CN110047006A - The method and device of portfolio investment decision is carried out based on neural network and machine learning - Google Patents

The method and device of portfolio investment decision is carried out based on neural network and machine learning Download PDF

Info

Publication number
CN110047006A
CN110047006A CN201910321426.XA CN201910321426A CN110047006A CN 110047006 A CN110047006 A CN 110047006A CN 201910321426 A CN201910321426 A CN 201910321426A CN 110047006 A CN110047006 A CN 110047006A
Authority
CN
China
Prior art keywords
model
neural network
decision
investment
data
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
Application number
CN201910321426.XA
Other languages
Chinese (zh)
Inventor
陈硕坚
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.)
Individual
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 CN201910321426.XA priority Critical patent/CN110047006A/en
Publication of CN110047006A publication Critical patent/CN110047006A/en
Pending legal-status Critical Current

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
    • G06N3/044—Recurrent networks, e.g. Hopfield 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
    • G06N3/045—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
    • G06N3/08—Learning methods
    • 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
    • G06Q40/06—Asset management; Financial planning or analysis

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Data Mining & Analysis (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Business, Economics & Management (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Artificial Intelligence (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Finance (AREA)
  • Development Economics (AREA)
  • Accounting & Taxation (AREA)
  • Strategic Management (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Technology Law (AREA)
  • General Business, Economics & Management (AREA)
  • Marketing (AREA)
  • Economics (AREA)
  • Operations Research (AREA)
  • Human Resources & Organizations (AREA)
  • Game Theory and Decision Science (AREA)
  • Financial Or Insurance-Related Operations Such As Payment And Settlement (AREA)

Abstract

The present invention relates to the method and devices that portfolio investment decision is carried out based on neural network and machine learning, the described method includes: the transactional related data of acquisition securities market；Data are pre-processed and are normalized；Build multiple neural network models；It trains and filters out good model；Multiple models are voted to obtain final investment decision after individually making investment decision；The franchise of each model is adjusted according to the Decision Quality of each model；Supplementary training constantly is carried out to each model using new data, to adapt to subsequent decision requirements.The present invention also provides a kind of computing devices for executing the above method.The present invention utilizes neural network and machine learning techniques, realizes the manual intelligent in investment securities field, investor does not need the mathematical knowledge or economy and finance knowledge that have any, does not need any investment experiences, and decision process can be completely to investor's black box.

