CN114254737A - Multi-factor neural network model management method and system - Google Patents
Multi-factor neural network model management method and system Download PDFInfo
- Publication number
- CN114254737A CN114254737A CN202111514218.5A CN202111514218A CN114254737A CN 114254737 A CN114254737 A CN 114254737A CN 202111514218 A CN202111514218 A CN 202111514218A CN 114254737 A CN114254737 A CN 114254737A
- Authority
- CN
- China
- Prior art keywords
- road section
- neural network
- running
- condition
- vehicle
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000003062 neural network model Methods 0.000 title claims abstract description 182
- 238000007726 management method Methods 0.000 title claims abstract description 32
- 238000005286 illumination Methods 0.000 claims abstract description 103
- 238000000034 method Methods 0.000 claims abstract description 9
- 239000000126 substance Substances 0.000 claims description 4
- 238000012549 training Methods 0.000 abstract description 7
- 238000001514 detection method Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 5
- 238000012545 processing Methods 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 239000004576 sand Substances 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 230000008447 perception Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 238000007664 blowing Methods 0.000 description 1
- 239000000428 dust Substances 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000007667 floating Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000002787 reinforcement Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/903—Querying
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Artificial Intelligence (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Databases & Information Systems (AREA)
- Navigation (AREA)
- Traffic Control Systems (AREA)
Abstract
The invention provides a multi-factor neural network model management method and a multi-factor neural network model management system, wherein the method comprises the following steps: judging whether the vehicle is in a navigation state or not; inquiring a corresponding neural network model according to the running condition, the meteorological condition and the illumination condition of the vehicle; sending a retrieval instruction to a terminal of the vehicle, and respectively retrieving the locally stored models by the terminal according to the inquired model numbers; the retrieved model is loaded to the memory after the retrieval is successful; sending a downloading instruction to the server when the failure occurs, and downloading the matched neural network model; loading the corresponding neural network models into the memory according to the sequence of each road section when the downloading is successful; and when the downloading fails, selecting the substitute neural network model with the highest conformity with the inquired neural network model, and loading the substitute neural network model into the memory. The invention improves the practicability and the applicability of the model under different conditions by dynamically managing the neural network model and considering multiple factors, reduces the complexity of the model and shortens the training period.
Description
Technical Field
The invention relates to the field of model management, in particular to a multi-factor neural network model management method and system.
Background
In recent years, AI technology is widely used in the field of automatic driving/driver assistance. Such as deep learning, reinforcement learning and the like, are good research results in the fields of automatic driving/auxiliary driving. And the perception processing is a typical application scenario in the AI technology. Because a vehicle can meet a complex environment in the driving process, the existing sensing technology cannot meet the requirement on the detection or identification precision, and the image processing based on deep learning gradually receives more and more attention. For example, the pedestrian detection technology based on the HOG features performs pedestrian detection through a support vector machine algorithm after the HOG features of the image are extracted; in the vehicle detection technology based on the laser radar and the camera, clustering processing needs to be carried out on laser radar data; linear regression algorithms, support vector machine algorithms, and artificial neural network algorithms are also commonly used for lane marking and traffic sign detection.
In practical application, deep learning has the characteristics that the more complex the model is, the larger the training set is, and the better the effect is, and a smaller neural network model has poor adaptability to application environment changes, so that actual use requirements cannot be met; the complex neural network model requires large computing resources configured on the terminal, affects real-time performance and has high cost, and meanwhile, the training period is long and the required training cost is high.
Disclosure of Invention
The invention provides a multi-factor neural network model management method and a multi-factor neural network model management system, which aim to solve the problem of adaptability of neural network model management under different environmental factors.
In order to solve the above technical problem, an embodiment of the present invention provides a multi-factor neural network model management method, including:
judging whether a running vehicle is in a navigation state or not, and inquiring a corresponding neural network model according to a judgment result by combining the vehicle running condition, the meteorological condition and the illumination condition of the running vehicle;
sending a retrieval instruction to a terminal of the running vehicle so that the terminal retrieves the neural network models stored locally by the terminal on the running vehicle according to the respective numbers corresponding to all the queried neural network models, and loading the retrieved neural network models to a memory of the terminal when the retrieval is successful; when the retrieval fails, sending a downloading request of all first neural network models which accord with the meteorological conditions and the illumination conditions of all road sections to the server;
when the downloading request is received, establishing connection with the terminal so that the terminal can download all the first neural network models, and then loading the first neural network models corresponding to all the road sections into a memory according to the sequence of all the road sections when the terminal is successfully downloaded; and when the downloading fails, selecting the substitute neural network model with the highest conformity with the inquired neural network model, and loading the substitute neural network model to the memory.
