function of extracting data from the business library.
The multidimensional analysis set model building unit 10 of the stock fluctuation monitoring data is used for building business dimensions, a fact base and forwarding data from the business base to the fact base.
A multidimensional analysis set data browsing unit 11, configured to autonomously select dimensions according to the established multidimensional analysis set, and combine various specific statistical reports; processing a multidimensional data warehouse according to a data structure of a stock fluctuation monitoring mode, and forming a summary table of monitoring data according to different parameters; data is organized into a data space.
And the report information analysis and display unit 12 is used for analyzing and displaying the stock fluctuation monitoring data and the stock fluctuation report data information.
And the software access end 13 is used for deploying the software database at the server end and dividing the software database into an entry management end and a statistical analysis end according to different use requirements.
The prediction module 5 provided in the embodiment of the present invention specifically includes:
the LSTM long-short term neural network takes the feature vector of each time node as input data, along with the advance of time, a forgetting gate of the LSTM is matched with the screening of an activation function sigmoid, the part with the large influence factor on the result is selectively transmitted into a next-stage input gate, the data of each time node is combined with all the previous input data to influence the total output, along with the advance of the time nodes, the forgetting degree of secondary information and the influence on the result are reduced at a large rate, the primary information is iterated for multiple times to influence the result, the final output is related to the data of all the previous time nodes, and the prediction is realized.
The technical solution of the present invention is further described with reference to the following specific examples.
Embodiments of the present invention search for available data sets on a network. The turn new wave finance and extreme width quantization provide a Tick data set and date line data for use. The data in the data set are arranged according to time, and the shortest time is 15 min. The 5 basic variables are open, high, low, close, volume, respectively. The five basic variables are independent of each other and are key factors influencing the stock trend. Is expressed as X ═ X1，x2，...xN}. In addition, there are several derived quantities derived from these 5 basic variables, denoted as Y ═ Y1，y2，...yN}. Since the Y parameter and the X parameter are not independent from each other, the dimension reduction method cannot be directly utilized. Therefore, the present invention discusses these two data sets separately. Meanwhile, in order to avoid the defects of the principal component analysis method, the factor analysis method is adopted for analysis. Factor analysis is also a technique for reducing dimensions and simplifying data. By studying the internal dependencies among a multitude of variables, a few "abstract" variables are used to represent their underlying data structure. These several abstract variables are called "factors" and reflect the primary information of the original many variables. The original variables are observable variables, while the factors are generally non-observable latent variables.
The R-type factor model is as follows:
X＝AF+ε。
Figure BSA0000193651000000141
in the formula, A is a factor load matrix, F is a common factor, and epsilon is a special factor. The invention calculates the factor load matrix by using a main factor estimation method, which is described as follows:
let the covariance matrix of the random vector X be sigma, lambda1≥λ2≥...≥λpCharacteristic root, u > 0 ∑1，u2，...，upFor the corresponding orthonormal feature vector, the spectral decomposition of Σ is:
Figure BSA0000193651000000152
Figure BSA0000193651000000153
factor load aijDenotes xiDependent on FjDegree of (d), the greater the value, the degree of dependenceThe greater the degree. The invention carries out factor analysis on the X and Y data sets respectively, and respectively takes out the first class and the second class of the quality factors.
The invention respectively finds out two parameters and analyzes the relation of the two parameters to stock rise and fall along with the change of time.
The invention substitutes the two found parameters into the LSTM model for prediction. The optimized LSTM incorporates a forgetting gate at each stimulus to better screen previous information that is closely related to the time series for the stock due to this memory.
LSTM principle of operation:
forget gate: choosing to forget certain information
Input gate: memorize some current information
Figure BSA0000193651000000155
3. Merge past and present memory:
Figure BSA0000193651000000157
output gate 4: and (3) outputting:
Figure BSA0000193651000000161
ht＝ot*tanh(Ct)。
LSTM model derivation, the composition of each cell is as follows:
(1) input node (gc): as in RNN, the output of the hidden node at the last time point and the current input are accepted as inputs, and then passed through an activation function of tanh.
(2) Input gate (ic): the input of the gate is the output of the hidden node at the last time point and the current input, and the activation function is sigmoid (the reason is that the output of the sigmoid is between 0 and 1, and the multiplication of the output of the input gate and the output of the input node can play a role in controlling the information quantity).