Description

The method and device of portfolio investment decision is carried out based on neural network and machine learning
Technical field
Applied technical field the present invention relates to artificial intelligence in investment securities field, it is specifically a kind of based on nerve Network and machine learning carry out the method and device of portfolio investment decision.
Background technique
At present in investment securities field, investment decision mainly is made by the mankind.The mankind, which carry out investment decision, following limitation:
1, the mankind are slow to the processing speed of information, can not read rapidly and understand bulk information.
2, inevitably in mood swing, changes of the mood, these situations are easy to bring incorrect decision the mankind.
3, investment decision is relied in personal experience, economy and finance knowledge, mathematical knowledge, the investment decision essence of high quality It is upper to become a few peoples' exclusive privilege.
The remote superman's class of the information processing rate of computer, can be in sufficiently obtaining historical experience in the short time；Computer does not have Emotion is not influenced by emotion and psychological factor.Using artificial intelligence, investment decision is carried out in conjunction with the operational capability of computer, is One technology being well worth doing.And with neural network and machine learning for important technique direction in artificial intelligence.
Obtain at present widely applied, the technology for being comparatively close to computer investment decision is " quantization investment " (Quantitative investment).The essence of quantization investment is to pre-set mathematical model by the mankind, by computer root Mathematical model carries out decision accordingly.This technology needs investment decision people to have deep since decision model is still to be formulated by the mankind Mathematical background, and this model be it is static, will not be evolved automatically according to the variation of market situation, can only go to modify by people Model.
The present invention utilizes neural network and machine learning techniques, realizes the manual intelligent in investment securities field, invests People does not need the mathematical knowledge or economy and finance knowledge that have any, does not need any investment experiences, and decision process can be complete To investor's black box.And artificial intelligence is with the variation of market situation, continuous self-teaching, self adjustment, self evolve with Adapt to the market of variation.
Summary of the invention
The various limitations that investment decision is carried out for the mankind, the present invention provides one kind to be based on neural network and engineering Habit technology carries out the method for portfolio investment decision and the device of adopting said method.
The embodiment of the invention provides a kind of method for carrying out portfolio investment decision based on neural network and machine learning, institutes The method stated the following steps are included:
Step 1: transaction history data, economical operation data, the enterprise operation data of continuous collecting securities market.
Step 2: the data obtained in step 1 are normalized.Before normalized, first to outlier (Outlier) the too far variable in bias data compact district carries out interval mapping, reduces between its outlier and data-intensive district Distance.
Step 3: on computers, with different model structures and different parameters, building multiple based on neural network The model of (Neural Networks).
Step 4: being calculated using the data obtained in step 2 in the method for machine learning (Machine Learning) The multiple models created in training step 3 on machine abandon the high model of error, it is low to leave error.
Step 5: each model screened in step 4, separately to the security product tendency instantly with future Predict and makes investment decision accordingly.
Step 6: the investment decision of oneself is taken out progress " ballot ", comprehensive each model respectively by each model in step 5 Voting results, obtain final investment decision.
Step 7: each to what is screened in step 4 with data updated in steps 1 and 2 in next investment cycle A model carries out supplementary training.Then step 5,6 are repeated, is so moved in endless cycles.
In the method step 2, for the initial data obtained from step 1, if the outlier of certain variables (outlier) bias data compact district is far, will have an adverse effect to data application thereafter, at this time first to these variables into Row logarithm (log) opens the transformation of n times root, data distribution can be made more uniform after transformation.
Normalization in the method step 2 is that all different dimensions, different size of variable are passed through interval mapping, The real number interval of [0,1] (or [- 1,1]) is transformed to, in case the small variable of absolute value is flooded by the big variable of absolute value and can not Embody its feature.
In the method step 3, convolutional neural networks (Convolutional Neural Networks), circulation are utilized Neural network (Recurrent Neural Network), shot and long term memory network (Long Short Term Memory Network), feedforward neural network (feedforward neural network) is joined by different combinations and network Number is built into the model that multiple structures are different, parameter is different.In the method step 4, neuron (Neurons) number of model If measure it is very few will lead to poor fitting, it is excessive and will lead to over-fitting.Therefore the present invention carries out certainly the neuronal quantity of neural network It is dynamic to adjust.If some model poor fitting, automatic re -training after increasing neuronal quantity；If some model over-fitting, subtracts automatically Re -training after few neuronal quantity；Until fitting is good.