Further, the querying a corresponding neural network model according to the judgment result and in combination with the vehicle running condition, the meteorological condition and the illumination condition specifically comprises:
if the judgment result is that the running vehicle is in the navigation state, acquiring a running route of the running vehicle, and dividing the running route into a plurality of road sections; predicting the running time of each road section, the meteorological condition of each road section in the running time and the illumination condition of each road section in the running time; inquiring all first neural network models which meet the conditions according to the predicted meteorological conditions of all road sections and the predicted illumination conditions of all road sections; the first neural network model corresponds to each road section one by one;
if the judgment result is that the running vehicle is not in the navigation state, predicting the running time of the running vehicle on the current road section, the running time of the next road section in the current running direction, the meteorological condition of the current road section in the running time, the illumination condition of the current road section in the running time, the meteorological condition of the next road section in the running time and the illumination condition of the next road section in the running time; inquiring all second neural network models meeting the conditions according to the predicted meteorological conditions and illumination conditions; wherein the second neural network model corresponds to a current road segment or the next road segment.
Further, before sending a retrieval instruction to the terminal of the traveling vehicle, the method further includes:
and matching the position of the vehicle running condition with the current position of the running vehicle, and if the position matching result shows that the running vehicle has yaw, re-executing the following steps until the position matching result shows that the running vehicle has no yaw:
and judging whether the running vehicle is in a navigation state or not, and inquiring a corresponding neural network model according to a judgment result by combining the vehicle running condition, the meteorological condition and the illumination condition of the running vehicle.
Further, before sending a retrieval instruction to the terminal of the traveling vehicle, the method further includes:
if the running vehicle is in a navigation state, sending the road section information of each divided road section, the predicted running time of each road section, the meteorological condition of each road section in the predicted running time, the illumination condition of each road section in the predicted running time and the number corresponding to each inquired first neural network model to the terminal of the running vehicle;
and if the running vehicle is not in the navigation state, sending the road section information of the current running road section, the road section information of the next road section, the predicted running time of the current road section, the predicted running time of the next road section, the meteorological condition of the current road section in the predicted running time, the illumination condition of the current road section in the predicted running time, the meteorological condition of the next road section in the predicted running time, the illumination condition of the next road section in the predicted running time and the inquired numbers corresponding to all the second neural network models to a terminal.
Further, the predicting of the travel time of each road section, the meteorological condition of each road section during the travel time, and the illumination condition of each road section during the travel time specifically includes:
and predicting the running time of each road section according to the road grade of each road section, and predicting the meteorological condition of each road section in the running time and the illumination condition of each road section in the running time according to the weather service.
Further, the predicting the driving time of the current road section of the driving vehicle, the driving time of the next road section in the current driving direction, the weather condition of the current road section during the driving time, the illumination condition of the current road section during the driving time, the weather condition of the next road section during the driving time, and the illumination condition of the next road section during the driving time specifically includes:
predicting the running time of the current road section of the running vehicle according to the road grade of the current road section, predicting the running time of the next road section according to the road grade of the next road section, and predicting the meteorological condition of the current road section in the running time, the illumination condition of the current road section in the running time, the meteorological condition of the next road section in the running time and the illumination condition of the next road section in the running time according to weather service.
Correspondingly, the invention also provides a multi-factor neural network model management system, which comprises an inquiry module, a retrieval module, a terminal on a running vehicle and a server; wherein the content of the first and second substances,
the query module is used for judging whether the running vehicle is in a navigation state or not, and querying a corresponding neural network model according to a judgment result by combining the vehicle running condition, the meteorological condition and the illumination condition of the running vehicle;
the retrieval module is used for sending a retrieval instruction to a terminal of the running vehicle so that the terminal retrieves the neural network models stored locally by the terminal on the running vehicle according to the respective numbers corresponding to all the queried neural network models, and loads the retrieved neural network models to a memory of the terminal when the retrieval is successful; when the retrieval fails, sending downloading requests of all first neural network models which accord with the meteorological conditions and the illumination conditions of all road sections to the server;
when the server receives the downloading request, the server establishes connection with the terminal so that the terminal can download all the first neural network models, and then when the terminal succeeds in downloading, the first neural network models corresponding to all the road sections are loaded into the memory according to the sequence of all the road sections; and when the downloading fails, selecting the substitute neural network model with the highest conformity with the inquired neural network model, and loading the substitute neural network model to the memory.
Further, the query module queries the corresponding neural network model according to the judgment result by combining the vehicle running condition, the meteorological condition and the illumination condition, and specifically comprises:
if the judgment result is that the running vehicle is in the navigation state, the query module acquires a running route of the running vehicle and divides the running route into a plurality of road sections; predicting the running time of each road section, the meteorological condition of each road section in the running time and the illumination condition of each road section in the running time; inquiring all first neural network models which meet the conditions according to the predicted meteorological conditions of all road sections and the predicted illumination conditions of all road sections; the first neural network model corresponds to each road section one by one;
if the judgment result is that the running vehicle is not in the navigation state, the query module predicts the running time of the running vehicle on the current road section, the running time of the next road section in the current running direction, the meteorological condition of the current road section in the running time, the illumination condition of the current road section in the running time, the meteorological condition of the next road section in the running time and the illumination condition of the next road section in the running time; inquiring all second neural network models meeting the conditions according to the predicted meteorological conditions and illumination conditions; wherein the second neural network model corresponds to a current road segment or the next road segment.
