CN109870164A - Navigation terminal and its route preferences prediction technique - Google Patents
Navigation terminal and its route preferences prediction technique Download PDFInfo
- Publication number
- CN109870164A CN109870164A CN201711251015.5A CN201711251015A CN109870164A CN 109870164 A CN109870164 A CN 109870164A CN 201711251015 A CN201711251015 A CN 201711251015A CN 109870164 A CN109870164 A CN 109870164A
- Authority
- CN
- China
- Prior art keywords
- data
- training
- neural network
- route
- route preferences
- 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
Landscapes
- Navigation (AREA)
Abstract
This application involves a kind of navigation terminal and its route preferences prediction technique, the route preferences prediction technique includes: to obtain the history preference data of active user when navigation terminal starts navigation programming；Learning training is carried out using neural network model to the history preference data；Planning forecast, which is carried out, using trained neural network model obtains the navigation programming route for meeting active user's route preferences；The navigation programming route is prompted to the active user.The application can accomplish that, according to different user progress intelligent planning, provided route planning can largely meet the preference demand of user well, can be according to the subsequent navigational suggestion scheme of actual travel situation adjust automatically of user.
Description
Technical field
This application involves navigation system technical fields, and in particular to a kind of navigation terminal, a kind of route of navigation terminal are inclined
Good prediction technique.
Background technique
With the development of society, the improvement of people's living standards, automobile increasingly becomes the important traffic work of people's trip
Tool.But many times user knows nothing the destination to be gone or real-time road, road are opened and closed situation etc. and all can not
It learns in time, therefore, the Related product of navigation is come into being.
Wherein, the prior art navigation device based on GPS (global positioning system) is well-known, and is widely used as vehicle
Interior navigation system.In general, the navigation device based on GPS is related to a kind of computing device, the computing device with it is external (or
It is internal) it can determine its global location in the functional connection of GPS receiver.In addition, the computing device can determine beginning
Route between address and destination-address, the address can be inputted by the user of computing device.In general, the computing device by
Software enables, for from " best " between map data base calculating start address position and destination-address position or " most
It is excellent " route." best " or " optimal " route is the determination based on predetermined criterion, and without that must be most fast or minimal path.
Its preferred routes that also may be based on the route previously driven and determination.
In the prior art, navigation terminal is by using the location information derived from GPS receiver, computing device applicable rule
Interval determine its position, and the vehicles or the current location of user itself can be shown to user.Moreover, its can provide about
How to be navigated by appropriate navigation direction the instruction of determined route, described instruction shows over the display and/or generates
As the earcon (for example, " turning left at the 100m of front ") from loudspeaker.Describe the figure (example of movement to be done
Such as, the left arrow that instruction front is turned left) it can be shown in status bar, and applicable crossroad can also be overlapped in map itself
On mouth/turning etc..This is highly useful in the case where the vehicles meet with construction operation or heavy congestion.In specific work
During work, user can select the route disposed by navigation device, such as from " normal " mode and " quick " mode (Fast Modular
Formula calculates route in the shortest time, but can't detect the alternative route as normal mode as many) in make a choice.Or
Person, user, which more gladly may be calculated to be suitble to go on a tour by device, sees the route of landscape on the way.
However, can not accomplish to carry out intelligent planning according to different user in terms of the route planning of the prior art, it is provided
Route planning is not necessarily able to satisfy the preference demand of user, in addition user wish by oneself idea when driving, no matter how
Operation can not all change the rigid programme path of navigation terminal, cause user experience very bad.
Summary of the invention
The purpose of the application is, provides a kind of navigation terminal and its route preferences prediction technique, can solve above-mentioned
Technical problem can accomplish that provided route planning can be largely according to different user progress intelligent planning well
Meet the preference demand of user, it can be according to the subsequent navigational suggestion scheme of actual travel situation adjust automatically of user.
In order to solve the above technical problems, the application provides a kind of route preferences prediction technique of navigation terminal, the route
Preference prediction technique includes:
When navigation terminal starts navigation programming, the history preference data of active user is obtained；
Learning training is carried out using neural network model to the history preference data；
Planning forecast is carried out using trained neural network model to obtain meeting leading for active user's route preferences
Navigate programme path；
The navigation programming route is prompted to the active user.
Wherein, the history preference data include whether to hide congestion, under jam situation whether to be inclined apart from length
Good, the different route preferences of morning and evening peak period, the preference of overpass and ground section, long-distance high speed and provincial highway selection preference,
Whether route preferences, active user under different weather environment are actively ready to detour in share-car and any two kinds of feelings
The preference of active user when condition combination of the above.
Wherein, described the step of learning training is carried out using neural network model to the history preference data, specific to wrap
It includes:
The history preference data is subjected to data cleansing, normalization；
History preference data after cleaning normalization is divided into training dataset and test data set according to the time；
Off-line model training is carried out to the training data of the training dataset, shot and long term memory nerve net is respectively trained
Multiple neural network models of network LSTM；
Training data is obtained for the prediction value list of multiple neural network models output after training, by the predicted value
List is compared with actual route preferences value, shared power when multiple neural network models are calculated as built-up pattern
Weight values；
Using the test data of test data set to multiple neural network model assessment prediction effects in built-up pattern, root
It is predicted that the weighted value that effect is shared when adjusting the multiple neural network model as built-up pattern.
Wherein, shared weight when adjusting the multiple neural network model as built-up pattern according to prediction effect
After value, further includes:
The route preferences specific value of the following predetermined amount of time is predicted using the mode of rolling time window.