(3) Internal state node (sc): the inputs are the current input filtered by the input gate and the internal state node output at the previous point in time.
(4) Forget to record gate (fc): the gate input is the output of the hidden node at the last time point and the current input, and the activation function is sigmoid (the reason is that the output of the sigmoid is between 0 and 1, and the multiplication of the output of the internal state node and the output of the forgotten gate can play a role in controlling the information quantity).
(5) Output gate (oc): the gate input is the output of the hidden node at the last time point and the current input, and the activation function is sigmoid (the reason is that the output of the sigmoid is between 0 and 1, and the multiplication of the output gate and the output of the internal state node can play a role in controlling the information quantity).
The calculation of the LSTM layer can be expressed as follows:
g(t)＝Φ(Wgxx(t)+Wghh(t-1)+bg)
i(t)＝σ(Wixx(t)+Wihh(t-1)+bi)
f(t)＝σ(Wfxx(t)+Wfhh(t-1)+bf)
o(t)＝σ(Woxx(t)+Wohh(t-1)+bo)
s(t)＝g(t)e i(t)+s(t-1)e f(t)
h(t)＝s(t)e o(t)。
fig. 5 is a schematic diagram of the prediction results provided by the embodiment of the present invention.
The technical effects of the present invention will be described in detail with reference to the tests below.
The present invention obtains the prediction by analyzing a stock of stock A in a certain time period as shown in FIG. 4. The prediction method of the invention highlights the advantages of the dimensionality reduction technology and the accuracy of the LSTM network, so that the stock market prediction is more credible than the traditional analysis method and can be applied to the practice.
The working principle of the invention is as follows:
firstly, a factor analysis module acquires source data of a past period (real-time replacement) from a stock market data set, namely a two-dimensional matrix between trading time and transaction data (opening price, closing price, highest price, lowest price and volume of transaction), performs expansion analysis in tushare on the acquired data to acquire more than 30 indexes in the financial field, converts the indexes into a group of new characteristics which are different in result contribution degree but linearly unrelated through orthogonal transformation on the matrix, is called as a result of principal component analysis, is called as a factor analysis module, and is used for preprocessing the data; secondly, the parameter variable acquisition module establishes a database or a variable-capacity multidimensional array, downloads financial data of a new time node from a financial data website in real time in a networking mode, and performs training set grouping on the data to meet the requirement of distinguishing data sets of long lines and short lines. The data set content can be viewed through other visualization ways; finally, the prediction module analyzes by adopting an LSTM long-short term neural network, the characteristic vector of each time node is used as input data according to the principle, along with the advancing of time, a forgetting gate of the LSTM is matched with the screening of an activation function sigmoid, the part with large influence factors on the result is selectively transmitted into a next-stage input gate, the steps are repeated, the data of each time node is combined with all the previous input data to influence the total output, along with the advancing of the time nodes, the forgetting degree of the secondary information and the influence on the result are reduced at a large rate, the primary information is iterated for multiple times to influence the result, and the final output is related to the data of all the previous time nodes, so that the prediction is realized.
Meanwhile, other visualization and interaction modules comprise a dimension setting module, a fact database data structure and key index building module and a multidimensional analysis set data browsing module, training of a user on specific stocks is achieved by establishing an App, the specific stocks select specific data sets, a relation between the data sets in the parameter acquisition module and user requirements is established, and the predicted results can be conveniently viewed.
It should be noted that the embodiments of the present invention can be realized by hardware, software, or a combination of software and hardware. The hardware portion may be implemented using dedicated logic; the software portions may be stored in a memory and executed by a suitable instruction execution system, such as a microprocessor or specially designed hardware. Those skilled in the art will appreciate that the apparatus and methods described above may be implemented using computer executable instructions and/or embodied in processor control code, such code being provided on a carrier medium such as a disk, CD-or DVD-ROM, programmable memory such as read only memory (firmware), or a data carrier such as an optical or electronic signal carrier, for example. The apparatus and its modules of the present invention may be implemented by hardware circuits such as very large scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., or by software executed by various types of processors, or by a combination of hardware circuits and software, e.g., firmware.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.