In the method step 4, with the different model of the data training of different time span.Several moons are as short as, it is long to number 10 years.In securities market, sometimes historical experience abundant is advantage；But other when, because of market style cataclysm, mistake More historical experiences may instead result in erroneous judgement.At this point, the model of only short-term experience has advantage instead.
In the method step 6, the decision that each model is made is not consistent, need to be voted by each model, with ballot It as a result is final decision.The franchise weight of each model is also inconsistent, and the high model influence power of weight is bigger.And each model Weight is determined by following methods: each decision of each model is recorded, the true tendency phase with securities market thereafter It contrasts, to count the Decision Quality of the model.
The Decision Quality of model can be measured by the prediction error to investment target, and calculation formula is as follows:
Wherein: E predicts error, and it is R that n, which is total periodicity of prediction,iFor the investment target period i true value；FiFor Predicted value of the investment target in period i.
Optionally, the Decision Quality of model can also be measured by investment return, and calculation formula is as follows:
Wherein: G is accumulated earnings, and n is total periodicity of prediction, giFor the income in i-th of period.The high mould of Decision Quality Type improves its ballot weight；The low model of Decision Quality, reduces its ballot weight, until cancelling its franchise.
The embodiment of the invention also provides a kind of device for carrying out portfolio investment decision based on neural network and machine learning, The device includes:
Storage unit, for storing instruction；And an at least processor, it is coupled with the storage unit；Wherein, work as institute When stating an at least processor execution described instruction, it is according to claims 1-8 that described instruction causes the processor to execute Method.
It is read the beneficial effects of the present invention are: mankind investor progress data can be substituted by computer in investment securities field Reading, data analysis and investment decision, can make the investor of financial knowledge scarcity that investment decision can also be effectively performed；People can be overcome The weakness such as class investor information processing speed is slow, meeting is tired, mood easily fluctuates, improve the quality of investment decision, avoid a large amount of Incorrect decision；It enables computer capacity by continuous self-teaching, self evolve, adapt to fast changing market, grow with each passing hour ground The investment decision of high quality is provided；It enables investment no longer be monopolized by a small number of professional persons, lacks investment experiences, mathematical knowledge, economy The ordinary people of financial knowledge can also make efficient investment.
Detailed description of the invention
Fig. 1 is the overall flow figure of the embodiment of the present invention；
Fig. 2 is the flow chart of " each Model Independent decision and ballot " in the embodiment of the present invention；
Fig. 3 is the flow chart for " adjusting model franchise according to accuracy of determination " in the embodiment of the present invention；
Fig. 4 is the function structure chart of device provided in an embodiment of the present invention.
Specific embodiment
Technical solution of the present invention will be clearly and completely described by embodiment below, it is clear that described reality Applying example is only a part of the embodiments of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, this field skill Art personnel every other embodiment obtained without making creative work belongs to the model that the present invention protects It encloses.
It determines as shown in Figure 1, carrying out investment securities based on neural network and machine learning the embodiment of the invention provides one kind The method of plan, the described method comprises the following steps:
Step 101: data acquisition
Transaction history data, enterprise operation data and the economical operation macro-indicators of continuous collecting China A share market. Data all derive from internet public information.The information of acquisition is stored to MySql database, is inquired and is adjusted for down-stream With.
Step 102: data prediction and normalization
The data acquired in inspection step 101, find out outlier (Outlier) the bias data compact district of which variable too Far.These variables are carried out log transformation.The distance between outlier and data-intensive district can significantly shorten after transformation.
Then, all variables are normalized, they is normalized to real number [0,1] section.Normalize formula Are as follows:
Xnorm=(X-Xmin)/(Xmax-Xmin)
Wherein: Xnorm is the value after normalization；X is the value before normalization；Xmax is the very big of the variable before normalizing Value；Xmin is the minimum of the variable before normalizing.
Step 103: building model
On computers, convolutional neural networks (Convolutional Neural Networks), circulation nerve net are utilized Network (Recurrent Neural Network), shot and long term memory network (Long Short Term Memory Network), Feedforward neural network (feedforward neural network) is built into 50 by different combination and network parameter The neural network model that a structure is different, parameter is different.
Step 104: training pattern
The data obtained in step 102, each variable takes 80% as training sample at random, remaining 20% conduct is surveyed Sample sheet.In the method for machine learning, on computers with 50 models created in training sample training step 103, with survey Sample training effect is tried, mean square error (MSE) or the high model of mean absolute error (MAE) is abandoned, leaves the low person of error.