Further, the neural network model management system further comprises a position matching module, the position matching module is used for sending a retrieval instruction to the terminal of the running vehicle,
and matching the position of the vehicle running condition with the current position of the running vehicle, and if the position matching result shows that the running vehicle has yaw, re-executing the following steps until the position matching result shows that the running vehicle has no yaw:
and judging whether the running vehicle is in a navigation state or not, and inquiring a corresponding neural network model according to a judgment result by combining the vehicle running condition, the meteorological condition and the illumination condition of the running vehicle.
Further, the neural network model management system further comprises a sending module, wherein the sending module is used for sending a retrieval instruction to the terminal of the running vehicle,
if the running vehicle is in a navigation state, the sending module sends the road section information of each divided road section, the predicted running time of each road section, the meteorological condition of each road section in the predicted running time, the illumination condition of each road section in the predicted running time and the number corresponding to each inquired first neural network model to the terminal of the running vehicle;
if the running vehicle is not in the navigation state, the sending module sends the road section information of the current running road section, the road section information of the next road section, the predicted running time of the current road section, the predicted running time of the next road section, the meteorological condition of the current road section in the predicted running time, the illumination condition of the current road section in the predicted running time, the meteorological condition of the next road section in the predicted running time, the illumination condition of the next road section in the predicted running time and the inquired numbers corresponding to all the second neural network models to the terminal.
Compared with the prior art, the embodiment of the invention has the following beneficial effects:
the embodiment of the invention provides a multi-factor neural network model management method and a system, wherein the neural network model management method comprises the following steps: judging whether a running vehicle is in a navigation state or not, and inquiring a corresponding neural network model according to a judgment result by combining the vehicle running condition, the meteorological condition and the illumination condition of the running vehicle; sending a retrieval instruction to a terminal of the running vehicle so that the terminal retrieves the neural network models stored locally by the terminal on the running vehicle according to the respective numbers corresponding to all the queried neural network models, and loading the retrieved neural network models to a memory of the terminal when the retrieval is successful; when the retrieval fails, sending a downloading request of all first neural network models which accord with the meteorological conditions and the illumination conditions of all road sections to the server; when the downloading request is received, establishing connection with the terminal so that the terminal can download all the first neural network models, and then loading the first neural network models corresponding to all the road sections into a memory according to the sequence of all the road sections when the terminal is successfully downloaded; and when the downloading fails, selecting the substitute neural network model with the highest conformity with the inquired neural network model, and loading the substitute neural network model to the memory. The invention updates and uses the neural network model by considering multiple factors, improves the practicability and the applicability of the neural network model under different meteorological and illumination conditions, ensures that the adopted neural network model has higher matching degree under various conditions and has better application effect; meanwhile, through dynamic management of the neural network model, the complexity of each model in the aspect of design is reduced, the training period is shortened, and meanwhile, the effectiveness of the model under different conditions or scenes is kept.
Drawings
FIG. 1: the invention provides a flow diagram of an embodiment of a multi-factor-based neural network model management method.
FIG. 2: the invention provides a structural schematic diagram of an embodiment of a multi-factor-based neural network model management system.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The first embodiment is as follows:
referring to fig. 1, fig. 1 is a diagram illustrating a multi-factor neural network model management method according to an embodiment of the present invention, including steps S1 to S3; wherein the content of the first and second substances,
and step S1, after the driving behavior starts, the terminal of the running vehicle acquires the position information of the vehicle and judges whether the running vehicle is in the navigation state, and according to the judgment result, the terminal queries the corresponding neural network model by combining the vehicle running condition, the meteorological condition and the illumination condition of the running vehicle. The vehicle-mounted terminal also has the functions of performing intelligent perception processing such as road recognition, vehicle recognition, pedestrian recognition and the like on data such as video information and the like acquired by the vehicle-mounted equipment. In the running process of the vehicle, the terminal uploads data such as position information and video information to the server in real time. And the server is mainly used for providing cloud services such as training of a neural network model, road planning and matching based on a GIS (geographic information system). The server sends a neural network model, structured data and the like to the terminal, wherein the structured data comprises road sections, predicted passing time, weather, illumination, model numbers and the like.