Wherein, the mode using rolling window predicts the route preferences of the following predetermined amount of time, comprising:
The amount of increase and amount of decrease of Combined model forecast is converted to the prediction numerical value for being predicted the moment, then the prediction that current predictive is gone out
Numerical value inserts next time window for being predicted the moment, and alternate cycles according to this；
When getting the actual numerical value of active user's actual change trend, prediction numerical value and actual numerical value are compared, and
According to the comparing result training data that actual numerical value is new as one group, model is substituted into update model parameter.
Wherein, described that the history preference data is subjected to data cleansing, normalization, it specifically includes:
According to the data distribution feature of active user, using receiving-refusal method of sampling, with reference to other users cost/
Time optimization navigation data cleans the data of active user described in normalizing.
Wherein, the history preference data after the normalization by cleaning is divided into training dataset and test number according to the time
According to collection, comprising:
Early time data before time in the history preference data is located at given time is divided into training dataset, will
The time is located at the advanced stage data after given time and is divided into test data set in the history preference data.
Wherein: the training data to the training dataset carries out before off-line model training, further includes: uses instruction
The training data for practicing data set generates the time series data of different spans；
The training data to the training dataset carries out off-line model training, and correspondence includes:
Multiple nerves of LSTM are respectively trained using every part of time series data in the time series data of different spans
Network model.
Wherein, the training data to the training dataset carries out off-line model training, specifically includes:
The training data of the training dataset is trained using the distributed training method calculated based on memory,
In, training data is distributed on each node and the original model parameter of neural network model is broadcast to each node, often
A node obtains current gradient and model parameter renewal amount according to the training data of current model parameter and certain scale, leads to
The model parameter renewal amount for summarizing each node feeding back is crossed to update model parameter, and updated model parameter is broadcast to respectively
A node, iterative repetition according to this, according to the training for requiring to complete single LSTM neural network model；
Shared weighted value when multiple neural network models are calculated as built-up pattern, specifically includes:
It obtains each LSTM neural network model using the method for linear regression by the training data of multiple periods and exists
Weighted value in final built-up pattern output.
In order to solve the above technical problems, the application also provides a kind of navigation terminal, the navigation terminal includes processor, institute
Processor is stated for executing program data, the step of to realize any above-mentioned route preferences prediction technique, the navigation terminal
For smart phone, tablet computer or vehicle-mounted computer.
The application navigation terminal and its route preferences prediction technique are used when navigation terminal starts navigation programming current
The history preference data at family carries out learning training using neural network model, is planned using trained neural network model
Prediction obtains the navigation programming route for meeting active user's route preferences, and the navigation programming route is prompted to described work as
Preceding user.In this way, the application can accomplish to carry out intelligent planning, provided route according to different user well
Planning can largely meet the preference demand of user, can be according to the subsequent navigation of actual travel situation adjust automatically of user
Proposed projects.In addition, the neural network model that the application uses, can speculate according to the historical data of user and use more accurately
The best preference at family is predicted the next behavior of user, is largely realized artificial by analyzing the previous behavior of user
It is intelligent.
Above description is only the general introduction of technical scheme, in order to better understand the technological means of the application,
And it can be implemented in accordance with the contents of the specification, and in order to allow the above and other objects, features and advantages of the application can
It is clearer and more comprehensible, it is special below to lift preferred embodiment, and cooperate attached drawing, detailed description are as follows.
Detailed description of the invention
Fig. 1 is the flow diagram of the route preferences prediction technique of the application navigation terminal.
Fig. 2 is the module diagram of the application navigation terminal.
Specific embodiment
Further to illustrate that the application is the technological means and effect reaching predetermined application purpose and being taken, below in conjunction with
Attached drawing and preferred embodiment, the specific embodiment that foundation the application navigation terminal and its route preferences prediction technique are proposed,
Method, step, structure, feature and its effect, detailed description are as follows.
Aforementioned and other technology contents, feature and effect in relation to the application refer to the preferable reality of schema in following cooperation
Applying in the detailed description of example can clearly appear from.By the explanation of specific embodiment, when can be that reach predetermined mesh to the application
The technological means taken and effect be able to more deeply and it is specific understand, however institute's accompanying drawings are only to provide with reference to and say
It is bright to be used, not it is used to limit the application.
Referring to Fig. 1, Fig. 1 is the flow diagram of the route preferences prediction technique of the application navigation terminal, the application is real
Applying route preferences prediction technique described in mode includes but is not limited to the following steps.
Step S101 obtains the history preference data of active user when navigation terminal starts navigation programming.
It should be strongly noted that history preference data described in present embodiment, can specifically include whether hide congestion,
Under jam situation whether using apart from length as preference, the different route preferences of morning and evening peak period, overpass and ground section
Preference, long-distance high speed and provincial highway selection preference, the route preferences under different weather environment, active user are in share-car
No active is ready to detour and the preference of whens any two kinds of situation combination of the above active user.
Wherein, the history preference data can store in navigation terminal itself, be stored in network side server
In, it is not limited thereto.In addition, intelligent knowledge can be carried out by fingerprint, account number or other modes for the identity of user
Not, for example active user is when using navigation terminal, logs in its personal system using fingerprint, and then navigation terminal is by active user
All data all synchronize update, for subsequent use.
Step S102 carries out learning training using neural network model to the history preference data.
Specifically, in step s 102, described that the history preference data is learnt using neural network model
Trained step can specifically include following treatment processes:
The history preference data is subjected to data cleansing, normalization；
History preference data after cleaning normalization is divided into training dataset and test data set according to the time；
Off-line model training is carried out to the training data of the training dataset, shot and long term memory nerve net is respectively trained
Multiple neural network models of network LSTM (Long Short-Term Memory, i.e. shot and long term Memory Neural Networks)；
Training data is obtained for the prediction value list of multiple neural network models output after training, by the predicted value
List is compared with actual route preferences value, shared power when multiple neural network models are calculated as built-up pattern
Weight values；
Using the test data of test data set to multiple neural network model assessment prediction effects in built-up pattern, root
It is predicted that the weighted value that effect is shared when adjusting the multiple neural network model as built-up pattern.