When training, the learning cycle of this 50 models is different.Learning cycle is shortest, it is only allowed to learn nearest 6 The data of the moon；Learning cycle is longest, it is allowed to learn nearest 20 years data.Because sometimes abundant in securities market Historical experience is advantage；But other when because market style cataclysm, excessive historical experience may instead result in erroneous judgement. At this point, the model of only short-term experience has advantage instead.
In training process, if the error of some model cannot synchronize reduction as frequency of training increases, this mould can determine whether Type is poor fitting, increases the neuronal quantity and re -training of the model.If some model increases with frequency of training, training is missed The synchronous decline of difference, but test error goes up not down, and can determine whether this model over-fitting reduces the neuronal quantity of the model and again Training.
It is used in next step finally, retaining 9 that error is minimum in 30 models.
Step 105: each Model Independent decision
As shown in Fig. 2, 9 models, independently the tendency of each stock is predicted daily and selects oneself think Optimal stock portfolio.
Step 106: each model ballot
As shown in Fig. 2, 9 models are voted respectively with the stock portfolio that oneself is selected, but the throwing of each model Ticket power be it is different, before ballot will first apply this model franchise weight.The voting results of comprehensive all models, obtain journey The consequently recommended stock portfolio of sequence.
As shown in figure 3, the franchise of each model is the same, but program will track the throwing of each model when starting Provide the historical yield of decision.Profit calculating formula is as follows:
Wherein: G is accumulated earnings, and n is total periodicity of prediction, giFor the income in i-th of period.According to historical return Height, dynamically adjust the franchise weight of each model, weight is up to 1, minimum 0.The bad model of investment return, When being voted in the future, its influence power will be reduced.
Step 107: supplementary training all is carried out to above-mentioned 9 models with data updated in step 101,102 daily, Then the stock portfolio for recommending next day according to step 105,106 again, so moves in endless cycles.
The embodiment of the invention also provides for realizing the above-mentioned corresponding device of method, device is a computer, packet It includes:
Memory: for storing instruction and data；
Processor: coupling with the storage unit, is used for operating instruction；
Internet connection apparatus: it is used for from internet hunt, download information；
Input-output equipment: for inputting instruction, data, and operation result is exported.
As shown in figure 4, the device includes with lower module:
Data acquisition module 401: transaction history data, enterprise operation data for continuous collecting China A share market, warp Ji operation macro-indicators, and the information of acquisition is stored to MySql database data and pre-processes and normalizes module 402: it is used for The variable too far to outlier bias data compact district carries out log transformation, is then normalized；
Neural network model builds module 403: for building 50 different neural network models of structure difference, parameter；
Neural network model is trained with screening module 404: for training neural network model, eliminating the high mould of error Type leaves the low model of error；
Each Model Independent decision-making module 405: for enabling each neural network model independently make investment decision；Each model is thrown Ticket module 406: for ballot weight to be applied to each model, then allow each model that their own investment decision is taken to throw Ticket obtains final investment decision；
Each model decision quality assessment modules 407: for assessing the investment decision quality of each model, and it is modified accordingly Franchise weight；
Each model continues training module 408: training each model for continuing with newest data, prepares in next investment Decision is carried out in period；
Output module 409: for exporting investment decision.
Effect: using this method and device, carries out the simulation investment under simulated conditions, investment cycle to Chinese A share market From the beginning of 2010 in April, 2019,11 times of the appreciation of fixed assets, and same period Index of Shanghai Stock Exchange amount of increase is 0.
The present invention realize investment securities field manual intelligent, can bring it is below the utility model has the advantages that
1, overcome the mankind in investment decision process, information processing rate is slow, mood swing, changes of the mood etc. it is unfavorable because Element avoids many investments from making mistakes.
2, it enables computer capacity by continuous self-teaching, self evolve, adapt to fast changing market, grow with each passing hour ground The investment decision of high quality is provided.
3, investment is enabled no longer to be monopolized by a small number of professional persons.Lack investment experiences, mathematical knowledge, economy and finance knowledge it is general Logical people can also make efficient investment.
Embodiment described above only indicates embodiments of the present invention, and the description thereof is more specific and detailed, but can not manage Solution is limitation of the scope of the invention.It should be pointed out that for those skilled in the art, not departing from structure of the present invention Under the premise of think of, various modifications and improvements can be made, these belong to the scope of the present invention.