In this embodiment, the querying a corresponding neural network model according to the determination result and in combination with the vehicle driving condition, the weather condition, and the illumination condition specifically includes:
if the judgment result is that the running vehicle is in a navigation state, namely route planning is not used, obtaining a running route of the running vehicle, and dividing the running route into a plurality of road sections; predicting the running time of each road section, the meteorological condition of each road section in the running time and the illumination condition of each road section in the running time; inquiring all first neural network models which meet the conditions according to the predicted meteorological conditions of all road sections and the predicted illumination conditions of all road sections; the first neural network model corresponds to each road section one by one;
specifically, the predicting of the travel time of each road segment, the meteorological condition of each road segment during the travel time, and the illumination condition of each road segment during the travel time includes:
and predicting the running time of each road section according to the road grade of each road section, and predicting the meteorological condition of each road section in the running time and the illumination condition of each road section in the running time according to the weather service.
If the judgment result is that the running vehicle is not in a navigation state, namely a roaming state or route planning is not carried out, predicting the running time of the current road section of the running vehicle, the running time of the next road section in the current running direction, the meteorological condition of the current road section in the running time, the illumination condition of the current road section in the running time, the meteorological condition of the next road section in the running time and the illumination condition of the next road section in the running time; inquiring all second neural network models meeting the conditions according to the predicted meteorological conditions and illumination conditions; wherein the second neural network model corresponds to a current road segment or the next road segment.
Specifically, the predicting of the driving time of the current road section of the driving vehicle, the driving time of the next road section in the current driving direction, the weather condition of the current road section during the driving time, the lighting condition of the current road section during the driving time, the weather condition of the next road section during the driving time, and the lighting condition of the next road section during the driving time is specifically:
predicting the running time of the current road section of the running vehicle according to the road grade of the current road section, predicting the running time of the next road section according to the road grade of the next road section, and predicting the meteorological condition of the current road section in the running time, the illumination condition of the current road section in the running time, the meteorological condition of the next road section in the running time and the illumination condition of the next road section in the running time according to weather service.
In a preferred embodiment, after the step S1 is executed and before the step S2 is executed to transmit a retrieval instruction to the terminal of the traveling vehicle, the method further includes:
and matching the position of the vehicle running condition with the current position of the running vehicle, and if the position matching result shows that the running vehicle has yaw, re-executing the following steps until the position matching result shows that the running vehicle has no yaw:
and judging whether the running vehicle is in a navigation state or not, and inquiring a corresponding neural network model according to a judgment result by combining the vehicle running condition, the meteorological condition and the illumination condition of the running vehicle.
Under the condition that the running vehicle is confirmed not to yaw, if the running vehicle is in a navigation state, the road section information of each divided road section, the predicted running time of each road section, the meteorological condition of each road section in the predicted running time, the illumination condition of each road section in the predicted running time and the number corresponding to each inquired first neural network model are sent to the terminal of the running vehicle;
and if the running vehicle is not in the navigation state, sending the road section information of the current running road section, the road section information of the next road section, the predicted running time of the current road section, the predicted running time of the next road section, the meteorological condition of the current road section in the predicted running time, the illumination condition of the current road section in the predicted running time, the meteorological condition of the next road section in the predicted running time, the illumination condition of the next road section in the predicted running time and the inquired numbers corresponding to all the second neural network models to a terminal.
Step S2, sending a retrieval instruction to the terminal of the running vehicle, so that the terminal retrieves the neural network models stored locally by the terminal of the running vehicle according to the respective numbers corresponding to all the queried neural network models, and loading the retrieved neural network models to the memory of the terminal when the retrieval is successful; and when the retrieval fails, sending a downloading request of all the first neural network models which accord with the meteorological conditions and the illumination conditions of all the road sections to the server.
Step S3, when receiving the download request, establishing a connection with the terminal to allow the terminal to download all the first neural network models, and then loading the first neural network models corresponding to each road segment into a memory according to the sequence of each road segment when the terminal succeeds in downloading; and when the downloading fails, selecting the substitute neural network model with the highest conformity with the inquired neural network model, and loading the substitute neural network model to the memory.
In the present embodiment, the conformity is mainly considered from two dimensions of weather, illumination similarity and space-time continuity. Specifically, in one aspect, embodiments of the invention consider 33 major weather types, including: sunny, cloudy, rainy, thunderstorm with hail, sleet, light rain, medium rain, heavy rainstorm, extra heavy rainstorm, snowfall, small snow, medium snow, heavy snow, fog, frozen rain, sand storm, light rain-medium rain, medium rain-heavy rain, heavy rain-heavy rainstorm, heavy rain-extra heavy rainstorm, small snow-medium snow, medium snow-heavy snow, heavy snow-heavy snow, floating dust, blowing sand, strong sand storm and haze; if the current road section is a cloudy road section, the corresponding neural network model is the neural network model under the cloudy condition, if the neural network model under the cloudy condition is not found, the neural network model under the sunny condition can be used preferentially, and the neural network model under the cloudy condition can be used preferentially, namely the similarity of the sunny environment and the cloudy environment is greater than the similarity of the cloudy environment and the cloudy environment.