Claims (10)

1. A stock market data analysis method of a recurrent neural network based on dimension reduction technology optimization is characterized by comprising the following steps of:
first, the data sets are separately processedFirstly, taking a factor analysis method to carry out factor analysis on the R-type factor model, and respectively taking out a first type and a second type of the data set quality factors; the data in the data set are arranged according to time, the shortest 15min is a unit, 5 basic variables are open, high, low, close and volume respectively, and the five basic variables are independent from each other and are marked as X ═ X1，x2，…xN}; several derived quantities derived from 5 basic variables, denoted as Y ═ Y1，y2，…yN}；
The R-type factor model is as follows:
X＝AF+ε；
Figure FSA0000193650990000011
in the formula, A is a factor load matrix, F is a common factor, epsilon is a special factor, the factor load matrix is calculated by using a main factor estimation method, and the factor estimation method is described as follows:
the covariance matrix of the random vector X is sigma, lambda1≥λ2≥...≥λpCharacteristic root, u > 0 ∑1，u2，...，upFor the corresponding orthonormal feature vector, the spectral decomposition of Σ is:
Figure FSA0000193650990000012
Figure FSA0000193650990000014
factor load aijDenotes xiDependent on FjThe greater the value of (a), the greater the degree of dependence; respectively carrying out factor analysis on the X and Y data sets, and respectively taking a first class and a second class of the factors; respectively finding out two parameters and analyzing them over timeThe relationship of changes to the rise and fall of stocks;
secondly, finding out the parameters and analyzing the relation of the parameters to the stock rise and fall along with the change of time;
thirdly, substituting the parameters into the LSTM model for prediction, adding a forgetting gate to each excitation source of the optimized LSTM, and screening previous information;
the LSTM model comprises:
(1) forget gate: choose to forget some information in the past:
Figure FSA0000193650990000021
(2) input gate: memorize some information now:
Figure FSA0000193650990000022
Figure FSA0000193650990000023
(3) merge past and present memory:
(4) output gate: and (3) outputting:
Figure FSA0000193650990000025
ht＝ot*tanh(Ct)；
the LSTM model is calculated as follows:
g(t)＝Φ(Wgxx(t)+Wghh(t-1)+bg)
i(t)＝σ(Wixx(t)+Wihh(t-1)+bi)
f(t)＝σ(Wfxx(t)+Wfhh(t-1)+bf)
o(t)＝σ(Woxx(t)+Wohh(t-1)+bo)
s(t)＝g(t)e i(i)+s(t-1)e f(t)
h(t)＝s(t)e o(t)。
2. the stock market data analysis method based on the dimensionality reduction technology-optimized recurrent neural network of claim 1, wherein the first step is further preceded by:
source data over a period of time is obtained from the RESSET financial research database, the Wind information database, and from the stock market dataset, and converted into a normalized dataset arranged in time.
3. The method for analyzing stock market data based on recurrent neural network optimized by dimension reduction technique as claimed in claim 1, wherein in the second step, the method for analyzing the relationship between the change of the parameter with time and the stock fluctuation comprises:
step one, establishing a parameter change-over-time stock rise and fall relation information database, permanently storing the parameter change-over-time stock rise and fall relation information, and allowing the accessed by a business library and a mobile phone APP after networking;
step two, establishing dimension setting; the dimension setting comprises but is not limited to design time, region, stock fluctuation, monitoring object, monitoring type key business dimension;
step three, establishing a data structure and key indexes of a fact database; the function of extracting data from the service library is realized;
establishing a multidimensional analysis set model of the stock fluctuation monitoring data, wherein the multidimensional analysis set model comprises business dimensions, a fact library and the forwarding of data from the business library to the fact library;
step five, browsing the data of the multi-dimensional analysis set: the dimensionality is selected independently according to the established multidimensional analysis set, and various specific statistical reports are combined; processing a multidimensional data warehouse according to a data structure of a stock fluctuation monitoring mode, and forming a summary table of monitoring data according to different parameters; organizing data into a data space;
analyzing and displaying the stock fluctuation monitoring data and the stock fluctuation report data information;
and step seven, deploying the software database at the server side, and dividing the software access side into an entry management side and a statistical analysis side according to different use requirements.
4. The method for analyzing stock market data based on recurrent neural network optimized by dimension reduction technique as claimed in claim 1, wherein in step six, the group dimensions are selected autonomously for monitoring information and stock fluctuation information, so as to summarize data and generate a specific statistical report; the monitoring results are transversely compared and analyzed with the data in the same period and the previous period according to the year, month, day and hour, and an analysis chart is generated;
the display implementation method for generating the analysis chart comprises the following steps: the method comprises the steps of setting corresponding parameters for stock fluctuation monitoring and data of a stock fluctuation information database, displaying block distribution characteristics of different stock fluctuations in a gradient mode by using different data values and block diagram colors, and finally generating a stock fluctuation monitoring information result statistical distribution diagram.
5. The stock market data analysis method of recurrent neural networks based on dimension reduction technique optimization as claimed in claim 1, wherein in the third step, the composition of each cell of the LSTM model is as follows:
(1) input node gc: as in RNN, accepting the output of the hidden node at the previous time point and the current input as inputs, and then passing through an activation function of tanh;
(2) the input gates ic: the function of controlling input information is realized, the input of the gate is the output of the hidden node at the last time point and the current input, and the activation function is sigmoid;
(3) internal state node sc: the input is the current input filtered by the input gate and the internal state node output of the previous time point;