Claims (10)

1. the method for carrying out portfolio investment decision based on neural network and machine learning, it is characterised in that: the method includes Following steps:
Step 1: transaction history data, economical operation data, the enterprise operation data of continuous collecting securities market；
Step 2: the data obtained in step 1 are normalized；Before normalized, first to outlier (Outlier) The too far variable in bias data compact district carries out interval mapping, reduces the distance between its outlier and data-intensive district；
Step 3: on computers, with different model structures and different parameters, building multiple based on neural network (Neural Networks model)；
Step 4: using the data obtained in step 2, in the method for machine learning (Machine Learning), on computers The multiple models created in training step 3 abandon the high model of error, it is low to leave error；
Step 5: each model screened in step 4, in each investment cycle, all separately to instantly with it is following Security product tendency predict and makes investment decision accordingly；
Step 6: the investment decision of oneself is taken out progress " ballot ", the throwing of comprehensive each model respectively by each model in step 5 Ticket is as a result, obtain final investment decision；
Step 7: in next investment cycle, with data updated in steps 1 and 2, to each mould screened in step 4 Type carries out supplementary training, then repeatedly step 5,6, so recycles.
2. the method according to claim 1 for carrying out portfolio investment decision based on neural network and machine learning, feature It is: in the step 2, for the initial data obtained from step 1, if the outlier (outlier) of certain variables deviates Data-intensive district is far, will have an adverse effect to data application thereafter, at this time first these variables carry out logarithm (log) or The transformation of n times root is opened, data distribution can be made more uniform after transformation.
3. the method according to claim 1 for carrying out portfolio investment decision based on neural network and machine learning, feature Be: the normalized in the step 2 is that all different dimensions, different size of variable are passed through interval mapping, transformation To the real number interval of [0,1] or [- 1,1], in case the small variable of absolute value is flooded by the big variable of absolute value and can not embody it Feature.
4. the method according to claim 1 for carrying out portfolio investment decision based on neural network and machine learning, feature It is: in step 3, utilizes convolutional neural networks (Convolutional Neural Networks), Recognition with Recurrent Neural Network It is (Recurrent Neural Network), shot and long term memory network (Long Short Term Memory Network), preceding Neural network (feedforward neural network) is presented by different combination and network parameter, is built into multiple The model that structure is different, parameter is different.
5. the method according to claim 1 for carrying out portfolio investment decision based on neural network and machine learning, feature Be: in step 4, the neuron (Neurons) of model will lead to poor fitting if quantity is very few, excessive and will lead to over-fitting, Therefore it automatically adjusts to the neuronal quantity of neural network；If some model poor fitting, automatic weight after increasing neuronal quantity New training；If some model over-fitting, automatic re -training after reducing neuronal quantity；Until fitting is good.
6. the method according to claim 1 for carrying out portfolio investment decision based on neural network and machine learning, feature It is: in step 4, with the different model of the data training of different time span；Several moons are as short as, it is long to many decades；Its meaning Be: some models are good at the experience for summarizing long history, some models, which are then good at, adapts to fast-changing environment and game rule Then.
7. the method according to claim 1 for carrying out portfolio investment decision based on neural network and machine learning, feature Be: in step 6, each model vote when franchise weight be it is inconsistent, the high model influence power of weight is bigger；Each The each decision of model is recorded, and shines compared with the true tendency of securities market thereafter, to count the decision of the model Quality.The high model of Decision Quality, improves its ballot weight；The low model of Decision Quality, reduces its ballot weight, until cancelling Its franchise.
8. the method according to claim 7 for carrying out portfolio investment decision based on neural network and machine learning, feature Be: the Decision Quality of model can be measured by the prediction error to investment target, and calculation formula is as follows:
Wherein: E predicts error, and it is R that n, which is total periodicity of prediction,iFor the investment target period i true value；FiFor the throwing Target is provided in the predicted value of period i；
Optionally, the Decision Quality of model can also be measured by investment return, and calculation formula is as follows:Wherein: G For accumulated earnings, n is total periodicity of prediction, giFor the income in i-th of period；
The high model of Decision Quality, improves its ballot weight；The low model of Decision Quality, reduces its ballot weight, until cancelling Its franchise.
9. carrying out the device of portfolio investment decision based on neural network and machine learning characterized by comprising refer to for storing The storage unit of order, and at least a processor, the processor are coupled with the storage unit；Wherein, when at least one When processor executes described instruction, described instruction causes the processor to execute method according to claims 1-8.
10. the device according to claim 9 for carrying out portfolio investment decision based on neural network and machine learning, feature It is, comprising with lower module:
Data acquisition module: transaction history data, enterprise operation data for continuous collecting China A share market, economical operation Macro-indicators, and the information of acquisition is stored to MySql database；
Data prediction and normalization module: for carrying out log transformation to the too far variable in outlier bias data compact district, so After be normalized；
Neural network model builds module: for building 50 different neural network models of structure difference, parameter.Neural network Model training and screening module: for training neural network model, the high model of error is eliminated, the low model of error is left；
Each Model Independent decision-making module: for enabling each neural network model independently make investment decision；
Each model vote module: for ballot weight to be applied to each model, each model is then allowed to take their own throwing Decision is provided to vote, obtains final investment decision；
Each model decision quality assessment modules: for assessing the investment decision quality of each model, and their throwing is modified accordingly Ticket weighs weight；
Each model continues training module: training each model for continuing with newest data, prepares in next investment cycle Carry out decision；
Output module: for exporting investment decision.
CN201910321426.XA 2019-04-22 2019-04-22 The method and device of portfolio investment decision is carried out based on neural network and machine learning Pending CN110047006A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910321426.XA CN110047006A (en) 2019-04-22 2019-04-22 The method and device of portfolio investment decision is carried out based on neural network and machine learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910321426.XA CN110047006A (en) 2019-04-22 2019-04-22 The method and device of portfolio investment decision is carried out based on neural network and machine learning