On the other hand, the embodiment of the invention also considers the space-time continuity dimension, namely the weather and the illumination condition of the current road section are measured aiming at the preorder section and the subsequent section of the current road section. The closer the interval from the current road section form time is, the higher the similarity of weather and illumination is; and the closer the spatial distance of the path in the form of the current road section is, the higher the similarity of weather and illumination. The embodiment of the invention also sets a time continuity threshold and a space continuity threshold aiming at the space-time continuity dimension, for example, the time continuity is set to be 6 hours before and after, and the space continuity is set to be within 60 kilometers before and after.
Correspondingly, referring to fig. 2, the invention also provides a multi-factor neural network model management system, which comprises an inquiry module 101, a retrieval module 102, a terminal 103 on a running vehicle and a server 104; wherein the content of the first and second substances,
the query module 101 is configured to determine whether the running vehicle is in a navigation state, and query a corresponding neural network model according to a determination result in combination with a vehicle running condition, a weather condition, and an illumination condition of the running vehicle;
the retrieval module 102 is configured to send a retrieval instruction to a terminal of the traveling vehicle, so that the terminal retrieves, according to respective numbers corresponding to all queried neural network models, the neural network models locally stored by the terminal 103 on the traveling vehicle, and loads the retrieved neural network models to a memory of the terminal 103 when the retrieval is successful; and when the retrieval fails, sending a download request of all first neural network models according with the meteorological conditions and the illumination conditions of all the road sections to the server 104;
when the server 104 receives the download request, establishing connection with the terminal 103 to allow the terminal 103 to download all the first neural network models, and then loading the first neural network models corresponding to the road segments into a memory according to the sequence of the road segments when the terminal 103 succeeds in downloading; and when the downloading fails, selecting the substitute neural network model with the highest conformity with the inquired neural network model, and loading the substitute neural network model to the memory.
In this embodiment, the query module 101 queries, according to the determination result, the corresponding neural network model in combination with the vehicle driving condition, the weather condition, and the illumination condition, specifically:
if the judgment result is that the driving vehicle is in the navigation state, the query module 101 acquires a driving route of the driving vehicle, and divides the driving route into a plurality of road sections; predicting the running time of each road section, the meteorological condition of each road section in the running time and the illumination condition of each road section in the running time; inquiring all first neural network models which meet the conditions according to the predicted meteorological conditions of all road sections and the predicted illumination conditions of all road sections; the first neural network model corresponds to each road section one by one;
in this embodiment, the query module 101 predicts the travel time of each road segment, the weather condition of each road segment during the travel time, and the illumination condition of each road segment during the travel time, specifically:
the query module 101 predicts the driving time of each road section according to the road grade of each road section, and predicts the weather condition of each road section in the driving time and the illumination condition of each road section in the driving time according to the weather service.
If the determination result is that the driving vehicle is not in the navigation state, the query module 101 predicts the driving time of the current road section of the driving vehicle, the driving time of the next road section in the current driving direction, the meteorological condition of the current road section in the driving time, the illumination condition of the current road section in the driving time, the meteorological condition of the next road section in the driving time and the illumination condition of the next road section in the driving time; inquiring all second neural network models meeting the conditions according to the predicted meteorological conditions and illumination conditions; wherein the second neural network model corresponds to a current road segment or the next road segment.
In this embodiment, the query module 101 predicts the driving time of the current road segment of the driving vehicle, the driving time of the next road segment in the current driving direction, the weather condition of the current road segment during the driving time, the illumination condition of the current road segment during the driving time, the weather condition of the next road segment during the driving time, and the illumination condition of the next road segment during the driving time, specifically:
the query module 101 predicts the driving time of the current road section of the driving vehicle according to the road grade of the current road section, predicts the driving time of the next road section according to the road grade of the next road section, and predicts the weather condition of the current road section within the driving time, the illumination condition of the current road section within the driving time, the weather condition of the next road section within the driving time and the illumination condition of the next road section within the driving time according to the weather service.
In this embodiment, the neural network model management system further includes a position matching module, where the position matching module is configured to perform position matching on the vehicle driving condition and the current position of the driving vehicle before the retrieval module 102 sends the retrieval instruction to the terminal of the driving vehicle, and if the position matching result is that the driving vehicle has drifted, re-perform the following steps until the position matching result is that the driving vehicle has not drifted:
and judging whether the running vehicle is in a navigation state or not, and inquiring a corresponding neural network model according to a judgment result by combining the vehicle running condition, the meteorological condition and the illumination condition of the running vehicle.
In this embodiment, the neural network model management system further includes a sending module, configured to, before the retrieving module 102 sends the retrieving instruction to the terminal of the running vehicle,
if the running vehicle is in a navigation state, the sending module sends the road section information of each divided road section, the predicted running time of each road section, the meteorological condition of each road section in the predicted running time, the illumination condition of each road section in the predicted running time and the number corresponding to each inquired first neural network model to the terminal of the running vehicle;
if the running vehicle is not in the navigation state, the sending module sends the road section information of the current running road section, the road section information of the next road section, the predicted running time of the current road section, the predicted running time of the next road section, the meteorological condition of the current road section in the predicted running time, the illumination condition of the current road section in the predicted running time, the meteorological condition of the next road section in the predicted running time, the illumination condition of the next road section in the predicted running time and the inquired numbers corresponding to all the second neural network models to the terminal.