(4) forgetting to record the door fc: the method has the advantages that the method plays a role in controlling internal state information, the input of a gate is the output of a hidden node at the last time point and the current input, and the activation function is sigmoid;
(5) an output gate oc: the gate has the function of controlling output information, the input of the gate is the output of the hidden node at the last time point and the current input, and the activation function is sigmoid.
6. A stock market data analysis system of the recurrent neural network based on dimension reduction technology optimization, which implements the stock market data analysis method of the recurrent neural network based on dimension reduction technology optimization according to claim 1, wherein the stock market data analysis system of the recurrent neural network based on dimension reduction technology optimization comprises:
a data acquisition module: the system comprises a data processing module, a data processing module and a data processing module, wherein the data processing module is used for acquiring source data of a past period from a RESSET financial research database, a Wind information database and a stock market data set and processing the source data into a standardized data set arranged according to time;
a data preprocessing module: the method is used for carrying out development analysis in the tune on the obtained data set to obtain more than 30 indexes in the financial field; the source data is a two-dimensional matrix between transaction time and transaction data formed by parameters related to opening price, closing price, highest price, lowest price and transaction amount;
the factor analysis module is used for carrying out orthogonal transformation on the two-dimensional matrix obtained by the data preprocessing module and converting the two-dimensional matrix into a group of new characteristics which have different contribution degrees but are linearly uncorrelated, namely principal component analysis results;
the parameter variable acquisition module is used for establishing a database or a variable-capacity multidimensional array, downloading financial data of a new time node from a financial data website in real time in a networking mode, and grouping training sets of the data according to the principle of meeting the data set of long-line short-line data;
the prediction module is used for analyzing and predicting by adopting an LSTM long-term and short-term neural network;
and the result output module is used for displaying the related data, information and the prediction result by utilizing the display equipment.
7. The dimensionality reduction technology optimized recurrent neural network-based stock market data analysis system of claim 6, wherein the parameter acquisition module comprises:
the parameter variable acquisition module can select a specific data set from specific stocks to realize training of the specific stocks, and comprises a database, a dimension setting unit, a fact library data structure and key index construction unit, a stock fluctuation monitoring data multidimensional analysis set model construction unit, a multidimensional analysis set data browsing unit, a report information analysis and display unit and a software access end;
the database is used for permanently storing the stock rising and falling relation information of the parameter change along with the time and allowing the accessed by the business library and the mobile phone APP after the networking;
the dimension setting unit is used for designing time, area, stock fluctuation, monitoring object and monitoring type key business dimensions;
the system comprises a fact library data structure and key index construction unit, a business library data extraction unit and a database management unit, wherein the fact library data structure and key index construction unit is used for realizing the function of extracting data from a business library;
the stock fluctuation monitoring data multidimensional analysis set model establishing unit is used for establishing business dimensions, a fact library and forwarding data from the business library to the fact library;
the multidimensional analysis set data browsing unit is used for autonomously selecting dimensions according to the established multidimensional analysis set and combining various specific statistical reports; processing a multidimensional data warehouse according to a data structure of a stock fluctuation monitoring mode, and forming a summary table of monitoring data according to different parameters; organizing data into a data space;
the report information analysis and display unit is used for analyzing and displaying the stock fluctuation monitoring data and the stock fluctuation report data information;
and the software access end is used for deploying the software database at the server end and is divided into an input management end and a statistical analysis end according to different use requirements.
8. The dimensionality reduction technology optimization-based stock market data analysis system of the recurrent neural network of claim 6, wherein the prediction module specifically comprises:
the LSTM long-short term neural network takes the feature vector of each time node as input data, along with the advance of time, a forgetting gate of the LSTM is matched with the screening of an activation function sigmoid, the part with the large influence factor on the result is selectively transmitted into a next-stage input gate, the data of each time node is combined with all the previous input data to influence the total output, along with the advance of the time nodes, the forgetting degree of secondary information and the influence on the result are reduced at a large rate, the primary information is iterated for multiple times to influence the result, the final output is related to the data of all the previous time nodes, and the prediction is realized.
9. An information data processing terminal for realizing the stock market data analysis method based on the recurrent neural network optimized by the dimension reduction technology according to any one of claims 1 to 6.
10. A computer-readable storage medium comprising instructions that, when executed on a computer, cause the computer to perform the method for stock market data analysis based on a recurrent neural network optimized by dimension reduction techniques according to any one of claims 1 to 6.
CN201911057108.3A 2019-10-24 2019-10-24 Stock market data analysis method of recurrent neural network based on dimension reduction technology optimization Active CN110827148B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911057108.3A CN110827148B (en) 2019-10-24 2019-10-24 Stock market data analysis method of recurrent neural network based on dimension reduction technology optimization

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911057108.3A CN110827148B (en) 2019-10-24 2019-10-24 Stock market data analysis method of recurrent neural network based on dimension reduction technology optimization

Publications (2)

Publication Number Publication Date
CN110827148A true CN110827148A (en) 2020-02-21
CN110827148B CN110827148B (en) 2020-07-24

Family

ID=69552091

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911057108.3A Active CN110827148B (en) 2019-10-24 2019-10-24 Stock market data analysis method of recurrent neural network based on dimension reduction technology optimization

Country Status (1)

Country Link
CN (1) CN110827148B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111309842A (en) * 2020-03-04 2020-06-19 重庆工程职业技术学院 Special instrument for geographic mapping, mapping method and application

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2014235564A (en) * 2013-06-02 2014-12-15 データ・サイエンティスト株式会社 Evaluation method, evaluation device, and program
CN105787582A (en) * 2015-12-24 2016-07-20 清华大学 Stock risk prediction method and apparatus
CN107480857A (en) * 2017-07-10 2017-12-15 武汉楚鼎信息技术有限公司 One B shareB gene pool diagnostic method and system
CN108154435A (en) * 2017-12-26 2018-06-12 浙江工业大学 A kind of stock index price expectation method based on Recognition with Recurrent Neural Network
AU2018101512A4 (en) * 2018-10-11 2018-11-15 Dong, Xun Miss A comprehensive stock trend predicting method based on neural networks
CN109272406A (en) * 2018-10-25 2019-01-25 安徽磐众信息科技有限公司 Prediction of Stock Price method and its system based on modified LSTM
CN109816140A (en) * 2018-12-12 2019-05-28 哈尔滨工业大学（深圳） Forecasting of Stock Prices method, apparatus, equipment and the storage medium influenced based on cross-market

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2014235564A (en) * 2013-06-02 2014-12-15 データ・サイエンティスト株式会社 Evaluation method, evaluation device, and program
CN105787582A (en) * 2015-12-24 2016-07-20 清华大学 Stock risk prediction method and apparatus
CN107480857A (en) * 2017-07-10 2017-12-15 武汉楚鼎信息技术有限公司 One B shareB gene pool diagnostic method and system
CN108154435A (en) * 2017-12-26 2018-06-12 浙江工业大学 A kind of stock index price expectation method based on Recognition with Recurrent Neural Network
AU2018101512A4 (en) * 2018-10-11 2018-11-15 Dong, Xun Miss A comprehensive stock trend predicting method based on neural networks
CN109272406A (en) * 2018-10-25 2019-01-25 安徽磐众信息科技有限公司 Prediction of Stock Price method and its system based on modified LSTM
CN109816140A (en) * 2018-12-12 2019-05-28 哈尔滨工业大学（深圳） Forecasting of Stock Prices method, apparatus, equipment and the storage medium influenced based on cross-market

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
李科嘉: "《 基于长短期记忆模型的期货价格预测》", 《中国优秀硕士学位论文全文数据库 经济与管理科学辑》 *
裴大卫 等: "《基于多因子与多变量长短期记忆网络的股票价格预测》", 《计算机系统应用》 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111309842A (en) * 2020-03-04 2020-06-19 重庆工程职业技术学院 Special instrument for geographic mapping, mapping method and application

Also Published As

Publication number Publication date
CN110827148B (en) 2020-07-24

Similar Documents

Publication Publication Date Title
Ramírez-Gallego et al. An information theory-based feature selection framework for big data under apache spark
Maliar et al. Numerical methods for large-scale dynamic economic models
Cerqueira et al. Vest: Automatic feature engineering for forecasting
Tank et al. Granger causality networks for categorical time series
Hosszejni et al. Approaches toward the Bayesian estimation of the stochastic volatility model with leverage
Kurbalija et al. A framework for time-series analysis
Hull Machine learning for economics and finance in tensorflow 2
Tepegjozova et al. Nonparametric C-and D-vine-based quantile regression
Zhang et al. Feature relevance term variation for multi-label feature selection
Yue et al. Dynamic network reconstruction from heterogeneous datasets
Fan et al. Penalized quantile regression for distributed big data using the slack variable representation
Qiu et al. Gradient-based sparse principal component analysis with extensions to online learning
Killiches et al. Model distances for vine copulas in high dimensions
CN110827148B (en) Stock market data analysis method of recurrent neural network based on dimension reduction technology optimization
Della Vecchia et al. Online instrumental variable regression: Regret analysis and bandit feedback
He et al. Stationary-Sparse Causality Network Learning.
Park et al. Three iteratively reweighted least squares algorithms for L 1-norm principal component analysis
Xiao et al. A geometric proximal gradient method for sparse least squares regression with probabilistic simplex constraint
Gorantla et al. Socially Fair Center-Based and Linear Subspace Clustering
Gao et al. Multi-label feature selection based on min-relevance label
Sumalatha et al. Rough set based decision rule generation to find behavioural patterns of customers
Ballester-Ripoll et al. High-dimensional scalar function visualization using principal parameterizations
Sun et al. Regularization for stationary multivariate time series
Solaro et al. Algorithmic-type imputation techniques with different data structures: alternative approaches in comparison
Hmamouche et al. Predictors extraction in time series using authorities-hubs ranking

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
GR01 Patent grant
GR01 Patent grant
TR01 Transfer of patent right
TR01 Transfer of patent right

Effective date of registration: 20230104

Address after: 518000 Room 201, room A, general office building of Qianhai Shenzhen Hong Kong cooperation zone, No.1 Qianhai Shenzhen Cooperation District, Qianhai Shenzhen Hong Kong cooperation zone, Shenzhen City, Shenzhen and Hong Kong cooperation zone 1 (to Shenzhen Qianhai business secretary Co., Ltd.)

Patentee after: Shenzhen Chengqi Asset Management Co.,Ltd.

Address before: 710126 South Campus of Xi'an University of Electronic Science and technology, No. 266, Xinglong section, Xifeng Road, Chang'an District, Xi'an City, Shaanxi Province

Patentee before: Song Yatong