Publications (1)

Publication Number Publication Date
CN110047006A true CN110047006A (en) 2019-07-23

Family

ID=67278232

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910321426.XA Pending CN110047006A (en) 2019-04-22 2019-04-22 The method and device of portfolio investment decision is carried out based on neural network and machine learning

Country Status (1)

Country Link
CN (1) CN110047006A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113496335A (en) * 2020-04-07 2021-10-12 厦门邑通软件科技有限公司 Method, system and equipment for recording decision-making behaviors
CN114509679A (en) * 2022-04-20 2022-05-17 中汽信息科技(天津)有限公司 Battery SOH prediction model construction method based on deep learning

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113496335A (en) * 2020-04-07 2021-10-12 厦门邑通软件科技有限公司 Method, system and equipment for recording decision-making behaviors
CN114509679A (en) * 2022-04-20 2022-05-17 中汽信息科技(天津)有限公司 Battery SOH prediction model construction method based on deep learning

Similar Documents

Publication Publication Date Title
Li et al. Risk prediction in financial management of listed companies based on optimized BP neural network under digital economy
CN110766212B (en) Ultra-short-term photovoltaic power prediction method for historical data missing electric field
Sayavong et al. Research on stock price prediction method based on convolutional neural network
Xie et al. Blockchain financial investment based on deep learning network algorithm
CN110751318B (en) Ultra-short-term power load prediction method based on IPSO-LSTM
CN107194803A (en) A kind of P2P nets borrow the device of borrower's assessing credit risks
Patni et al. Air quality prediction using artificial neural networks
Zhu et al. Coke price prediction approach based on dense GRU and opposition-based learning salp swarm algorithm
CN110378382A (en) Novel quantization transaction system and its implementation based on deeply study
CN108734220A (en) Adaptive Financial Time Series Forecasting method based on k lines cluster and intensified learning
CN111489259A (en) Stock market risk prediction intelligent implementation method based on deep learning
CN110047006A (en) The method and device of portfolio investment decision is carried out based on neural network and machine learning
CN112862182A (en) Investment prediction method and device, electronic equipment and storage medium
Zhao et al. A novel cryptocurrency price time series hybrid prediction model via machine learning with MATLAB/Simulink
Zhou Sports achievement prediction and influencing factors analysis combined with deep learning model
Zhuang et al. Research on Quantitative Stock Selection Strategy Based on CNN-LSTM
CN115423499A (en) Model training method, price prediction method, terminal device, and storage medium
Liu The prediction model and system of stock rise and fall based on BP neural network
Wang et al. Research on quantitative trading strategy based on LSTM
Ruciński Neural modelling of electricity prices quoted on the Day-Ahead Market of TGE SA shaped by environmental and economic factors
CN113935413A (en) Distribution network wave recording file waveform identification method based on convolutional neural network
Song et al. Dynamic adjustment of stock position based on hybrid deep neural network
Gorbachev et al. Cognitive Neural Network Foresight to Forecast Scientific and Technological Development of the State
Wang et al. Compare CNN and LSTM Model to Forecast the Stock Price
Ji Application research of SGD algorithm based on PCA dimensionality reduction technique for stock price prediction

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