Compared with the prior art, the embodiment of the invention has the following beneficial effects:
the embodiment of the invention provides a multi-factor neural network model management method and a system, wherein the neural network model management method comprises the following steps: judging whether a running vehicle is in a navigation state or not, and inquiring a corresponding neural network model according to a judgment result by combining the vehicle running condition, the meteorological condition and the illumination condition of the running vehicle; sending a retrieval instruction to a terminal of the running vehicle so that the terminal retrieves the neural network models stored locally by the terminal on the running vehicle according to the respective numbers corresponding to all the queried neural network models, and loading the retrieved neural network models to a memory of the terminal when the retrieval is successful; when the retrieval fails, sending a downloading request of all first neural network models which accord with the meteorological conditions and the illumination conditions of all road sections to the server; when the downloading request is received, establishing connection with the terminal so that the terminal can download all the first neural network models, and then loading the first neural network models corresponding to all the road sections into a memory according to the sequence of all the road sections when the terminal is successfully downloaded; and when the downloading fails, selecting the substitute neural network model with the highest conformity with the inquired neural network model, and loading the substitute neural network model to the memory. The invention updates and uses the neural network model by considering multiple factors, improves the practicability and the applicability of the neural network model under different meteorological and illumination conditions, ensures that the adopted neural network model has higher matching degree under various conditions and has better application effect; meanwhile, through dynamic management of the neural network model, the complexity of each model in the aspect of design is reduced, the training period is shortened, and meanwhile, the effectiveness of the model under different conditions or scenes is kept.
The above-mentioned embodiments are provided to further explain the objects, technical solutions and advantages of the present invention in detail, and it should be understood that the above-mentioned embodiments are only examples of the present invention and are not intended to limit the scope of the present invention. It should be understood that any modifications, equivalents, improvements and the like, which come within the spirit and principle of the invention, may occur to those skilled in the art and are intended to be included within the scope of the invention.
Claims (10)
1. A multi-factor neural network model management method is characterized by comprising the following steps:
judging whether a running vehicle is in a navigation state or not, and inquiring a corresponding neural network model according to a judgment result by combining the vehicle running condition, the meteorological condition and the illumination condition of the running vehicle;
sending a retrieval instruction to a terminal of the running vehicle so that the terminal retrieves the neural network models stored locally by the terminal on the running vehicle according to the respective numbers corresponding to all the queried neural network models, and loading the retrieved neural network models to a memory of the terminal when the retrieval is successful; when the retrieval fails, sending a downloading request of all first neural network models which accord with the meteorological conditions and the illumination conditions of all road sections to the server;
when the downloading request is received, establishing connection with the terminal so that the terminal can download all the first neural network models, and then loading the first neural network models corresponding to all the road sections into a memory according to the sequence of all the road sections when the terminal is successfully downloaded; and when the downloading fails, selecting the substitute neural network model with the highest conformity with the inquired neural network model, and loading the substitute neural network model to the memory.
2. The method according to claim 1, wherein the querying of the neural network model according to the determination result in combination with the vehicle driving condition, the weather condition and the illumination condition specifically comprises:
if the judgment result is that the running vehicle is in the navigation state, acquiring a running route of the running vehicle, and dividing the running route into a plurality of road sections; predicting the running time of each road section, the meteorological condition of each road section in the running time and the illumination condition of each road section in the running time; inquiring all first neural network models which meet the conditions according to the predicted meteorological conditions of all road sections and the predicted illumination conditions of all road sections; the first neural network model corresponds to each road section one by one;
if the judgment result is that the running vehicle is not in the navigation state, predicting the running time of the running vehicle on the current road section, the running time of the next road section in the current running direction, the meteorological condition of the current road section in the running time, the illumination condition of the current road section in the running time, the meteorological condition of the next road section in the running time and the illumination condition of the next road section in the running time; inquiring all second neural network models meeting the conditions according to the predicted meteorological conditions and illumination conditions; wherein the second neural network model corresponds to a current road segment or the next road segment.
3. The multi-factor neural network model management method according to claim 1, further comprising, before sending a retrieval instruction to the terminal of the running vehicle:
and matching the position of the vehicle running condition with the current position of the running vehicle, and if the position matching result shows that the running vehicle has yaw, re-executing the following steps until the position matching result shows that the running vehicle has no yaw:
and judging whether the running vehicle is in a navigation state or not, and inquiring a corresponding neural network model according to a judgment result by combining the vehicle running condition, the meteorological condition and the illumination condition of the running vehicle.