It should be pointed out that adjusting the multiple neural network model as group according to prediction effect described in present embodiment
It can also include: the mode using rolling time window to the following predetermined amount of time when molding type after shared weighted value
Route preferences specific value is predicted.
Specifically, the mode using rolling window predicts the route preferences of the following predetermined amount of time,
It may include following treatment processes in present embodiment: the amount of increase and amount of decrease of Combined model forecast be converted to the prediction for being predicted the moment
Numerical value, then the prediction numerical value that current predictive is gone out insert next time window for being predicted the moment, and alternate cycles according to this；When
When getting the actual numerical value of active user's actual change trend, prediction numerical value and actual numerical value are compared, and tied according to comparison
The fruit training data that actual numerical value is new as one group substitutes into model to update model parameter.
It is noted that the history preference data is carried out data cleansing, normalization, tool described in present embodiment
Body may include following treatment processes: according to the data distribution feature of active user, use receiving-refusal method of sampling, reference
The cost of other users/time optimization navigation data cleans the data of active user described in normalizing.In other words, this implementation
Mode can go to judge whether there is the navigation number of other optimizations with reference to the driving cost or driving time of other users
According to being added in prediction result with lesser weight, selected for user, the usage experience of user is enriched with this.
It should be noted that the history preference data after cleaning normalization is divided into according to the time described in present embodiment
Training dataset and test data set, can specifically include: the time in the history preference data is located at before given time
Early time data be divided into training dataset, the time in the history preference data is located at the advanced stage data after given time
It is divided into test data set.For example, given time can be the moment such as season alternation, on and off duty or month.
In a preferred embodiment, the training data to the training dataset carries out off-line model training
Before, it can also include: the time series data that different spans are generated using the training data of training dataset.Correspondingly, described
Off-line model training is carried out to the training data of the training dataset, correspondence may include: the time sequence using different spans
Multiple neural network models of LSTM are respectively trained in every part of time series data in column data.
In the present embodiment, the training data to the training dataset carries out off-line model training, specifically may be used
To include: to be trained to the training data of the training dataset using the distributed training method calculated based on memory,
In, training data is distributed on each node and the original model parameter of neural network model is broadcast to each node, often
A node obtains current gradient and model parameter renewal amount according to the training data of current model parameter and certain scale, leads to
The model parameter renewal amount for summarizing each node feeding back is crossed to update model parameter, and updated model parameter is broadcast to respectively
A node, iterative repetition according to this, according to the training for requiring to complete single LSTM neural network model.
Correspondingly, shared weight when multiple neural network models being calculated described in present embodiment as built-up pattern
Value, can specifically include: obtaining each LSTM nerve net using the method for linear regression by the training data of multiple periods
Weighted value of the network model in final built-up pattern output.
Step S103 carries out planning forecast using trained neural network model and obtains meeting active user's route
The navigation programming route of preference.
It should be noted that present embodiment training obtain the built-up pattern of LSTM neural network model after, according to institute
The starting point and terminal for stating active user carry out unified plan prediction according to all situations factor that may be present on the way.
The navigation programming route is prompted to the active user by step S104.
It is noted that the navigation programming route can be for one, two or more, it can be by different
Display mode reminds user, so that user carries out final choice.
In the present embodiment, such as in one of application examples, navigation terminal is in active user's driving process, analysis
The route selection preference of active user, although such as when congestion can select a little stifled route often walked, or can attempt to detour
Avoid congested link；Or morning and evening summit selects selection when overhead section or ground section or long-distance running to walk high speed also
It is preferences such as provincial highway etc., and cooks up and be best suitable for using the preference of active user as the condition of screening route when calculating route
The route of active user's habit.
The application can accomplish that, according to different user progress intelligent planning, provided route planning can larger journey well
Meet the preference demand of user on degree, it can be according to the subsequent navigational suggestion scheme of actual travel situation adjust automatically of user.This
Outside, the neural network model that the application uses can speculate that user's is best inclined according to the historical data of user more accurately
It is good, by analyzing the previous behavior of user, predicts the next behavior of user, largely realize manual intelligent.
Please refer to figure 2, and Fig. 2 is the module diagram of the application navigation terminal.In the present embodiment, the navigation
Terminal includes but is not limited to processor 21, memory 22, display 23 and loudspeaker 24, and the navigation terminal can be intelligent hand
Machine, navigator, tablet computer or vehicle-mounted computer etc..
Wherein, for the processor 21 for executing program data, the route preferences to realize any above embodiment are pre-
The step of survey method.
The memory 22 can be used for storing above procedure data, can be used for a number of storage different user
According to.
Navigation programming route can be shown to active user by the display 23, or will be led by the loudspeaker 24
Boat programme path plays to active user.The display 23 can be curved-surface display device, projection display equipment or virtual
Reality display device is not limited thereto.
Specifically, the processor 21 obtains the history preference of active user when navigation terminal starts navigation programming
Data.
It should be strongly noted that history preference data described in present embodiment, can specifically include whether hide congestion,
Under jam situation whether using apart from length as preference, the different route preferences of morning and evening peak period, overpass and ground section
Preference, long-distance high speed and provincial highway selection preference, the route preferences under different weather environment, active user are in share-car
No active is ready to detour and the preference of whens any two kinds of situation combination of the above active user.
Wherein, the history preference data can store in navigation terminal itself, be stored in network side server
In, it is not limited thereto.In addition, intelligent knowledge can be carried out by fingerprint, account number or other modes for the identity of user
Not, for example active user is when using navigation terminal, logs in its personal system using fingerprint, and then navigation terminal is by active user
All data all synchronize update, for subsequent use.
The processor 21 carries out learning training using neural network model to the history preference data.
Specifically, the processor 21 carries out learning training using neural network model to the history preference data
Step can specifically include following treatment processes:
The history preference data is subjected to data cleansing, normalization；
History preference data after cleaning normalization is divided into training dataset and test data set according to the time；
Off-line model training is carried out to the training data of the training dataset, multiple nerve nets of LSTM are respectively trained
Network model；
Training data is obtained for the prediction value list of multiple neural network models output after training, by the predicted value
List is compared with actual route preferences value, shared power when multiple neural network models are calculated as built-up pattern
Weight values；
Using the test data of test data set to multiple neural network model assessment prediction effects in built-up pattern, root
It is predicted that the weighted value that effect is shared when adjusting the multiple neural network model as built-up pattern.
It should be pointed out that adjusting the multiple neural network model as group according to prediction effect described in present embodiment
It can also include: that the processor 21 uses the mode of rolling time window to future when molding type after shared weighted value
The route preferences specific value of predetermined amount of time is predicted.
Specifically, the mode using rolling window predicts the route preferences of the following predetermined amount of time,
May include following treatment processes in present embodiment: the processor 21 is converted to the amount of increase and amount of decrease of Combined model forecast pre-
The prediction numerical value at moment, then the prediction numerical value that current predictive is gone out are surveyed, inserts next time window for being predicted the moment, and according to this
Alternate cycles；When getting the actual numerical value of active user's actual change trend, prediction numerical value and actual numerical value are compared, and
According to the comparing result training data that actual numerical value is new as one group, model is substituted into update model parameter.
It is noted that processor 21 described in present embodiment by the history preference data carry out data cleansing,
Normalization, can specifically include following treatment processes: the processor 21 is according to the data distribution feature of active user, using connecing
By-refuse the method for sampling, with reference to cost/time optimization navigation data of other users, clean active user's described in normalizing
Data.In other words, present embodiment can go to judge whether there is with reference to the driving cost or driving time of other users
Other navigation datas optimized are added in prediction result with lesser weight, are selected for user, enrich user with this
Usage experience.
It should be noted that processor 21 described in present embodiment will cleaning normalization after history preference data according to when
Between be divided into training dataset and test data set, can specifically include: the processor 21 will be in the history preference data
Time be located at given time before early time data be divided into training dataset, by the time in the history preference data be located at refer to
Advanced stage data after timing is carved are divided into test data set.For example, given time can be season alternation, it is on and off duty or
The moment such as month.
In a preferred embodiment, the processor 21 carries out off-line model to the training data of the training dataset
It can also include: the time sequence that the processor 21 generates different spans using the training data of training dataset before training
Column data.Correspondingly, the processor 21 carries out off-line model training to the training data of the training dataset, and correspondence can be with
It include: that the processor 21 using every part of time series data in the time series data of different spans is respectively trained LSTM's
Multiple neural network models.
In the present embodiment, the processor 21 carries out off-line model instruction to the training data of the training dataset
Practice, can specifically include: the processor 21 is to the training data of the training dataset using the distribution calculated based on memory
Formula training method is trained, wherein training data is distributed on each node and by the initial model of neural network model
Parameter is broadcast to each node, and each node obtains according to the training data of current model parameter and certain scale and works as front ladder
Degree and model parameter renewal amount, update model parameter by summarizing the model parameter renewal amount of each node feeding back, and will more
Model parameter after new is broadcast to each node, according to this iterative repetition, with according to requiring to complete single LSTM neural network model
Training.
Correspondingly, multiple neural network models are calculated as built-up pattern when institute in processor 21 described in present embodiment
The weighted value accounted for, can specifically include: the processor 21 passes through the training data of multiple periods, uses the side of linear regression
Method obtains weighted value of each LSTM neural network model in final built-up pattern output.
The processor 21 carries out planning forecast using trained neural network model and obtains meeting the active user
The navigation programming route of route preferences.
It should be noted that present embodiment training obtain the built-up pattern of LSTM neural network model after, according to institute
The starting point and terminal for stating active user carry out unified plan prediction according to all situations factor that may be present on the way.
Then, the navigation programming route is prompted to the current use by the display 23 and/or the loudspeaker 24
Family.
It is noted that the navigation programming route can be for one, two or more, it can be by different
Display mode reminds user, so that user carries out final choice.
The application can accomplish that, according to different user progress intelligent planning, provided route planning can larger journey well
Meet the preference demand of user on degree, it can be according to the subsequent navigational suggestion scheme of actual travel situation adjust automatically of user.This
Outside, the neural network model that the application uses can speculate that user's is best inclined according to the historical data of user more accurately
It is good, by analyzing the previous behavior of user, predicts the next behavior of user, largely realize manual intelligent.
The application also provides a kind of vehicle, and navigation terminal described in above-described embodiment can be set.
As previously mentioned, the navigation terminal includes but is not limited to processor 21, memory 22, display 23 and loudspeaker
24, in the present embodiment, the navigation terminal can be vehicle-mounted computer on vehicle or to connect with the vehicle network
Dedicated navigator.
Wherein, for the processor 21 for executing program data, the route preferences to realize any above embodiment are pre-
The step of survey method.
The memory 22 can be used for storing above procedure data, can be used for a number of storage different user
According to.
Navigation programming route can be shown to active user by the display 23, or will be led by the loudspeaker 24
Boat programme path plays to active user.The display 23 can be curved-surface display device, projection display equipment or virtual
Reality display device is not limited thereto.
Specifically, the processor 21 obtains the history preference of active user when navigation terminal starts navigation programming
Data.
It should be strongly noted that history preference data described in present embodiment, can specifically include whether hide congestion,
Under jam situation whether using apart from length as preference, the different route preferences of morning and evening peak period, overpass and ground section
Preference, long-distance high speed and provincial highway selection preference, the route preferences under different weather environment, active user are in share-car
No active is ready to detour and the preference of whens any two kinds of situation combination of the above active user.
Wherein, the history preference data can store in navigation terminal itself, be stored in network side server
In, it is not limited thereto.In addition, intelligent knowledge can be carried out by fingerprint, account number or other modes for the identity of user
Not, for example active user is when using navigation terminal, logs in its personal system using fingerprint, and then navigation terminal is by active user
All data all synchronize update, for subsequent use.
The processor 21 carries out learning training using neural network model to the history preference data.
Specifically, the processor 21 carries out learning training using neural network model to the history preference data
Step can specifically include following treatment processes:
The history preference data is subjected to data cleansing, normalization；
History preference data after cleaning normalization is divided into training dataset and test data set according to the time；
Off-line model training is carried out to the training data of the training dataset, multiple nerve nets of LSTM are respectively trained
Network model；
Training data is obtained for the prediction value list of multiple neural network models output after training, by the predicted value
List is compared with actual route preferences value, shared power when multiple neural network models are calculated as built-up pattern
Weight values；
Using the test data of test data set to multiple neural network model assessment prediction effects in built-up pattern, root
It is predicted that the weighted value that effect is shared when adjusting the multiple neural network model as built-up pattern.
It should be pointed out that adjusting the multiple neural network model as group according to prediction effect described in present embodiment
It can also include: that the processor 21 uses the mode of rolling time window to future when molding type after shared weighted value
The route preferences specific value of predetermined amount of time is predicted.
Specifically, the mode using rolling window predicts the route preferences of the following predetermined amount of time,
May include following treatment processes in present embodiment: the processor 21 is converted to the amount of increase and amount of decrease of Combined model forecast pre-
The prediction numerical value at moment, then the prediction numerical value that current predictive is gone out are surveyed, inserts next time window for being predicted the moment, and according to this
Alternate cycles；When getting the actual numerical value of active user's actual change trend, prediction numerical value and actual numerical value are compared, and
According to the comparing result training data that actual numerical value is new as one group, model is substituted into update model parameter.
It is noted that processor 21 described in present embodiment by the history preference data carry out data cleansing,
Normalization, can specifically include following treatment processes: the processor 21 is according to the data distribution feature of active user, using connecing
By-refuse the method for sampling, with reference to cost/time optimization navigation data of other users, clean active user's described in normalizing
Data.In other words, present embodiment can go to judge whether there is with reference to the driving cost or driving time of other users
Other navigation datas optimized are added in prediction result with lesser weight, are selected for user, enrich user with this
Usage experience.
It should be noted that processor 21 described in present embodiment will cleaning normalization after history preference data according to when
Between be divided into training dataset and test data set, can specifically include: the processor 21 will be in the history preference data
Time be located at given time before early time data be divided into training dataset, by the time in the history preference data be located at refer to
Advanced stage data after timing is carved are divided into test data set.For example, given time can be season alternation, it is on and off duty or
The moment such as month.
In a preferred embodiment, the processor 21 carries out off-line model to the training data of the training dataset
It can also include: the time sequence that the processor 21 generates different spans using the training data of training dataset before training
Column data.Correspondingly, the processor 21 carries out off-line model training to the training data of the training dataset, and correspondence can be with
It include: that the processor 21 using every part of time series data in the time series data of different spans is respectively trained LSTM's
Multiple neural network models.
In the present embodiment, the processor 21 carries out off-line model instruction to the training data of the training dataset
Practice, can specifically include: the processor 21 is to the training data of the training dataset using the distribution calculated based on memory
Formula training method is trained, wherein training data is distributed on each node and by the initial model of neural network model
Parameter is broadcast to each node, and each node obtains according to the training data of current model parameter and certain scale and works as front ladder
Degree and model parameter renewal amount, update model parameter by summarizing the model parameter renewal amount of each node feeding back, and will more
Model parameter after new is broadcast to each node, according to this iterative repetition, with according to requiring to complete single LSTM neural network model
Training.
Correspondingly, multiple neural network models are calculated as built-up pattern when institute in processor 21 described in present embodiment
The weighted value accounted for, can specifically include: the processor 21 passes through the training data of multiple periods, uses the side of linear regression
Method obtains weighted value of each LSTM neural network model in final built-up pattern output.
The processor 21 carries out planning forecast using trained neural network model and obtains meeting the active user
The navigation programming route of route preferences.
It should be noted that present embodiment training obtain the built-up pattern of LSTM neural network model after, according to institute
The starting point and terminal for stating active user carry out unified plan prediction according to all situations factor that may be present on the way.
Then, the navigation programming route is prompted to the current use by the display 23 and/or the loudspeaker 24
Family.
It is noted that the navigation programming route can be for one, two or more, it can be by different
Display mode reminds user, so that user carries out final choice.
The application can accomplish that, according to different user progress intelligent planning, provided route planning can larger journey well
Meet the preference demand of user on degree, it can be according to the subsequent navigational suggestion scheme of actual travel situation adjust automatically of user.This
Outside, the neural network model that the application uses can speculate that user's is best inclined according to the historical data of user more accurately
It is good, by analyzing the previous behavior of user, predicts the next behavior of user, largely realize manual intelligent.
Then, one concrete application example of the application is explained below.
Route preferences prediction technique in the application example, by processor 21 according to the road of the map data base in memory 22
Segment data plans the route between departure place and destination, comprising: several ways line is wrapped between analysis departure place and destination
The weight of several candidate road sections and each candidate road section that contain and at least one road type；From several candidates
Choose at least one specified section in section and receive the weight option for corresponding to specified section, weight option promoted indicating or
Reduce the weight with the candidate road section of specified section road type having the same；Recognize road class belonging to the specified section
Type；According to the road type in specified section and corresponding weight option, adjustment includes road type identical with specified section
Each candidate road section weight, with generate respectively each candidate road section one update weight, gravity treatment item of holding power instruction promoted should
When weight, the weight of candidate road section increases according to corresponding weight proportion, when gravity treatment item of holding power instruction reduces the weight, candidate road
The weight of section is reduced according to corresponding weight proportion；And it is inclined according to the update weight of candidate road section and the history of active user
Good data, the navigation programming route of planning departure place to the preference between destination.
The section data of the map data base can be two dimensional map data (latitude and longitude), but also may include
Three dimensionality (height), the map datum can further include additional information, for example, about the letter for adding oil/gas station, point of interest
Breath, the map datum also may include the information about the shape of the building and object along road.
In addition, navigation terminal also may include input unit, for example, touch screen, allows the user to call navigation menu
(not shown).From this menu, other navigation features can be originated or controlled.Allow to be easy to called menu screen from itself
In (for example, from map display to menu screen need a step) selection navigation feature greatly simplify user interaction and
Keep it very fast and is easier to.The navigation menu includes the option that destination is inputted for user.
It should be strongly noted that the application is when carrying out route preferences prediction, used route calculation algorithm is described
Algorithm is handling many potential different routes, such as complete Mode scans, the museum with scenic route, process
The potential different navigation programme path is assessed with the user-defined criterion such as no speed camera (or device default value).
The above is only the preferred embodiment of the application, not makes any form of restriction to the application, though
Right the application has been disclosed in a preferred embodiment above, however is not limited to the application, any technology people for being familiar with this profession
Member, is not departing within the scope of technical scheme, when the technology contents using the disclosure above make a little change or modification
For the equivalent embodiment of equivalent variations, but all technical spirits pair without departing from technical scheme content, according to the application
Any simple modification, equivalent change and modification made by above embodiments, in the range of still falling within technical scheme.
Claims (10)
1. a kind of route preferences prediction technique of navigation terminal, which is characterized in that the route preferences prediction technique includes:
When navigation terminal starts navigation programming, the history preference data of active user is obtained；
Learning training is carried out using neural network model to the history preference data；
Planning forecast, which is carried out, using trained neural network model obtains the navigation rule for meeting active user's route preferences
Draw route；
The navigation programming route is prompted to the active user.
2. route preferences prediction technique according to claim 1, which is characterized in that the history preference data includes whether
Hide congestion, under jam situation whether using apart from length as preference, different route preferences, overpass and the ground of morning and evening peak period
The preference in face section, long-distance high speed and provincial highway selection preference, the route preferences under different weather environment, active user are in share-car
In the case of whether be actively ready to detour and the preference of whens any two kinds of situation combination of the above active user.
3. route preferences prediction technique according to claim 1 or 2, which is characterized in that described to the history preference number
According to the step of carrying out learning training using neural network model, specifically include:
The history preference data is subjected to data cleansing, normalization；
History preference data after cleaning normalization is divided into training dataset and test data set according to the time；
Off-line model training is carried out to the training data of the training dataset, shot and long term Memory Neural Networks are respectively trained
Multiple neural network models of LSTM；
Training data is obtained for the prediction value list of multiple neural network models output after training, by the prediction value list
It is compared with actual route preferences value, shared weight when multiple neural network models are calculated as built-up pattern
Value；
Using the test data of test data set to multiple neural network model assessment prediction effects in built-up pattern, according to pre-
Survey weighted value shared when effect adjusts the multiple neural network model as built-up pattern.
4. route preferences prediction technique according to claim 3, which is characterized in that it is described adjusted according to prediction effect described in
When multiple neural network models are as built-up pattern after shared weighted value, further includes:
The route preferences specific value of the following predetermined amount of time is predicted using the mode of rolling time window.
5. route preferences prediction technique according to claim 4, which is characterized in that the mode pair using rolling window
The route preferences of the following predetermined amount of time are predicted, comprising:
The amount of increase and amount of decrease of Combined model forecast is converted to the prediction numerical value for being predicted the moment, then the prediction number that current predictive is gone out
Value inserts next time window for being predicted the moment, and alternate cycles according to this；
When getting the actual numerical value of active user's actual change trend, prediction numerical value and actual numerical value are compared, and according to
The comparing result training data that actual numerical value is new as one group substitutes into model to update model parameter.
6. route preferences prediction technique according to claim 3, which is characterized in that it is described by the history preference data into
Row data cleansing, normalization, specifically include:
According to the data distribution feature of active user, using receiving-refusal method of sampling, with reference to cost/time of other users
Navigation data is optimized, the data of active user described in normalizing are cleaned.
7. route preferences prediction technique according to claim 3, which is characterized in that the history after the normalization by cleaning
Preference data is divided into training dataset and test data set according to the time, comprising:
Early time data before time in the history preference data is located at given time is divided into training dataset, will be described
The time is located at the advanced stage data after given time and is divided into test data set in history preference data.
8. route preferences prediction technique according to claim 3, it is characterised in that:
The training data to the training dataset carries out before off-line model training, further includes:
The time series data of different spans is generated using the training data of training dataset；
The training data to the training dataset carries out off-line model training, and correspondence includes:
Multiple neural networks of LSTM are respectively trained using every part of time series data in the time series data of different spans
Model.
9. route preferences prediction technique according to claim 3, which is characterized in that the instruction to the training dataset
Practice data and carry out off-line model training, specifically includes:
The training data of the training dataset is trained using the distributed training method calculated based on memory, wherein
Training data is distributed on each node and the original model parameter of neural network model is broadcast to each node, Mei Gejie
Point obtains current gradient and model parameter renewal amount, passes through remittance according to the training data of current model parameter and certain scale
Updated model parameter is broadcast to each section to update model parameter by the model parameter renewal amount of total each node feeding back
Point, iterative repetition according to this, according to the training for requiring to complete single LSTM neural network model；
Shared weighted value when multiple neural network models are calculated as built-up pattern, specifically includes:
Each LSTM neural network model is obtained final using the method for linear regression by the training data of multiple periods
Built-up pattern output in weighted value.
10. a kind of navigation terminal, which is characterized in that the navigation terminal includes processor, and the processor is for executing program
Data, with realize according to claim 1-9 described in any item route preferences prediction techniques the step of, the navigation terminal be intelligence
It can mobile phone, tablet computer or vehicle-mounted computer.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711251015.5A CN109870164A (en) | 2017-12-01 | 2017-12-01 | Navigation terminal and its route preferences prediction technique |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711251015.5A CN109870164A (en) | 2017-12-01 | 2017-12-01 | Navigation terminal and its route preferences prediction technique |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109870164A true CN109870164A (en) | 2019-06-11 |
Family
ID=66913634
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711251015.5A Pending CN109870164A (en) | 2017-12-01 | 2017-12-01 | Navigation terminal and its route preferences prediction technique |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109870164A (en) |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110260864A (en) * | 2019-06-20 | 2019-09-20 | 维智汽车电子(天津)有限公司 | Construction method, device and the electronic equipment of optimal reference trace route |
| CN110570948A (en) * | 2019-09-09 | 2019-12-13 | 深圳市伊欧乐科技有限公司 | User future weight prediction method, device, server and storage medium |
| CN111982144A (en) * | 2020-08-20 | 2020-11-24 | 北京百度网讯科技有限公司 | Navigation method, navigation device, electronic equipment and computer readable medium |
| CN112393775A (en) * | 2021-01-21 | 2021-02-23 | 上海傅利叶智能科技有限公司 | Method and device for detecting user state based on intelligent walking stick and intelligent walking stick |
| CN112449676A (en) * | 2019-06-28 | 2021-03-05 | 谷歌有限责任公司 | Generating a navigation route and identifying carpool options according to a compromise between calculated parameters |
| CN112721743A (en) * | 2021-02-04 | 2021-04-30 | 武汉理工大学 | Fuel cell automobile remaining mileage prediction method based on charging-hydrogen using mode |
| CN113034894A (en) * | 2021-02-04 | 2021-06-25 | 招商新智科技有限公司 | ETC portal system, and highway section closing early warning method and device |
| CN113705827A (en) * | 2021-07-29 | 2021-11-26 | 阿里巴巴新加坡控股有限公司 | Model evaluation method and device and electronic equipment |
| CN113970338A (en) * | 2020-07-22 | 2022-01-25 | 阿里巴巴集团控股有限公司 | Travel mode recommendation method, related method, device and system |
| CN114153934A (en) * | 2021-12-13 | 2022-03-08 | 高德软件有限公司 | Machine learning model training, navigation route recommendation method and computer storage medium |
| CN114383617A (en) * | 2021-12-29 | 2022-04-22 | 深圳市跨越新科技有限公司 | Navigation route calculation method, device, equipment and storage medium based on map neural network |
| CN115038935A (en) * | 2020-02-04 | 2022-09-09 | 通腾导航技术股份有限公司 | Navigation system |
| WO2024179127A1 (en) * | 2023-02-28 | 2024-09-06 | 北京高德云图科技有限公司 | Navigation route recommendation method and apparatus, familiar-route prediction model training method and apparatus, and medium |
| CN118960774A (en) * | 2024-10-21 | 2024-11-15 | 南京英莫特信息科技有限公司 | Trip planning method based on vehicle historical navigation data |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101739839A (en) * | 2008-11-14 | 2010-06-16 | 上海经达实业发展有限公司 | Vehicle dynamic path navigational system |
| US20140189587A1 (en) * | 2012-12-28 | 2014-07-03 | Alticast Corporation | Method and apparatus for screen navigation history management |
| CN103956043A (en) * | 2014-04-29 | 2014-07-30 | 南京理工大学 | Auxiliary vehicle traveling path system based on mobile terminal |
| CN104165634A (en) * | 2014-07-28 | 2014-11-26 | 广州视源电子科技股份有限公司 | Path planning method based on user use habit |
| CN105139301A (en) * | 2015-08-13 | 2015-12-09 | 厦门维途信息技术有限公司 | Guiding method based on BP neural network |
| CN106225794A (en) * | 2016-06-30 | 2016-12-14 | 百度在线网络技术（北京）有限公司 | A kind of method and device for planning of navigation way |
| CN106991506A (en) * | 2017-05-16 | 2017-07-28 | 深圳先进技术研究院 | Intelligent terminal and its stock trend forecasting method based on LSTM |
-
2017
- 2017-12-01 CN CN201711251015.5A patent/CN109870164A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101739839A (en) * | 2008-11-14 | 2010-06-16 | 上海经达实业发展有限公司 | Vehicle dynamic path navigational system |
| US20140189587A1 (en) * | 2012-12-28 | 2014-07-03 | Alticast Corporation | Method and apparatus for screen navigation history management |
| CN103956043A (en) * | 2014-04-29 | 2014-07-30 | 南京理工大学 | Auxiliary vehicle traveling path system based on mobile terminal |
| CN104165634A (en) * | 2014-07-28 | 2014-11-26 | 广州视源电子科技股份有限公司 | Path planning method based on user use habit |
| CN105139301A (en) * | 2015-08-13 | 2015-12-09 | 厦门维途信息技术有限公司 | Guiding method based on BP neural network |
| CN106225794A (en) * | 2016-06-30 | 2016-12-14 | 百度在线网络技术（北京）有限公司 | A kind of method and device for planning of navigation way |
| CN106991506A (en) * | 2017-05-16 | 2017-07-28 | 深圳先进技术研究院 | Intelligent terminal and its stock trend forecasting method based on LSTM |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110260864A (en) * | 2019-06-20 | 2019-09-20 | 维智汽车电子(天津)有限公司 | Construction method, device and the electronic equipment of optimal reference trace route |
| CN110260864B (en) * | 2019-06-20 | 2021-11-09 | 维智汽车电子(天津)有限公司 | Method and device for constructing optimal reference trail route and electronic equipment |
| CN112449676A (en) * | 2019-06-28 | 2021-03-05 | 谷歌有限责任公司 | Generating a navigation route and identifying carpool options according to a compromise between calculated parameters |
| CN110570948A (en) * | 2019-09-09 | 2019-12-13 | 深圳市伊欧乐科技有限公司 | User future weight prediction method, device, server and storage medium |
| CN115038935A (en) * | 2020-02-04 | 2022-09-09 | 通腾导航技术股份有限公司 | Navigation system |
| CN113970338A (en) * | 2020-07-22 | 2022-01-25 | 阿里巴巴集团控股有限公司 | Travel mode recommendation method, related method, device and system |
| CN111982144A (en) * | 2020-08-20 | 2020-11-24 | 北京百度网讯科技有限公司 | Navigation method, navigation device, electronic equipment and computer readable medium |
| CN112393775A (en) * | 2021-01-21 | 2021-02-23 | 上海傅利叶智能科技有限公司 | Method and device for detecting user state based on intelligent walking stick and intelligent walking stick |
| CN112721743A (en) * | 2021-02-04 | 2021-04-30 | 武汉理工大学 | Fuel cell automobile remaining mileage prediction method based on charging-hydrogen using mode |
| CN113034894A (en) * | 2021-02-04 | 2021-06-25 | 招商新智科技有限公司 | ETC portal system, and highway section closing early warning method and device |
| CN113705827A (en) * | 2021-07-29 | 2021-11-26 | 阿里巴巴新加坡控股有限公司 | Model evaluation method and device and electronic equipment |
| CN113705827B (en) * | 2021-07-29 | 2024-04-12 | 阿里巴巴创新公司 | Model evaluation method and device and electronic equipment |
| CN114153934A (en) * | 2021-12-13 | 2022-03-08 | 高德软件有限公司 | Machine learning model training, navigation route recommendation method and computer storage medium |
| CN114383617A (en) * | 2021-12-29 | 2022-04-22 | 深圳市跨越新科技有限公司 | Navigation route calculation method, device, equipment and storage medium based on map neural network |
| WO2024179127A1 (en) * | 2023-02-28 | 2024-09-06 | 北京高德云图科技有限公司 | Navigation route recommendation method and apparatus, familiar-route prediction model training method and apparatus, and medium |
| CN118960774A (en) * | 2024-10-21 | 2024-11-15 | 南京英莫特信息科技有限公司 | Trip planning method based on vehicle historical navigation data |
| CN118960774B (en) * | 2024-10-21 | 2024-12-20 | 南京英莫特信息科技有限公司 | Journey road condition planning method based on vehicle history navigation data |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109870164A (en) | Navigation terminal and its route preferences prediction technique | |
| CN112074845B (en) | Deep reinforcement learning for optimizing carpooling policies | |
| US10671086B2 (en) | Generation of trip estimates using real-time data and historical data | |
| US10175054B2 (en) | Predicting and utilizing variability of travel times in mapping services | |
| US20080300778A1 (en) | System and method for personalizing trip assistance on a digital map | |
| US9417069B2 (en) | Familiarity modeling | |
| WO2012066951A1 (en) | Data processing device, data processing method, and program | |
| US10816351B1 (en) | Generation of trip estimates using real-time data and historical data | |
| CN102759362B (en) | Navigation system and path planning process | |
| CN110517492A (en) | Traffic route recommendation method, system and device based on parallel ensemble learning | |
| CN107228677A (en) | Driftage recognition methods and device | |
| AU2006297550A1 (en) | Methods for predicting destinations from partial trajectories employing open-and closed-world modeling methods | |
| CN102538813A (en) | Method and apparatus for route searching | |
| US12163793B2 (en) | Methods and apparatus for ascertaining a driving route for a motor vehicle | |
| JP2012251954A (en) | Destination estimation device and navigation system including the same, destination estimation method, destination estimation program | |
| KR102775768B1 (en) | Navigation system | |
| CN107705607A (en) | A kind of information processing method and information processor | |
| CN109631922A (en) | Information processing unit, information processing method and the non-transitory storage medium for storing program | |
| CN109726852A (en) | Based on route planning method, device, terminal and the medium for improving ant group algorithm | |
| JP2016038741A (en) | Movement state determination device, movement state determination method and program | |
| CN109073402A (en) | For determining that safety returns to the method and system of mileage | |
| US20200292344A1 (en) | Information processing apparatus, information processing method, and non-transitory computer readable storage medium storing program | |
| US10557713B1 (en) | Generation of trip estimates using real-time data and historical data | |
| CN112556717A (en) | Travel mode screening method and travel route recommending method and device | |
| CN115183789A (en) | Navigation route determination method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190611 |