CN109948930A - Deep learning method and its application for driving training planning - Google Patents
Deep learning method and its application for driving training planning Download PDFInfo
- Publication number
- CN109948930A CN109948930A CN201910204504.8A CN201910204504A CN109948930A CN 109948930 A CN109948930 A CN 109948930A CN 201910204504 A CN201910204504 A CN 201910204504A CN 109948930 A CN109948930 A CN 109948930A
- Authority
- CN
- China
- Prior art keywords
- feature
- training
- student
- pretreatment
- deep learning
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Abstract
This disclosure relates to a kind of deep learning method, apparatus, electronic equipment and computer-readable medium.It is a kind of for driving training planning deep learning method include: to be pre-processed to student's feature, obtain pretreatment student's feature；Customization training feature is pre-processed, pretreatment customization training feature is obtained；Student is acquired in the features of construction schedule of the first pre- timing point to the pre- timing point of N；Using the pretreatment student feature, pretreatment customization training feature and the features of construction schedule as input, timing Neural Network model predictive residue training duration is utilized.Pass through the predictable remaining training duration of this method, effectively progress cost management.
Description
Technical field
This disclosure relates to computer information processing field, in particular to a kind of depth for driving training planning
Learning method, device, electronic equipment and computer-readable medium.
Background technique
Self-driving trip has become a kind of common trip mode, for this purpose, obtaining driving license by examination has been many people
Rigid need.Currently, people mainly pass through the training class for participating in driving school, by training, driving Course Examination is then participated in.Driving school one
As to participate in the student of training preset several classes of types, student is generally according to itself economic conditions and time situation selection one
Kind class's type.
But present inventor has found, the real learning time of plan training time of these class of type and student are often
Not consistent, many students may need the training time longer than planned time.This just to the didactic code of student pilot and
Cost management brings certain difficulty.Therefore, it is necessary to a kind of driving training planing methods, can carry out to current residual training duration
Prediction.
Above- mentioned information are only used for reinforcing the understanding to the background of the disclosure, therefore it disclosed in the background technology part
It may include the information not constituted to the prior art known to persons of ordinary skill in the art.
Summary of the invention
The application provide it is a kind of for driving training planning deep learning method, apparatus, electronic equipment and computer can
Medium is read, residual time length can be predicted.
Other characteristics and advantages of the disclosure will be apparent from by the following detailed description, or partially by the disclosure
Practice and acquistion.
According to the one side of the disclosure, a kind of deep learning method for driving training planning is provided, comprising:
Student's feature is pre-processed, pretreatment student's feature is obtained；
Customization training feature is pre-processed, pretreatment customization training feature is obtained；
Student is acquired in the features of construction schedule of the first pre- timing point to the pre- timing point of N；
Using the pretreatment student feature, pretreatment customization training feature and the features of construction schedule as inputting, when utilization
Sequence Neural Network model predictive residue training duration.
Optionally, the features of construction schedule includes at least two in following characteristics: being commented with duration, progress ratio, progress
Valence, projects scoring.
Optionally, the progress ratio is the planned time that training project is completed and the planned time of all training projects
Between ratio.
Optionally, the progress evaluation includes subjective assessment of the coach to student training progress.
Optionally, projects scoring includes the scoring that project is completed in student, and the scoring for not completing project is zero
Point or predetermined score.
Optionally, preceding method further include: determine current whole cost to be paid according to the sum of all remaining training durations.
Optionally, the timing neural network model includes RNN, LSTM, GRU, two-way RNN or SRU.
Optionally, student's feature includes at least two in following characteristics: age, gender, occupation type, education journey
Degree, time guarantee degree, self-description option, manipulation class game test result.
Optionally, customization training feature include: time classification, teaching method, coach's quantity, vehicle, coach's evaluation,
At least one of trainer attributes.
Optionally, customization training feature includes examination surroundings feature.Therefore, optionally, examination surroundings feature includes examination hall
At least one of percent of pass, examination date, expected weather, pre- end-of-term examination road.
Optionally, it is described to student's feature carry out pretreatment include: by student's character representation be row vector form.It is described right
Customizing training feature and carrying out pretreatment includes: that customization is trained character representation as row vector form.
According to another aspect of the present invention, a kind of deep learning device for driving training planning is provided, comprising:
First preprocessing module obtains pretreatment student's feature for pre-processing to student's feature；
Second preprocessing module obtains pretreatment customization training feature for pre-processing to customization training feature；
When point module, for acquiring student in the features of construction schedule of the first pre- timing point to the pre- timing point of N；
Prediction module, for making the pretreatment student feature, pretreatment customization training feature and the features of construction schedule
For input, timing Neural Network model predictive residue training duration is utilized.
In accordance with a further aspect of the present invention, a kind of electronic equipment is provided, comprising: one or more processors；Storage device,
For storing one or more programs；When one or more of programs are executed by one or more of processors, so that institute
It states one or more processors and realizes aforementioned any method.
In accordance with a further aspect of the present invention, a kind of computer-readable medium is provided, computer program is stored thereon with, it is special
Sign is, aforementioned any method is realized when described program is executed by processor.
According to some embodiments of the present invention, residual time length can be predicted at any time, to can determine current entirety
Cost to be paid.
Other embodiments according to the present invention, according to the prediction to residual time length, when training can be adjusted flexibly in student
Between arrange or adjustment customization training contents.
It should be understood that the above general description and the following detailed description are merely exemplary, this can not be limited
It is open.
Detailed description of the invention
Its example embodiment is described in detail by referring to accompanying drawing, above and other target, feature and the advantage of the disclosure will
It becomes more fully apparent.
Figure 1A show accoding to exemplary embodiment using according to the method for the embodiment of the present invention or the system of device
Block diagram；
Figure 1B shows the frame of the deep learning method according to an exemplary embodiment of the present invention for driving training planning
Schematic diagram；
Fig. 2 shows the processes of the deep learning method according to an exemplary embodiment of the present invention for driving training planning
Figure；
Fig. 3 Fig. 3 show it is according to an embodiment of the present invention for driving training planning LSTM network layer specific structure and
The information process of each control door；
Fig. 4 shows example embodiment according to the present invention and trains to student's feature, the customization training students such as feature or features of construction schedule
It instructs feature and carries out pretreated method；
Fig. 5 shows the mistake that example embodiment according to the present invention utilizes history training sample training timing neural network model
Journey；
Fig. 6 diagrammatically illustrates the deep learning device for driving training planning of example embodiment according to the present invention
Block diagram；
Fig. 7 shows the block diagram for the electronic equipment of driving training planning accoding to exemplary embodiment.
Specific embodiment
Example embodiment is described more fully with reference to the drawings.However, example embodiment can be real in a variety of forms
It applies, and is not understood as limited to embodiment set forth herein；On the contrary, thesing embodiments are provided so that the disclosure will be comprehensively and complete
It is whole, and the design of example embodiment is comprehensively communicated to those skilled in the art.Identical appended drawing reference indicates in figure
Same or similar part, thus repetition thereof will be omitted.
In addition, described feature, structure or characteristic can be incorporated in one or more implementations in any suitable manner
In example.In the following description, many details are provided to provide and fully understand to embodiment of the disclosure.However,
It will be appreciated by persons skilled in the art that can with technical solution of the disclosure without one or more in specific detail,
Or it can be using other methods, constituent element, device, step etc..In other cases, it is not shown in detail or describes known side
Method, device, realization or operation are to avoid fuzzy all aspects of this disclosure.
Block diagram shown in the drawings not necessarily must be corresponding with physically separate entity.I.e., it is possible to using software
Form realizes these functional entitys, or these functional entitys are realized in one or more hardware modules or integrated circuit, or
These functional entitys are realized in heterogeneous networks and/or processor device and/or microcontroller device.
Flow chart shown in the drawings is merely illustrative, it is not necessary to including all content and operation/step,
It is not required to execute by described sequence.For example, some operation/steps can also decompose, and some operation/steps can close
And or part merge, therefore the sequence actually executed is possible to change according to the actual situation.
It should be understood that although herein various assemblies may be described using term first, second, third, etc., these groups
Part should not be limited by these terms.These terms are to distinguish a component and another component.Therefore, first group be discussed herein below
Part can be described as the second component without departing from the teaching of disclosure concept.As used herein, term " and/or " include associated
All combinations for listing any of project and one or more.
It will be understood by those skilled in the art that attached drawing is the schematic diagram of example embodiment, module or process in attached drawing
Necessary to not necessarily implementing the disclosure, therefore it cannot be used for the protection scope of the limitation disclosure.
In driving training, although there is various class's types, the actual learning of the plan training time and student of these class of type
Time is often not consistent.Some students also have since capability problems may need the training time longer than planned time
Student can elongate the training time due to odjective cause.This just brings the didactic code of student pilot and cost management certain tired
It is difficult.In addition, student also wants to understand the practical training duration oneself also needed, thus reasonable arrangement other business.
Based on these realistic problems that inventor excavates, the present invention proposes a kind of technical concept and scheme, based on timing mind
Timing neural network model is established through network, based on student's feature, customization training feature and features of construction schedule, when predicting remaining training
It is long.
Inventors have found that although time correlation prediction can be carried out based on timing neural network, for driving training
Speech can not obtain preferable prediction result if only considering features of construction schedule and ignoring student's feature and customization training feature.Hair
Bright people discovery, this is because the residual time length of student is not only related to progress training feature, also with foundation characteristic (student's feature,
Customization training feature) it is related, and also these foundation characteristics not only only belong to primary condition feature, but in entirely training timing thing
It is all wielding influence and is acting in part.Therefore, inventor uses student's feature when predicting by timing neural network, determines
System training feature and features of construction schedule are inputted as node.
The embodiment of the present invention is described in detail with reference to the accompanying drawings.
Figure 1A show accoding to exemplary embodiment using according to the method for the embodiment of the present invention or the system of device
Block diagram.
As shown in Figure 1A, system architecture 100 may include terminal device 101,102,103, network 104 and server 105.
Network 104 between terminal device 101,102,103 and server 105 to provide the medium of communication link.Network 104 can be with
Including various connection types, such as wired, wireless communication link or fiber optic cables etc..
User can be used terminal device 101,102,103 and be interacted by network 104 with server 105, to receive or send out
Send message etc..Various telecommunication customer end applications, such as prediction application, webpage can be installed on terminal device 101,102,103
Browser application, searching class application, instant messaging tools, mailbox client, social platform software etc..
Terminal device 101,102,103 can be the various electronic equipments with display screen and supported web page browsing, packet
Include but be not limited to smart phone, tablet computer, pocket computer on knee and desktop computer etc..
Server 105 can be to provide the server of various services, such as utilize terminal device 101,102,103 to user
The information submitted provides the back-stage management server of prediction processing.Back-stage management server can use prediction model to reception
To the stored data of information and system carry out the processing such as calculating, and processing result is fed back into terminal device.Server 105
Other relevant operations and processing can be also carried out according to actual needs.Server 105 can be the server of an entity, can also example
As being to be made of multiple servers.
Figure 1B shows the frame of the deep learning method according to an exemplary embodiment of the present invention for driving training planning
Schematic diagram.
As shown in Figure 1B, S102 includes data acquisition cleaning and processing.Specifically, history training data is acquired, is cleaned different
Normal training data converts dirty data to and meets the quality of data using mathematical statistics, data mining or predefined cleaning rule
It is required that data.
In the process, generally require by consistency check, invalid value and missing values processing (including estimation, whole example
Deletion, variable deletion, in pairs deletion etc.).Main wash data type includes incomplete data, wrong data, repeated data etc..
The method of cleaning data generally comprises the detection and solution for solving the method, error value of deficiency of data (i.e. value lacks)
The certainly detection and solution of method, the detection of repetition record and removing method, inconsistency (inside data source and between data source)
Method etc..
Then, data can be counted and is analyzed, and can provide a kind of mode of statistical data result visualization.
S104 includes feature selecting and characteristic processing.For example, to historical data carry out characteristic processing, logarithm type data into
Line amplitude adjustment/normalization is encoded or is converted to term vector to classification type data (such as text information) and handled, right
Time data carry out successive value extraction (duration etc.) and discrete value extraction (time point analysis), to text type data into
Row natural language processing etc..
Feature generally may include plus-minus average, lane place line, order type, proportional-type etc..Plus-minus is averagely, for example, this student
Exam score be higher than all student's exam scores number the examination ability of one people (tradeoff), student's number of days of continuously taking an examination is super
Cross average how many (showing this student to the stickiness for driving training).Lane place line is, for example, how many points for training number of days and being in all students
At bit line.Order type is which position come.
S106 includes construction feature engineering.Historical data feature may include such as student's feature, training feature, features of construction schedule
And other features.Student's feature may include age, gender, occupation type, education degree, time guarantee degree, self-description choosing
Etc..Training feature may include time classification, teaching method, trainer attributes, coach's quantity, vehicle etc..Other features may include
Coach's evaluation etc..Features of construction schedule includes having used duration, progress ratio, progress evaluation, projects scoring etc..
Feature selecting and construction feature engineering can be carried out from these features.It is, for example, possible to use PCA principal component analysis,
LDA linear discriminant analysis, ICA independent component analysis carry out feature extraction.Feature selecting is to reject uncorrelated or redundancy spy
Sign, reduces the number of validity feature, reduces the time of model training, improve the accuracy of model.With statistical method, weighing apparatus
Measure the relationship between single feature and response variable (Lable).
Certainly, based on the use of deep neural network, it may not be necessary to it does a large amount of Feature Engineering or even omits, Ke Yizhi
It connects and data filling is entered, allow network oneself to train, self-recision.
S108 includes building deep neural network model, such as constructs RNN sequential depth neural net regression model or divide
Class model, and usage history data such as driving training data are trained model.Neural net layer may include three classes: defeated
Enter layer, hidden layer and output layer.First layer is input layer, and the last layer is output layer, and centre is all hidden layer.Before use
To and/or back-propagation algorithm, set loss function, carry out the training and prediction of neural network.
It can also construction logic regression model, linear regression mould in other methods in addition to building deep neural network model
Type, support vector regression model, polynomial regression model, ridge regression model, Lasso regression model, elastomeric network regression model
Deng.
S110 is predicted including the use of model.It is row vector form by student's character representation；Mark sheet is trained into customization
It is shown as row vector form；Features of construction schedule is expressed as row vector form.Using feature as input, model prediction residue training is utilized
Time.
The foregoing describe general principles, illustrate the technical solution of technical concept according to the present invention with reference to the accompanying drawings.
Fig. 2 shows the processes of the deep learning method according to an exemplary embodiment of the present invention for driving training planning
Figure.
As shown in Fig. 2, pre-processing in S202 to student's feature, pretreatment student's feature is obtained.
In accordance with the technical idea of the present invention, training concrete scheme is customized according to student's feature selecting.To customize driving training
For, according to some embodiments, student's feature includes at least two in following characteristics: age, gender, occupation type, education
Degree, time guarantee degree, self-description option.Occupation type can manipulate class for example to research and develop class, administrative sale class, equipment
Deng.Time guarantee degree, which can be divided into, to be completely secured, most of guarantee, partially guarantees, is difficult to ensure, can also be used and accordingly be commented
Divide system.Self-description may include such as carefulness degree, patience degree, harmony, driving hobby degree.
According to other embodiments, student's feature includes manipulation class game test result.Driving efficiency is one and manipulation
The relevant technical ability of ability.Therefore, it is tested by carrying out manipulation class game to student, such as driving game, it can be preferably
Predict training process and result.
In accordance with the technical idea of the present invention, prediction is giveed training using deep neural network.For this reason, it may be necessary to student's feature
It is pre-processed, is converted into the manageable input data of deep neural network.Further below this will exemplary retouch in detail
It states.
In S204, customization training feature is pre-processed, obtains pretreatment customization training feature.
Customization training feature mentioned here can be the feature for having each training class's type, be also possible to according to training side
The feature for the various customization trainings that formula is combined into.
By taking driving training as an example, there are class, all-day class, common honored guest's all-day class, honored guest class etc. on ordinary days in some driving schools.These classes
Difference essentially consist in about the vehicle time, coach whether fix.In addition, there is also one-to-one teaching and group instructions for type not of the same class
Difference and automatic catch vehicle and manual gear vehicle difference etc..
According to some embodiments of the present invention, when predicting remaining training duration, consider the factor of customization training.Work as training
When factor difference, prediction result be would also vary from.Customization training feature include: time classification, teaching method, coach quantity,
Vehicle.Time classification can be such as class or all-day class on ordinary days.Teaching method can be one-to-one or group instruction.Train quantity
It can be 1 or 1/N (when N number of coach, inverted i.e. 1/N).Vehicle can be automatic catch or manual gear.
According to some embodiments, customization trains feature on the basis of time classification, teaching method, coach's quantity, vehicle,
It may also include coach's evaluation that student provides, such as the overall score that provides of student or classification scoring, it is such as patient, whether be good at teaching
, attitude etc..
According to some embodiments, customization training feature includes trainer attributes.According to statistics, coach is largely determined
Result of training.In addition, different students also more adapt to different types of coach, vice versa.Trainer attributes may include coach's year
Age, personality classification (for example, introversive, export-oriented), the length of teaching, driving age, history percent of pass etc..When calculating history percent of pass, if one
The about excessively N number of coach of a student is then calculated by 1/N student of each coach's band.
According to some embodiments, customization training feature includes examination surroundings feature.Even for the same student, due to examining
Try the influence of environmental factor, it is also possible to which examination result is different.For example, the condition in different examination halls is different, it can reflect and pass the examination
In rate.In addition, on examination date such as season, month etc., also have an impact to examination result.It is expected that weather such as wind, snow, rain, mist etc.,
Also there is certain influence to examination result.Certainly, examination surroundings further include other factors, and details are not described herein again.Therefore, optionally,
Examination surroundings feature includes at least one of examination hall percent of pass, examination date, expected weather, pre- end-of-term examination road.
In accordance with the technical idea of the present invention, prediction is giveed training using deep neural network.For this reason, it may be necessary to train customization
Feature is pre-processed, and the manageable input data of deep neural network is converted into.This will be carried out further below exemplary detailed
Thin description.
In S206, student is acquired in the features of construction schedule of the first pre- timing point to the pre- timing point of N.
According to training process, in pre- timing point sequence node or period, point sequence node (for example, predetermined space) is acquired
The features of construction schedule, for example, the first pre- timing point, the second pre- timing point ..., the pre- timing point of N, N is natural number.Features of construction schedule
It may include at least two in following characteristics: having used duration, progress ratio, progress evaluation, projects scoring.
Optionally, the progress ratio is the planned time that training project is completed and the planned time of all training projects
Between ratio.According to some embodiments, a weighted value can be applied to the planned time of each training project.
Optionally, the progress evaluation includes subjective assessment of the coach to student training progress, including marking or text
Evaluation, for example, it is outstanding, good etc..
Optionally, projects scoring includes the scoring that project is completed in student, and the scoring for not completing project is zero
Point or predetermined score (for example, 50 points etc.).
In accordance with the technical idea of the present invention, prediction is giveed training using deep neural network.For this purpose, acquisition features of construction schedule is also
Including pre-processing to features of construction schedule, it is converted into the manageable input data of deep neural network.Further below will to this into
The exemplary detailed description of row.
In S208, using the pretreatment student feature, pretreatment customization training feature and the features of construction schedule as inputting,
Utilize timing Neural Network model predictive residue training duration.
Timing neural network model can be RNN or its improved model.RNN network is by multiple concatenated hiding network layer structures
At especially suitable for handling the data set based on time domain by deep learning.The calculation formula of the hidden layer neuron of RNN network
Are as follows:
S (t)=f (x (t) U+s (t-1) W)
Wherein U, W are the parameter of RNN network model, and f indicates activation primitive.Hidden layer neuron activation for time t
Value st uses the input xt of the hidden layer neuron of time t and a upper hidden layer neuron (corresponding to a upper time t-1)
Activation value st-1 carries out calculating acquisition.
Hiding layer state may be considered the memory unit of network, contain the hiding layer state of all steps in front.And it is defeated
The output of layer is related with the s (t) currently walked out.In practice, in order to reduce the complexity of network, before often s (t) only includes
Several steps in face rather than the hiding layer state of all steps.In traditional neural network, the parameter of each network layer is not shared
's.And in RNNs, one step of every input, each layer each shared parameter, each step in this reflection RNNs all do it is identical
Work, only input is different, therefore greatly reduces the parameter for needing to learn in network.
In traditional RNN, training algorithm is that (Back-propagation Through Time, passes through time reversal to BPTT
It propagates).But when the period is long, BPTT causes RNN network to need the residual error returned that can exponentially decline, and causes
Network weight updates slowly, can not embody the effect of the long-term memory of RNN, it is therefore desirable to which a storage unit is remembered to store
Recall.
Therefore, and the improved model of a kind of RNN: shot and long term memory models (Long-short Term Memory, letter is proposed
Claim LSTM).This special RNN network model is to solve the problems, such as RNN model gradient disperse.LSTM has " triple gate ": defeated
Get started i, out gate o, forgets door f, value range is restricted within (0,1) using Sigmoid function.It can be with using three doors
Different moments information flow direction is controlled, door and input gate are forgotten by control, suitable information is selected to enter the cell in center,
Irrelevant information is kept outside of the door；By controlling out gate, most suitable moment output cell is selected treated information.
Other than LSTM, according to some embodiments of the invention, it is used for it is also an option that GRU, two-way RNN or SRU are used as
The timing neural network model of residual time length prediction.
Deep learning method according to an embodiment of the present invention, using timing neural network timing neural network model,
Based on student's feature, customization training feature and features of construction schedule, remaining training duration is predicted.The residual time length of student not only with progress
It is related to train feature, it is also related to foundation characteristic (student's feature, customization training feature), and also these foundation characteristics not only belong to
In primary condition feature, but is all wielding influence and acting in entirely training temporal events.It is according to an embodiment of the present invention
Therefore method is capable of the remaining training duration of more accurate prediction science person.
According to some embodiments, current whole cost to be paid can be determined according to the sum of all remaining training durations.
Fig. 3 shows the specific structure of LSTM network layer according to an embodiment of the present invention for driving training planning and each
Control the information process of door.
Dotted line frame in module 301 shows the information process for forgeing door, and functional expression is as follows:
ft-σ(WtxXt+Wthht-1+bt)
Wherein Wfx、Wfh、bfDoor is respectively forgotten to the input X of current time point ttWith the hidden state h of previous time point t-1t-1
Network weight parameter and linear transformation parameter.
Dotted line frame in module 302 then shows the information process of input gate and intermediate memory state, and functional expression is such as
Under:
it=σ (W1xXt+W1hht-1+bi)
Ct=σ (WCxXt+WChht-1+bC)
Wherein W1x、W1h、biRespectively input X of the input gate to current time point ttWith the hidden state h of previous time point t-1t-1
Network weight parameter and linear transformation parameter, WCx、WCh、bCInput X of the respectively intermediate memory state to current time point ttWith
The hidden state h of previous time point t-1t-1Network weight parameter and linear transformation parameter.
Dotted line frame in module 303 then shows the process of memory state transmission.
Dotted line frame in module 304 shows the information process of out gate and hidden state, and functional expression is as follows:
ot=σ (WoxXt+Wohht-1+bo)
Wherein Wox、Woh、boRespectively input X of the out gate to current time point ttWith the hidden state h of previous time point t-1t-1
Network weight parameter and linear transformation parameter.
For the hidden state of current time point t, functional expression is seen above.
Fig. 4 shows example embodiment according to the present invention and trains to student's feature, the customization training students such as feature or features of construction schedule
It instructs feature and carries out pretreated method.
As shown in figure 4, being row vector form by student training character representation in S401.
Natural language is handled using computer, needs natural language processing becoming the symbol that machine can identify
Number, and in machine-learning process, it needs to quantize.
The original form of student training feature is generally character string forms, these primitive characters are being carried out head and the tail splicing
Afterwards, the feature of row vector form is obtained.
According to some embodiments of the present invention, can by the way that a part of list of feature values is shown as binary value form, meanwhile,
The transformation of another part characteristic value is characterized, and the list of feature values for the feature that transformation obtains is shown as binary value form.By this
A little Boolean sequences are connected, the feature of available row vector form.For example, can indicate manual with 0 or 00 for vehicle
Gear, indicates automatic catch with 1 or 01.For the age, age bracket can be divided, such as is expressed as within 18-28 years old the first age bracket feature, such as
Fruit student falls in this range the age, then the value of the first age bracket feature is 1, at this moment the value of other age bracket features then Xiang Yingwei
0.Certainly, age bracket can also be divided, each age bracket is indicated with such as 000,001,010,011,100 etc. respectively.Other features
It can be processed similarly with this.
Fig. 5 shows the mistake that example embodiment according to the present invention utilizes history training sample training timing neural network model
Journey.
As shown in figure 5, in S501, from the student training feature of history training sample extraction row vector form.
As before, student's feature, customization training feature, features of construction schedule original form be generally character string forms, by this
After a little primitive characters carry out head and the tail splicing, the feature of row vector form is obtained.
According to some embodiments of the present invention, can by the way that a part of list of feature values is shown as binary value form, meanwhile,
The transformation of another part characteristic value is characterized, and the list of feature values for the feature that transformation obtains is shown as binary value form.By this
A little Boolean sequences are connected, the feature of available row vector form.For example, can indicate manual with 0 or 00 for vehicle
Gear, indicates automatic catch with 1 or 01.For the age, age bracket can be divided, such as is expressed as within 18-28 years old the first age bracket feature, such as
Fruit student falls in this range the age, then the value of the first age bracket feature is 1, at this moment the value of other age bracket features then Xiang Yingwei
0.Certainly, age bracket can also be divided, each age bracket is indicated with such as 000,001,010,011,100 etc. respectively.Other features
It can be processed similarly with this.
In S503, student training feature training timing neural network model is utilized.
According to example embodiment, it is marked using the actual result in history training sample as sample, then utilizes mark
Sample training timing neural network model.
It will be appreciated by those skilled in the art that realizing that all or part of the steps of above-described embodiment is implemented as being executed by CPU
Computer program.When the computer program is executed by CPU, above-mentioned function defined by the above method that the disclosure provides is executed
Energy.Program can store in a kind of computer readable storage medium, which can be read-only memory, disk or
CD etc..
Further, it should be noted that above-mentioned attached drawing is only the place according to included by the method for disclosure exemplary embodiment
Reason schematically illustrates, rather than limits purpose.It can be readily appreciated that above-mentioned processing shown in the drawings is not indicated or is limited at these
The time sequencing of reason.In addition, be also easy to understand, these processing, which can be, for example either synchronously or asynchronously to be executed in multiple modules.
Following is embodiment of the present disclosure, can be used for executing disclosed method.For embodiment of the present disclosure
In undisclosed details, please refer to embodiments of the present disclosure.
Fig. 6 diagrammatically illustrates the deep learning device for driving training planning of example embodiment according to the present invention
Block diagram.
As shown in fig. 6, the deep learning device 600 for driving training planning of example embodiment includes according to the present invention
First preprocessing module 610, the second preprocessing module 620, when point module 630 and prediction module 640.
First preprocessing module 610 carries out pretreatment to student's feature to obtain pretreatment student's feature.Second pre- place
620 pairs of customization training features of reason module carry out pretreatment to obtain pretreatment customization training feature.When point module 630 for adopting
Collect student in the features of construction schedule of the first pre- timing point to the pre- timing point of N.Prediction module 640 is used for the pretreatment student is special
Sign, pretreatment customization training feature and the features of construction schedule utilize timing Neural Network model predictive residue training as input
Duration.
Fig. 6 shown device is corresponding with preceding method, omits the detailed description herein.
It will be appreciated by those skilled in the art that above-mentioned each module can be distributed in device according to the description of embodiment, it can also
Uniquely it is different from one or more devices of the present embodiment with carrying out corresponding change.The module of above-described embodiment can be merged into
One module, can also be further split into multiple submodule.
Fig. 7 shows the block diagram for the electronic equipment of driving training planning accoding to exemplary embodiment.
The electronic equipment 700 of this embodiment according to the disclosure is described referring to Fig. 7.The electronics that Fig. 7 is shown
Equipment 700 is only an example, should not function to the embodiment of the present disclosure and use scope bring any restrictions.
As shown in fig. 7, computer system 700 includes central processing unit (CPU) 701, it can be read-only according to being stored in
Program in memory (ROM) 702 or be loaded into the program in random access storage device (RAM) 703 from storage part 708 and
Execute various movements appropriate and processing.In RAM 703, it is also stored with various programs and data needed for system operatio.CPU
701, ROM 702 and RAM 703 is connected with each other by bus 704.Input/output (I/O) interface 705 is also connected to bus
704。
I/O interface 705 is connected to lower component: the importation 706 including touch screen, keyboard etc.；Including such as liquid crystal
The output par, c 707 of display (LCD) etc. and loudspeaker etc.；Storage part 708 including flash memory etc.；And including such as without
The communications portion 709 of gauze card, High_speed NIC etc..Communications portion 709 executes communication process via the network of such as internet.It drives
Dynamic device 710 is also connected to I/O interface 705 as needed.Detachable media 711, semiconductor memory, disk etc., according to
It needs to be mounted on driver 710, in order to be mounted into storage part as needed from the computer program read thereon
708。
By the description of above embodiment, those skilled in the art is it can be readily appreciated that example embodiment described herein
It can also be realized in such a way that software is in conjunction with necessary hardware by software realization.Therefore, implemented according to the disclosure
The technical solution of example can be embodied in the form of software products, which can store in a non-volatile memories
In medium (can be CD-ROM, USB flash disk, mobile hard disk etc.) or on network, including some instructions are so that a calculating equipment (can
To be personal computer, server, mobile terminal or network equipment etc.) it executes according to the method for the embodiment of the present disclosure.
The foregoing describe it is according to an embodiment of the present invention for driving training planning method and apparatus and electronic equipment and
Medium.By above detailed description, those skilled in the art is it can be readily appreciated that according to the method for the embodiment of the present invention and device
It has one or more of the following advantages.
According to some embodiments of the present invention, by the method for customization training, training efficiency is improved, and then it is logical to improve examination
Cross rate.
Other embodiments according to the present invention, according to student's situation and training method, exam score is trained in prediction, thus
Training method can be adjusted according to prediction result.
The present invention includes following implementation:
Embodiment 1, a kind of deep learning method for driving training planning characterized by comprising
Student's feature is pre-processed, pretreatment student's feature is obtained；
Customization training feature is pre-processed, pretreatment customization training feature is obtained；
Student is acquired in the features of construction schedule of the first pre- timing point to the pre- timing point of N；
Using the pretreatment student feature, pretreatment customization training feature and the features of construction schedule as inputting, when utilization
Sequence Neural Network model predictive residue training duration.
Embodiment 2, deep learning method as tdescribed in embodiment 1, which is characterized in that under the features of construction schedule includes
It states at least two in feature: having used duration, progress ratio, progress evaluation, projects scoring.
Embodiment 3, the deep learning method as described in embodiment 2, which is characterized in that the progress ratio is complete
At the ratio between the planned time of training project and the planned time of all training projects.
Embodiment 4, the deep learning method as described in embodiment 2, which is characterized in that the progress evaluation includes religion
Practice the subjective assessment to student training progress.
Embodiment 5, the deep learning method as described in embodiment 2, which is characterized in that described projects, which score, includes
The scoring of project is completed in student, and the scoring for not completing project is zero or predetermined score.
Embodiment 6, deep learning method as tdescribed in embodiment 1, which is characterized in that further include:
Current whole cost to be paid is determined according to the sum of all remaining training durations.
Embodiment 7, deep learning method as tdescribed in embodiment 1, which is characterized in that the timing neural network mould
Type includes RNN, LSTM, GRU, two-way RNN or SRU.
Embodiment 8, deep learning method as tdescribed in embodiment 1, which is characterized in that under student's feature includes
State at least two in feature: age, gender, occupation type, education degree, time guarantee degree, self-description option, manipulation class
Game test result.
Embodiment 9, deep learning method as tdescribed in embodiment 1, which is characterized in that feature packet is trained in the customization
It includes: at least one of time classification, teaching method, coach's quantity, vehicle, coach's evaluation, trainer attributes.
Embodiment 10, deep learning method as tdescribed in embodiment 1, which is characterized in that feature is trained in the customization
Including examination surroundings feature.
Embodiment 11, the deep learning method as described in embodiment 10, which is characterized in that the examination surroundings feature
Including at least one of examination hall percent of pass, examination date, expected weather, pre- end-of-term examination road.
Embodiment 12, deep learning method as tdescribed in embodiment 1, which is characterized in that
It is described to student's feature carry out pretreatment include:
It is row vector form by student's character representation；
Described pair of customization training feature carries out pretreatment and includes:
It is row vector form by customization training character representation.
Embodiment 13, a kind of deep learning device for driving training planning characterized by comprising
First preprocessing module obtains pretreatment student's feature for pre-processing to student's feature；
Second preprocessing module obtains pretreatment customization training feature for pre-processing to customization training feature；
When point module, for acquiring student in the features of construction schedule of the first pre- timing point to the pre- timing point of N；
Prediction module, for making the pretreatment student feature, pretreatment customization training feature and the features of construction schedule
For input, timing Neural Network model predictive residue training duration is utilized.
Embodiment 14, a kind of electronic equipment characterized by comprising
One or more processors；
Storage device, for storing one or more programs；
When one or more of programs are executed by one or more of processors, so that one or more of processing
Device realizes the deep learning method as described in any in embodiment 1-12.
Embodiment 15, a kind of computer-readable medium, are stored thereon with computer program, which is characterized in that the journey
The deep learning method as described in any in embodiment 1-12 is realized when sequence is executed by processor.
Those skilled in the art after considering the specification and implementing the invention disclosed here, will readily occur to of the invention its
Its embodiment.This application is intended to cover any variations, uses, or adaptations of the invention, these modifications, purposes or
Person's adaptive change follows general principle of the invention and including the undocumented common knowledge in the art of the present invention
Or conventional techniques.The description and examples are only to be considered as illustrative, and true scope and spirit of the invention are by following
Claim is pointed out.
It should be understood that the present invention is not limited to the precise structure already described above and shown in the accompanying drawings, and
And various modifications and changes may be made without departing from the scope thereof.The scope of the present invention is limited only by the attached claims.
Claims (10)
1. a kind of deep learning method for driving training planning characterized by comprising
Student's feature is pre-processed, pretreatment student's feature is obtained；
Customization training feature is pre-processed, pretreatment customization training feature is obtained；
Student is acquired in the features of construction schedule of the first pre- timing point to the pre- timing point of N；
Using the pretreatment student feature, pretreatment customization training feature and the features of construction schedule as input, timing mind is utilized
Remaining training duration is predicted through network model.
2. deep learning method as described in claim 1, which is characterized in that the features of construction schedule include in following characteristics extremely
It is two kinds few: duration, progress ratio, progress evaluation, projects scoring have been used,
Student's feature includes at least two in following characteristics: age, gender, occupation type, education degree, time guarantee
Degree, self-description option, manipulation class game test result,
The customization training feature includes: time classification, teaching method, coach's quantity, vehicle, trains and evaluate, in trainer attributes
At least one,
The customization training feature includes examination surroundings feature,
The examination surroundings feature includes at least one of examination hall percent of pass, examination date, expected weather, pre- end-of-term examination road,
It is described to student's feature carry out pretreatment include:
It is row vector form by student's character representation；
Described pair of customization training feature carries out pretreatment and includes:
It is row vector form by customization training character representation.
3. deep learning method as claimed in claim 2, which is characterized in that the progress ratio is that training project is completed
Ratio between planned time and the planned time of all training projects.
4. deep learning method as claimed in claim 2, which is characterized in that the progress evaluation includes coach to student training
The subjective assessment of progress.
5. deep learning method as claimed in claim 2, which is characterized in that projects scoring includes that item is completed in student
Purpose scoring, and the scoring for not completing project is zero or predetermined score.
6. deep learning method as described in claim 1, which is characterized in that further include:
Current whole cost to be paid is determined according to the sum of all remaining training durations.
7. deep learning method as described in claim 1, which is characterized in that the timing neural network model include RNN,
LSTM, GRU, two-way RNN or SRU.
8. a kind of deep learning device for driving training planning characterized by comprising
First preprocessing module obtains pretreatment student's feature for pre-processing to student's feature；
Second preprocessing module obtains pretreatment customization training feature for pre-processing to customization training feature；
When point module, for acquiring student in the features of construction schedule of the first pre- timing point to the pre- timing point of N；
Prediction module, for using the pretreatment student feature, pretreatment customization training feature and the features of construction schedule as defeated
Enter, utilizes timing Neural Network model predictive residue training duration.
9. a kind of electronic equipment characterized by comprising
One or more processors；
Storage device, for storing one or more programs；
When one or more of programs are executed by one or more of processors, so that one or more of processors are real
The now deep learning method as described in any in claim 1-7.
10. a kind of computer-readable medium, is stored thereon with computer program, which is characterized in that described program is held by processor
The deep learning method as described in any in claim 1-7 is realized when row.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910204504.8A CN109948930A (en) | 2019-03-18 | 2019-03-18 | Deep learning method and its application for driving training planning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910204504.8A CN109948930A (en) | 2019-03-18 | 2019-03-18 | Deep learning method and its application for driving training planning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109948930A true CN109948930A (en) | 2019-06-28 |
Family
ID=67009022
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910204504.8A Pending CN109948930A (en) | 2019-03-18 | 2019-03-18 | Deep learning method and its application for driving training planning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109948930A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111079020A (en) * | 2019-12-19 | 2020-04-28 | 联想(北京)有限公司 | Output method and device and electronic equipment |
| CN111158068A (en) * | 2019-12-31 | 2020-05-15 | 哈尔滨工业大学（深圳） | Short-term prediction method and system based on simple convolutional recurrent neural network |
| CN112905758A (en) * | 2021-01-29 | 2021-06-04 | 深圳通联金融网络科技服务有限公司 | Intelligent training management method and system based on telephone robot |
-
2019
- 2019-03-18 CN CN201910204504.8A patent/CN109948930A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111079020A (en) * | 2019-12-19 | 2020-04-28 | 联想(北京)有限公司 | Output method and device and electronic equipment |
| CN111158068A (en) * | 2019-12-31 | 2020-05-15 | 哈尔滨工业大学（深圳） | Short-term prediction method and system based on simple convolutional recurrent neural network |
| CN111158068B (en) * | 2019-12-31 | 2022-09-23 | 哈尔滨工业大学（深圳） | Short-term prediction method and system based on simple convolution cyclic neural network |
| CN112905758A (en) * | 2021-01-29 | 2021-06-04 | 深圳通联金融网络科技服务有限公司 | Intelligent training management method and system based on telephone robot |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Tolan et al. | Measuring the occupational impact of ai: tasks, cognitive abilities and ai benchmarks | |
| Tang et al. | An exploratory analysis of the latent structure of process data via action sequence autoencoders | |
| Lesh et al. | Assessing authentic mathematical performance | |
| Schwartz et al. | Perspectives on the open standards for the practice of conservation | |
| CN108701293A (en) | System and method for the data-driven identification to the talent | |
| CN109948930A (en) | Deep learning method and its application for driving training planning | |
| CN109543841A (en) | Deep learning method, apparatus, electronic equipment and computer-readable medium | |
| KR102506132B1 (en) | Method and device for personalized learning amount recommendation based on self-attention mechanism | |
| CN109783704A (en) | Man-machine mixed answer method, system, device | |
| CN107273979A (en) | The method and system of machine learning prediction are performed based on service class | |
| Winn | Model-based machine learning | |
| Welling | Are ML and statistics complementary | |
| KR102563986B1 (en) | Method and apparatus for recommending learning amount using artificial intelligence and clustering using k-means algorithm at the same time | |
| Simeunović et al. | Using data mining to predict success in studying | |
| Yang et al. | Machine learning-based student modeling methodology for intelligent tutoring systems | |
| Cheng | Data-mining research in education | |
| Pistilli et al. | Guiding early and often: Using curricular and learning analytics to shape teaching, learning, and student success in gateway courses | |
| Grossman et al. | A student-centered approach to the business school management science course | |
| Shapovalova et al. | Adaptive testing model as the method of quality knowledge control individualizing | |
| Zhang et al. | Students’ adaptive deep learning path and teaching strategy of contemporary ceramic art under the background of Internet+ | |
| Jayanthi et al. | Research contemplate on educational data mining | |
| Akhavan et al. | Knowledge management and adaptive e-learning: Iranian Schools Case Study | |
| Aslan et al. | Effect of Bayesian Student Modeling on Academic Achievement in Foreign Language Teaching (University Level English Preparatory School Example). | |
| Bell et al. | Assessment of extended tasks | |
| Sokolov et al. | Formation of the Electronic Educational Trajectory for Maritime Students |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190628 |