4. The multi-factor neural network model management method according to claim 2, further comprising, before sending a retrieval instruction to the terminal of the running vehicle:
if the running vehicle is in a navigation state, sending the road section information of each divided road section, the predicted running time of each road section, the meteorological condition of each road section in the predicted running time, the illumination condition of each road section in the predicted running time and the number corresponding to each inquired first neural network model to the terminal of the running vehicle;
and if the running vehicle is not in the navigation state, sending the road section information of the current running road section, the road section information of the next road section, the predicted running time of the current road section, the predicted running time of the next road section, the meteorological condition of the current road section in the predicted running time, the illumination condition of the current road section in the predicted running time, the meteorological condition of the next road section in the predicted running time, the illumination condition of the next road section in the predicted running time and the inquired numbers corresponding to all the second neural network models to a terminal.
5. The method for managing a multi-factor neural network model according to claim 2, wherein the predicting of the travel time of each link, the weather condition of each link during the travel time, and the illumination condition of each link during the travel time comprises:
and predicting the running time of each road section according to the road grade of each road section, and predicting the meteorological condition of each road section in the running time and the illumination condition of each road section in the running time according to the weather service.
6. The method according to claim 2, wherein the predicting the driving time of the current road section of the driving vehicle, the driving time of the next road section in the current driving direction, the weather condition of the current road section during the driving time, the lighting condition of the current road section during the driving time, the weather condition of the next road section during the driving time, and the lighting condition of the next road section during the driving time comprises:
predicting the running time of the current road section of the running vehicle according to the road grade of the current road section, predicting the running time of the next road section according to the road grade of the next road section, and predicting the meteorological condition of the current road section in the running time, the illumination condition of the current road section in the running time, the meteorological condition of the next road section in the running time and the illumination condition of the next road section in the running time according to weather service.
7. A multi-factor neural network model management system is characterized by comprising a query module, a retrieval module, a terminal on a running vehicle and a server; wherein the content of the first and second substances,
the query module is used for judging whether the running vehicle is in a navigation state or not, and querying a corresponding neural network model according to a judgment result by combining the vehicle running condition, the meteorological condition and the illumination condition of the running vehicle;
the retrieval module is used for sending a retrieval instruction to a terminal of the running vehicle so that the terminal retrieves the neural network models stored locally by the terminal on the running vehicle according to the respective numbers corresponding to all the queried neural network models, and loads the retrieved neural network models to a memory of the terminal when the retrieval is successful; when the retrieval fails, sending downloading requests of all first neural network models which accord with the meteorological conditions and the illumination conditions of all road sections to the server;
when the server receives the downloading request, the server establishes connection with the terminal so that the terminal can download all the first neural network models, and then when the terminal succeeds in downloading, the first neural network models corresponding to all the road sections are loaded into the memory according to the sequence of all the road sections; and when the downloading fails, selecting the substitute neural network model with the highest conformity with the inquired neural network model, and loading the substitute neural network model to the memory.
8. The multi-factor neural network model management system of claim 7, wherein the query module queries the corresponding neural network model according to the determination result in combination with the vehicle driving condition, the meteorological condition and the lighting condition, specifically:
if the judgment result is that the running vehicle is in the navigation state, the query module acquires a running route of the running vehicle and divides the running route into a plurality of road sections; predicting the running time of each road section, the meteorological condition of each road section in the running time and the illumination condition of each road section in the running time; inquiring all first neural network models which meet the conditions according to the predicted meteorological conditions of all road sections and the predicted illumination conditions of all road sections; the first neural network model corresponds to each road section one by one;
if the judgment result is that the running vehicle is not in the navigation state, the query module predicts the running time of the running vehicle on the current road section, the running time of the next road section in the current running direction, the meteorological condition of the current road section in the running time, the illumination condition of the current road section in the running time, the meteorological condition of the next road section in the running time and the illumination condition of the next road section in the running time; inquiring all second neural network models meeting the conditions according to the predicted meteorological conditions and illumination conditions; wherein the second neural network model corresponds to a current road segment or the next road segment.
9. The multi-factor neural network model management system of claim 7, further comprising a location matching module for, before the retrieval module sends a retrieval instruction to the terminal of the running vehicle,
and matching the position of the vehicle running condition with the current position of the running vehicle, and if the position matching result shows that the running vehicle has yaw, re-executing the following steps until the position matching result shows that the running vehicle has no yaw:
and judging whether the running vehicle is in a navigation state or not, and inquiring a corresponding neural network model according to a judgment result by combining the vehicle running condition, the meteorological condition and the illumination condition of the running vehicle.
10. The multi-factor neural network model management system of claim 8, further comprising a sending module for, before sending a retrieval instruction to a terminal of the running vehicle,
if the running vehicle is in a navigation state, the sending module sends the road section information of each divided road section, the predicted running time of each road section, the meteorological condition of each road section in the predicted running time, the illumination condition of each road section in the predicted running time and the number corresponding to each inquired first neural network model to the terminal of the running vehicle;
if the running vehicle is not in the navigation state, the sending module sends the road section information of the current running road section, the road section information of the next road section, the predicted running time of the current road section, the predicted running time of the next road section, the meteorological condition of the current road section in the predicted running time, the illumination condition of the current road section in the predicted running time, the meteorological condition of the next road section in the predicted running time, the illumination condition of the next road section in the predicted running time and the inquired numbers corresponding to all the second neural network models to the terminal.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111514218.5A CN114254737B (en) | 2021-12-10 | 2021-12-10 | Multi-factor neural network model management method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111514218.5A CN114254737B (en) | 2021-12-10 | 2021-12-10 | Multi-factor neural network model management method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114254737A true CN114254737A (en) | 2022-03-29 |
| CN114254737B CN114254737B (en) | 2024-04-02 |
Family
ID=80791979
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111514218.5A Active CN114254737B (en) | 2021-12-10 | 2021-12-10 | Multi-factor neural network model management method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114254737B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070129860A1 (en) * | 2005-12-06 | 2007-06-07 | Hunter Engineering Company | Vehicle Service Equipment Interface Drivers |
| CN110610260A (en) * | 2019-08-21 | 2019-12-24 | 南京航空航天大学 | Driving energy consumption prediction system, method, storage medium and equipment |
-
2021
- 2021-12-10 CN CN202111514218.5A patent/CN114254737B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070129860A1 (en) * | 2005-12-06 | 2007-06-07 | Hunter Engineering Company | Vehicle Service Equipment Interface Drivers |
| CN110610260A (en) * | 2019-08-21 | 2019-12-24 | 南京航空航天大学 | Driving energy consumption prediction system, method, storage medium and equipment |
Non-Patent Citations (1)
| Title |
|---|
| 何波;卢耀军;陈旭梅;林国鑫;朱琳;: "基于神经网络的公交车辆行程时间预测模型研究", 城市公共交通, no. 04, 15 April 2010 (2010-04-15) * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114254737B (en) | 2024-04-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108372857B (en) | Efficient context awareness by event occurrence and episode memory review for autonomous driving systems | |
| US20210302585A1 (en) | Smart navigation method and system based on topological map | |
| CN108372856A (en) | Effective context-aware of perception stream in autonomous driving system | |
| WO2020042348A1 (en) | Method for generating autonomous driving navigation map, system, vehicle-mounted terminal, and server | |
| CN108073170A (en) | For the automation collaborative Driving control of autonomous vehicle | |
| CN110356412B (en) | Method and apparatus for automatic rule learning for autonomous driving | |
| CN110435667A (en) | System and method for controlling autonomous vehicle | |
| KR20210087005A (en) | Method and apparatus of estimating road condition, and method and apparatus of establishing road condition estimation model | |
| KR20180046798A (en) | Method and apparatus for real time traffic information provision | |
| CN110827540B (en) | Motor vehicle movement mode recognition method and system based on multi-mode data fusion | |
| US20120271544A1 (en) | System and Method for Providing Georeferenced Predictive Information to Motor Vehicles | |
| CN112885134B (en) | Smart city traffic management method based on big data | |
| US20220201584A1 (en) | Method for predicting a modification of the linkage conditions of a terminal to a cellular network | |
| CN110789515B (en) | System and method for hardware validation in a motor vehicle | |
| US10953871B2 (en) | Transportation infrastructure communication and control | |
| US10974727B2 (en) | Transportation infrastructure communication and control | |
| CN113085891A (en) | Vehicle control method, vehicle control device, nonvolatile storage medium and processor | |
| CN114254737B (en) | Multi-factor neural network model management method and system | |
| CN111483465B (en) | MPP expansion method and device, electronic equipment and storage medium | |
| CN112258881B (en) | Vehicle management method based on intelligent traffic | |
| JP7025372B2 (en) | Object identification device, method and system | |
| CN113850111A (en) | Road condition identification method and system based on semantic segmentation and neural network technology | |
| CN112710316A (en) | Dynamic map generation focused on the field of construction and localization technology | |
| CN111860396A (en) | Method for identifying and summarizing congestion conditions of current area of vehicle | |
| CN112784824A (en) | Intelligent driving deep learning model management method, device and system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| CB02 | Change of applicant information | ||
| CB02 | Change of applicant information |
Address after: 510000 Room 601, building a, No. 23, Nanxiang Second Road, Huangpu District, Guangzhou, Guangdong Applicant after: GUANGZHOU HAIGE XINGHANG INFORMATION TECHNOLOGY CO.,LTD. Address before: 510000 room 3009, No.5, Wangjiang 2nd Street, Nansha District, Guangzhou City, Guangdong Province Applicant before: GUANGZHOU HAIGE XINGHANG INFORMATION TECHNOLOGY CO.,LTD. |
|
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |