CN111275182A - Deep learning simulation consolidation method based on cloud computing - Google Patents
Deep learning simulation consolidation method based on cloud computing Download PDFInfo
- Publication number
- CN111275182A CN111275182A CN202010029453.2A CN202010029453A CN111275182A CN 111275182 A CN111275182 A CN 111275182A CN 202010029453 A CN202010029453 A CN 202010029453A CN 111275182 A CN111275182 A CN 111275182A
- Authority
- CN
- China
- Prior art keywords
- data
- deep learning
- simulation
- environment
- learning
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
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
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a deep learning simulation consolidation method based on cloud computing, which belongs to the field of deep learning, can improve the precision and the operating sensitivity of the actual operation of a deep learning object by setting an error model between a data group similar to a simulation environment and ideal data, can continuously reduce the determining range of a next operation data point by setting a difference node and a mutual-tolerance correcting ring outside the difference node in the actual fine tuning correcting process, effectively improve the precision of the next operation data point, simultaneously can reduce the judging difficulty of the next operation data point, improve the overall operating efficiency of the deep learning object, and simultaneously expand the depth of the simulation environment and reduce the difference between the simulation environment and the actual operation environment by the difference between the data group which is obviously different from the simulation environment and the ideal data, and further improve the comprehensiveness of deep learning objects.
Description
Technical Field
The invention relates to the field of deep learning, in particular to a deep learning simulation consolidation method based on cloud computing.
Background
Deep Learning (DL) is a new research direction in the field of Machine Learning (ML), which is introduced into Machine Learning to make it closer to the original goal, Artificial Intelligence (AI).
Deep learning is the intrinsic law and expression level of the learning sample data, and the information obtained in the learning process is very helpful for the interpretation of data such as characters, images and sounds. The final aim of the method is to enable the machine to have the analysis and learning capability like a human, and to recognize data such as characters, images and sounds. Deep learning is a complex machine learning algorithm, and achieves the effect in speech and image recognition far exceeding the prior related art.
Deep learning has achieved many achievements in search technology, data mining, machine learning, machine translation, natural language processing, multimedia learning, speech, recommendation and personalization technologies, and other related fields. The deep learning enables the machine to imitate human activities such as audio-visual and thinking, solves a plurality of complex pattern recognition problems, and makes great progress on the artificial intelligence related technology.
The existing deep learning simulation only obtains related data in a simulation environment, the related data is used as the basis of actual operation of a deep learning object in the later period, however, the data obtained by deep learning is obtained in a simulated ideal environment, data pieces are idealized, and when the deep learning object is separated from the simulation environment to carry out actual operation, certain errors exist between final actual operation data and the ideal data, and the subsequent actual operation efficiency, sensitivity and precision of the deep learning object are influenced due to the fact that some conditions which do not appear in the simulation environment or are similar to but different from those in the simulation environment appear.
Disclosure of Invention
1. Technical problem to be solved
Aiming at the problems in the prior art, the invention aims to provide a deep learning simulation consolidation method based on cloud computing, which can set an error model through an error between a data group similar to a simulation environment and ideal data, can perform fine tuning correction on next operation data, thereby improving the actual operation precision and operation sensitivity of a deep learning object, can continuously reduce the determining range of next operation data points through setting a difference node and a mutual-capacitance correction ring outside the difference node in the actual fine tuning correction process, effectively improves the precision of the next operation data points, can reduce the judgment difficulty of the next operation data points, effectively improves the data fine tuning repair efficiency of the error model, namely improves the overall operation efficiency of the deep learning object, and simultaneously, through the difference between the data group which has obvious difference with the simulation environment and the ideal data, the depth of the simulation environment is expanded, the difference between the simulation environment and the actual operation environment is reduced, and the comprehensiveness of deep learning of a deep learning object is further improved.
2. Technical scheme
In order to solve the above problems, the present invention adopts the following technical solutions.
The deep learning simulation consolidation method based on cloud computing comprises the following steps:
s1, firstly, determining a deep learning object, and then listing influence factors influencing the object;
s2, simulating the environment through the three-dimensional simulator, simulating the scene of deep learning according to the influence factors, and modifying the influence factor data of the simulated environment through the remote control terminal to realize the simulation of various different simulated environments;
s3, after the simulation is completed, performing deep learning on the determined object in the simulated environment, storing deep learning data in the simulated environment in a cloud end, and obtaining optimal learning data through cloud computing;
s4, the cloud end imports the optimal learning data into a control system of a learning object, the learning object performs actual operation according to the optimal learning data, the actual operation environment is segmented according to the fact that the two actual environments are obviously different, a plurality of different data acquisition points are set, the data acquisition points are in signal connection, the data acquisition points acquire the actual data through a data acquisition module to obtain a plurality of bidirectional actual data groups, and the actual data groups are imported into the cloud end again;
s5, the bidirectional actual data set comprises a data set similar to the simulation environment and a data set with obvious difference between the actual environment and the simulation environment;
s51, the cloud end compares a data set similar to the simulation environment with the optimal learning data through cloud calculation, establishes an error model, and continuously performs fine adjustment and correction on the actual operation data of the learning object through the error model, and the actual operation data is used as the next actual operation data, so that the accuracy and the operation sensitivity of the actual operation of the deep learning object are well consolidated;
s52, the cloud end carries out data analysis on the data group which is obviously different from the simulation environment and the optimal learning data through cloud calculation, the difference factor between the simulation environment and the actual environment is reflected through the comparison difference, the difference factor is fed back to the three-dimensional simulator and serves as the influence factor of the three-dimensional simulator for carrying out environment simulation, the depth of the simulation environment is enlarged, the difference between the simulation environment and the actual operation environment is reduced, and the comprehensiveness of deep learning of a deep learning object is improved.
The error model is set through the error between the data group similar to the simulation environment and the ideal data, the precision and the operation sensitivity of the actual operation of the deep learning object can be improved, in the actual fine tuning correction process, the determining range of the next operation data point can be continuously reduced through the difference node and the arrangement of the mutual-capacitance correction ring outside the difference node, the precision of the next operation data point is effectively improved, meanwhile, the judgment difficulty of the next operation data point can be reduced, the overall operation efficiency of the deep learning object is improved, meanwhile, the depth of the simulation environment is expanded through the difference between the data group which is obviously different from the simulation environment and the ideal data, the difference between the simulation environment and the actual operation environment is reduced, and the comprehensiveness of the deep learning object in deep learning is further improved.
Furthermore, the number of the influencing factors in S1 is not less than three, and the influencing factors are too small, which may result in insufficient learning depth, so that there is a large error in the best learning data obtained in the later period, and it is easy to mislead the error correction of the actual operation data in the actual operation, when the learning object performs the actual operation after the deep learning, the data of the actual operation each time is used as the data source of the error model, so that the data source of the error model is continuously enlarged, so that the correction of the operation data is more and more accurate, the longer the time, the higher the operation precision of the learning object is, and the work efficiency and the effect thereof are improved.
Further, the method for acquiring the optimal learning data in S3 specifically includes:
s31, controlling the three-dimensional simulator to change different data of the influence factors through the remote control terminal, collecting deep learning data of the learning object under the condition of various different influence factors, and uploading the deep learning data to the cloud;
s32, repeating the deep learning of the multiple scenes, collecting multiple deep learning data in the same deep learning scene, and uploading the deep learning data to the cloud end;
and S33, comparing multiple deep learning data in the same deep learning scene and multiple groups of data in different deep learning scenes by the cloud to obtain the optimal learning data in different deep learning scenes.
Further, the data acquisition module in S4 includes a data collector and a memory, the data collector may be but is not limited to a temperature sensor, a speed sensor, an anti-collision sensor and a brightness sensor, the data collector is in signal connection with the memory, and the memory is in signal connection with the cloud.
Furthermore, the remote control terminal comprises a shell, a touch screen, a built-in control chip and a wireless signal transmitter, wherein the wireless signal transmitter and the control chip are in signal connection with the three-dimensional simulator, when deep learning of the deep learning object is carried out in a simulation environment, influence factors can be directly modified on the touch screen of the remote control terminal, then the wireless signal transmitter sends parameters corresponding to the modified influence factors to the three-dimensional simulator, and the three-dimensional simulator modifies the simulation parameters according to the received parameters, so that the simulation environment is changed, and the deep learning object can carry out deep learning in simulation environments with different influence factors.
Further, the error model is an error between the multiple operation data and the ideal data, and the error model is marked with a node of each error, so that the difference between the operation data of each time can be observed conveniently through the error node, the error model can be used for correcting the error, the error correction efficiency can be improved, and the ideal data is the best learning data in S3.
Furthermore, mutual capacitance correction rings are arranged outside each error node, the range of data fine adjustment correction of each error model is conveniently reduced through the mutual capacitance correction rings, the speed of determining the next operation data is accelerated, and the operation sensitivity and the working efficiency of the learning object are effectively improved.
Furthermore, the radius of the mutual capacitance correction ring is set according to the distance between the error node of the first three marks and the ideal data point, the distance between the three error nodes and the ideal data point is not obviously different, and one of the three error nodes has obvious difference with the other two error nodes, which indicates that the data has error, so that in the actual data correction, the data is taken to cause large error, and in the actual operation, the data needs to be skipped, and the fourth error node is selected to be used together with the other two error nodes to determine the radius of the mutual capacitance correction ring, thereby effectively improving the precision.
Furthermore, the radius of the mutual capacitance correction ring is smaller than the average distance between the three error nodes and the ideal data point and is larger than half of the average distance, an excessively large radius of the mutual capacitance correction ring easily causes a plurality of mutually crossed parts to be enlarged, the determination range of the next operation data point to be enlarged, the accuracy of the determined next operation data point to be reduced, and an excessively small radius easily causes the mutually crossed parts to be too small or even not to be arranged, so that the difficulty in judging the next operation data point is increased under the condition that the data source of the previous error model is less, and the weakening of the fine adjustment and repair capability of the error model to the data is reduced.
Further, the actual operation of the data trimming correction in S5 is: and taking the central point of the most intersected part of the mutual-capacitance correction circles as actual data of the next operation of the learning object to finish fine adjustment and correction of the data.
3. Advantageous effects
Compared with the prior art, the invention has the advantages that:
(1) according to the scheme, the error model can be set through errors between actual operation data and ideal data, fine adjustment and correction can be carried out on next operation data, the precision and the operation sensitivity of actual operation of a deep learning object are improved, in the actual fine adjustment and correction process, mutual capacitance correction rings outside a difference node and the difference node are arranged, the central point of the most intersected part of a plurality of mutual capacitance correction rings is used as actual data of next operation of the learning object, the determining range of next operation data points can be continuously reduced in the process, the precision of the next operation data points is effectively improved, meanwhile, the judgment difficulty of the next operation data points can be reduced, the fine adjustment and repair efficiency of the error model on the data is effectively improved, and the overall operation efficiency of the deep learning object is improved.
(2) The number of the influencing factors in the S1 is not less than three, and too few influencing factors may result in insufficient learning depth, so that the optimal learning data obtained in the later period has a large error, which is likely to mislead the error correction of the actual operation data in the next time in the actual operation, when the learning object performs the actual operation after deep learning, the data of each actual operation is used as the data source of the error model, so that the data source of the error model is continuously enlarged, the correction of the operation data is more and more accurate, the time is longer, the operation precision of the learning object is higher, and the work efficiency and the effect are improved.
(3) The data acquisition module in the S4 comprises a data acquisition unit and a memory, wherein the data acquisition unit can be but is not limited to a temperature sensor, a speed sensor, an anti-collision sensor and a brightness sensor, the data acquisition unit is in signal connection with the memory, and the memory is in signal connection with the cloud
(4) The remote control terminal comprises a shell, a touch screen, a built-in control chip, a wireless signal transmitter chip and a three-dimensional simulator, wherein the wireless signal transmitter and the control chip are in signal connection with the three-dimensional simulator, when a deep learning object performs deep learning in a simulation environment, influence factors can be directly modified on the touch screen of the remote control terminal, then the wireless signal transmitter transmits parameters corresponding to the modified influence factors to the three-dimensional simulator, the three-dimensional simulator performs simulation parameter modification according to the received parameters, and therefore the simulation environment is changed, and the deep learning object can perform deep learning in the simulation environments with different influence factors.
(5) The error model is an error between the multi-operation data and the ideal data, a node of each error is marked on the error model, the difference between the operation data of each time can be observed conveniently through the error node, the error model can be used for correcting reference, the error correction efficiency is improved, and the ideal data is the best learning data in S3.
(6) Each error node is externally provided with a mutual capacitance correction ring, the range of data fine adjustment correction of each error model is conveniently reduced through the mutual capacitance correction rings, the speed of determining next operation data is accelerated, and the operation sensitivity and the working efficiency of a learning object are effectively improved.
(7) The radius of the mutual-capacitance correction ring is set according to the distance between the error node of the first three marks and the ideal data point, the distance between the three error nodes and the ideal data point is not obviously different, one of the three error nodes is obviously different from the other two error nodes, and the difference indicates that the data has errors, so that the data is taken to cause large errors in actual data correction, and therefore, in actual operation, the data needs to be skipped over, and the fourth error node is selected to be used together with the other two error nodes to determine the radius of the mutual-capacitance correction ring, so that the precision is effectively improved.
(8) The radius of the mutual-capacitance correction ring is smaller than the average distance between the three error nodes and the ideal data point and larger than half of the average distance, the larger radius of the mutual-capacitance correction ring easily causes the larger cross sections among the error nodes, the larger range of the next operation data point is determined, the lower precision of the next operation data point is determined, the smaller radius of the mutual-capacitance correction ring easily causes the smaller or even no cross sections among the error nodes, the larger difficulty of judging the next operation data point is caused under the condition that the data source of the previous error model is less, and the weaker fine-tuning repair capability of the error model to the data is reduced.
(9) The actual operation of the data trimming correction in S5 is: and taking the central point of the most intersected part of the plurality of mutual-capacitance correction circles as actual data of the next operation of the learning object to finish fine adjustment and correction of the data.
Drawings
FIG. 1 is a principal flow diagram of the present invention;
FIG. 2 is a block diagram of the present invention for deep learning under a variety of influencing factors;
FIG. 3 is a schematic diagram of the structure of the next operation data point determined according to the present invention;
fig. 4 is a deep learning block diagram for controlling the driving speed of the unmanned vehicle in embodiment 2 of the present invention.
Detailed Description
The drawings in the embodiments of the invention will be combined; the technical scheme in the embodiment of the invention is clearly and completely described; obviously; the described embodiments are only some of the embodiments of the invention; but not all embodiments, are based on the embodiments of the invention; all other embodiments obtained by a person skilled in the art without making any inventive step; all fall within the scope of protection of the present invention.
In the description of the present invention, it should be noted that the terms "upper", "lower", "inner", "outer", "top/bottom", and the like indicate orientations or positional relationships based on those shown in the drawings, and are only for convenience of description and simplification of description, but do not indicate or imply that the referred device or element must have a specific orientation, be constructed in a specific orientation, and be operated, and thus should not be construed as limiting the present invention. Furthermore, the terms "first" and "second" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance.
In the description of the present invention, it should be noted that, unless otherwise explicitly specified or limited, the terms "mounted," "disposed," "sleeved/connected," "connected," and the like are to be construed broadly, e.g., "connected," which may be fixedly connected, detachably connected, or integrally connected; can be mechanically or electrically connected; they may be connected directly or indirectly through intervening media, or they may be interconnected between two elements. The specific meanings of the above terms in the present invention can be understood in specific cases to those skilled in the art.
Example 1:
referring to fig. 1-2, a deep learning simulation consolidation method based on cloud computing,
the deep learning simulation consolidation method based on cloud computing comprises the following steps:
s1, firstly, determining a deep learning object, and then listing influence factors influencing the object;
s2, simulating the environment through the three-dimensional simulator, simulating the scene of deep learning according to the influence factors, and modifying the influence factor data of the simulated environment through the remote control terminal to realize the simulation of various different simulated environments;
s3, after the simulation is completed, deep learning is carried out on the determined object in the simulated environment, data are collected and stored in the cloud, the best learning data are obtained through cloud computing, and the best learning data are obtained in the following specific mode:
s31, controlling the three-dimensional simulator to change different data of the influence factors through the remote control terminal, collecting deep learning data of the learning object under the condition of various different influence factors, and uploading the deep learning data to the cloud;
s32, repeating the deep learning of the multiple scenes, collecting multiple deep learning data in the same deep learning scene, and uploading the deep learning data to the cloud end;
s33, comparing multiple deep learning data in the same deep learning scene and multiple groups of data in different deep learning scenes by the cloud to obtain the optimal learning data in different deep learning scenes;
s4, the cloud end imports the optimal learning data into a control system of a learning object, the learning object performs actual operation according to the optimal learning data, the actual operation environment is segmented according to the fact that the two actual environments are obviously different, a plurality of different data acquisition points are set, the data acquisition points are in signal connection, the data acquisition points acquire the actual data through a data acquisition module to obtain a plurality of bidirectional actual data groups, and the actual data groups are imported into the cloud end again;
s5, the bidirectional actual data set comprises a data set similar to the simulation environment and a data set with obvious difference between the actual environment and the simulation environment;
s51, the cloud end compares a data set similar to the simulation environment with the optimal learning data through cloud calculation, establishes an error model, and continuously performs fine adjustment and correction on the actual operation data of the learning object through the error model, and the actual operation data is used as the next actual operation data, so that the accuracy and the operation sensitivity of the actual operation of the deep learning object are well consolidated;
s52, the cloud end carries out data analysis on the data group which is obviously different from the simulation environment and the optimal learning data through cloud calculation, the difference factor between the simulation environment and the actual environment is reflected through the comparison difference, the difference factor is fed back to the three-dimensional simulator and serves as the influence factor of the three-dimensional simulator for carrying out environment simulation, the depth of the simulation environment is enlarged, the difference between the simulation environment and the actual operation environment is reduced, and the comprehensiveness of deep learning of a deep learning object is improved.
The number of the influencing factors in the S1 is not less than three, the influencing factors are too small, the learning depth is insufficient, the best learning data obtained in the later period has large errors, the error correction of the next actual operation data in the subsequent actual operation is easy to mislead, when the learning object carries out the actual operation after the deep learning, the data of each actual operation is used as the data source of the error model, the data source of the error model is continuously enlarged, the correction of the operation data is more and more accurate, the time is longer, the operation precision of the learning object is higher, the working efficiency and the effect are improved, the error model is the error between the multi-time operation data and the ideal data, the node of each error is marked on the error model, the difference between each time of operation data is convenient to observe through the error node, the reference of the error model correction can be realized, and the error correction efficiency is improved, the ideal data is the best learning data in S3.
The data acquisition module in S4 comprises a data acquisition unit and a memory, the data acquisition unit can be but is not limited to a temperature sensor, a speed sensor, an anti-collision sensor and a brightness sensor, the data acquisition unit is in signal connection with the memory, the memory is in signal connection with the cloud, the remote control terminal comprises a shell, a touch screen, a built-in control chip and a wireless signal emitter, the wireless signal emitter and the control chip are in signal connection with the three-dimensional simulator, when the deep learning object performs deep learning in a simulation environment, the influence factors can be directly modified on the touch screen of the remote control terminal, then the wireless signal transmitter transmits the parameters corresponding to the modified influencing factors to the three-dimensional simulator, the three-dimensional simulator modifies the simulation parameters according to the received parameters, therefore, the change of the simulation environment is realized, and the deep learning object can carry out deep learning in the simulation environments with different influence factors.
Referring to fig. 3, each error node is externally provided with a mutual capacitance correction ring, the mutual capacitance correction ring is convenient to reduce the range of data fine tuning correction of each error model, the speed of determining the next operation data is increased, the operation sensitivity and the working efficiency of the learning object are effectively improved, the radius of the mutual capacitance correction ring is set according to the distance between the error node of the first three marks and the ideal data point, the distance between the three error nodes and the ideal data point is not obviously different, and one of the three error nodes is obviously different from the other two error nodes, which indicates that the data is wrong, therefore, in the actual data correction, the data is taken to cause a large error, so that in the actual operation, the data needs to be skipped over, and the fourth error node is selected to carry out the determination of the mutual capacitance correction circle radius together with the other two error nodes, thereby effectively improving the precision;
the radius of the mutual-capacitance correction ring is smaller than the average distance between the three error nodes and the ideal data point and is larger than half of the average distance, the too large radius of the mutual-capacitance correction ring easily causes the enlargement of a plurality of mutual cross parts, the enlargement of the determination range of the next operation data point, the reduction of the precision of the determined next operation data point, the too small radius easily causes the undersize or even no mutual cross parts, and the judgment difficulty of the next operation data point is increased under the condition that the data source of the previous error model is less, the weakening of the fine adjustment repair capability of the error model to the data is reduced, and the actual operation of the data fine adjustment in the S5 is as follows: and taking the central point of the most intersected part of the plurality of mutual-capacitance correction circles as actual data of the next operation of the learning object to finish fine adjustment and correction of the data.
Example 2:
referring to fig. 4, the deep learning simulation consolidation method based on cloud computing includes the following steps:
s1, firstly, determining a deep learning object, which can be deep learning of unmanned vehicle speed control, and then listing influence factors influencing the object;
s2, setting a simulation road section through the three-dimensional simulator, simulating a deep learning scene according to influence factors, and modifying influence factor data of the simulated environment through the remote control terminal to realize the simulation of various different simulation environments;
s3, after the simulation is completed, enabling the unmanned vehicle to run on the simulation road section for multiple times, controlling the three-dimensional simulator to change different data of influencing factors through the remote control terminal, continuously recording the running speed of the unmanned vehicle under various different conditions, and uploading the running speed to the cloud;
s4, repeating the multiple simulated driving for multiple times, recording the driving speed again, and uploading to the cloud;
s5, obtaining the optimal running speed under various conditions by the cloud computing and comparing multiple groups of running speeds obtained by the cloud end;
s6, the cloud end imports the result obtained in the step into an unmanned vehicle control system to be used as the driving speed of the unmanned vehicle in various conditions;
s7, the cloud end imports the learning data into a control system of the unmanned vehicle, the unmanned vehicle performs actual operation according to the optimal driving speed, segments the road section actually operated by the unmanned vehicle according to the fact that there is a significant difference between two actual environments, such as smoothness, dryness and wetness of the road section, and presence or absence of a curve, and sets up a plurality of different data acquisition points, and the plurality of data acquisition points are in signal connection, and acquire actual data thereof through the data acquisition module to obtain a plurality of bidirectional unmanned vehicle driving data sets, and imports the cloud end again, in this embodiment, the data acquisition module includes a speed sensor, a distance sensor, and a memory;
and S8, comparing the actual data with the optimal running speed through cloud calculation by the cloud end, establishing an error model, continuously performing fine adjustment and correction on the actual operation data of the unmanned vehicle through the error model, and taking the actual operation data as the next actual operation data, so that the accuracy and the operation sensitivity of the actual operation of the deep unmanned vehicle are well consolidated.
S8, the bidirectional actual data set comprises a data set similar to the simulation environment and a data set with obvious difference between the actual environment and the simulation environment;
s81, the cloud end compares a data set similar to the simulation environment with the optimal learning data through cloud calculation, establishes an error model, and continuously performs fine adjustment and correction on actual operation data of the unmanned vehicle through the error model, and the actual operation data is used as next actual operation data, so that the accuracy and the operation sensitivity of the actual operation of the unmanned vehicle are well consolidated;
s82, the cloud end carries out data analysis on the data group which is obviously different from the simulation environment and the optimal learning data through cloud calculation, the difference factor between the simulation environment and the actual environment is reflected through the comparison difference, the difference factor is fed back to the three-dimensional simulator and serves as the influence factor of the three-dimensional simulator for carrying out environment simulation, the depth of the simulation environment is enlarged, the difference between the simulation environment and the actual operation environment is reduced, and the comprehensiveness of deep learning of the unmanned vehicle is improved.
In the embodiment, the influence factors mainly include the obstacles, the states of the obstacles, the number of the obstacles and the distance between the unmanned vehicle and the obstacles, the speed control of the unmanned vehicle under different conditions can be deeply learned, the actual operation of the unmanned vehicle is more accurate and sensitive under the correction of an error model which continuously enlarges a data source in the later period, and the traffic safety of the unmanned vehicle in the unmanned state is improved.
The error model may be set by the error between a data set similar to the simulated environment and the ideal data, the fine adjustment and correction can be carried out on the next operation data, thereby improving the precision and the operation sensitivity of the actual operation of the deep learning object, in the actual fine adjustment and correction process, through the arrangement of the difference nodes and the mutual capacitance correction rings outside the difference nodes, can continuously reduce the determining range of the next operation data point, effectively improve the precision of the next operation data point, meanwhile, the difficulty in judging the data points of the next operation can be reduced, the efficiency of the error model for fine adjustment and restoration of the data can be effectively improved, the whole operation efficiency of the deep learning object is improved, meanwhile, the depth of the simulation environment is expanded through the difference between the data group which has obvious difference with the simulation environment and ideal data, the difference between the simulation environment and the actual operation environment is reduced, and further the comprehensiveness of the deep learning object in deep learning is improved.
The above; but are merely preferred embodiments of the invention; the scope of the invention is not limited thereto; any person skilled in the art is within the technical scope of the present disclosure; the technical scheme and the improved concept of the invention are equally replaced or changed; are intended to be covered by the scope of the present invention.
Claims (10)
1. The deep learning simulation consolidation method based on cloud computing is characterized by comprising the following steps: the method comprises the following steps:
s1, firstly, determining a deep learning object, and then listing influence factors influencing the object;
s2, simulating the environment through the three-dimensional simulator, simulating the scene of deep learning according to the influence factors, and modifying the influence factor data of the simulated environment through the remote control terminal to realize the simulation of various different simulated environments;
s3, after the simulation is completed, performing deep learning on the determined object in the simulated environment, storing deep learning data in the simulated environment in a cloud end, and obtaining optimal learning data through cloud computing;
s4, the cloud end imports the optimal learning data into a control system of a learning object, the learning object performs actual operation according to the optimal learning data, the actual operation environment is segmented according to the fact that the two actual environments are obviously different, a plurality of different data acquisition points are set, the data acquisition points are in signal connection, the data acquisition points acquire the actual data through a data acquisition module to obtain a plurality of bidirectional actual data groups, and the actual data groups are imported into the cloud end again;
s5, the bidirectional actual data set comprises a data set similar to the simulation environment and a data set with obvious difference between the actual environment and the simulation environment;
s51, the cloud end compares a data set similar to the simulation environment with the optimal learning data through cloud calculation, establishes an error model, and continuously performs fine adjustment and correction on the actual operation data of the learning object through the error model, and the actual operation data is used as the next actual operation data, so that the accuracy and the operation sensitivity of the actual operation of the deep learning object are well consolidated;
s52, the cloud end carries out data analysis on the data group which is obviously different from the simulation environment and the optimal learning data through cloud calculation, the difference factor between the simulation environment and the actual environment is reflected through the comparison difference, the difference factor is fed back to the three-dimensional simulator and serves as the influence factor of the three-dimensional simulator for carrying out environment simulation, the depth of the simulation environment is enlarged, the difference between the simulation environment and the actual operation environment is reduced, and the comprehensiveness of deep learning of a deep learning object is improved.
2. The deep learning simulation consolidation method based on cloud computing of claim 1, characterized in that: the number of the influencing factors in the step S1 is not less than three, and when the learning object performs actual operation after deep learning, data of each actual operation is used as a data source of the error model.
3. The deep learning simulation consolidation method based on cloud computing of claim 1, characterized in that: the method for acquiring the optimal learning data in S3 specifically includes:
s31, controlling the three-dimensional simulator to change different data of the influence factors through the remote control terminal, collecting deep learning data of the learning object under the condition of various different influence factors, and uploading the deep learning data to the cloud;
s32, repeating the deep learning of the multiple scenes, collecting multiple deep learning data in the same deep learning scene, and uploading the deep learning data to the cloud end;
and S33, comparing multiple deep learning data in the same deep learning scene and multiple groups of data in different deep learning scenes by the cloud to obtain the optimal learning data in different deep learning scenes.
4. The deep learning simulation consolidation method based on cloud computing of claim 1, characterized in that: the data acquisition module in the S4 includes a data acquisition device and a memory, the data acquisition device may be but is not limited to a temperature sensor, a speed sensor, an anti-collision sensor and a brightness sensor, the data acquisition device is in signal connection with the memory, and the memory is in signal connection with the cloud.
5. The deep learning simulation consolidation method based on cloud computing of claim 1, characterized in that: the remote control terminal comprises a shell, a touch screen, a built-in control chip and a wireless signal transmitter, wherein the wireless signal transmitter and the control chip are in signal connection with the three-dimensional simulator.
6. The deep learning simulation consolidation method based on cloud computing of claim 1, characterized in that: the error model is an error between the multi-operation data and ideal data, and the error model is marked with a node of each error, and the ideal data is the best learning data in S3.
7. The deep learning simulation consolidation method based on cloud computing of claim 6, characterized in that: and a mutual capacitance correction ring is arranged outside each error node.
8. The deep learning simulation consolidation method based on cloud computing of claim 7, characterized in that: the radius of the mutual capacitance correction ring is set according to the distance between the error nodes of the first three marks and the ideal data point, and the distance between the three error nodes and the ideal data point is not obviously different.
9. The deep learning simulation consolidation method based on cloud computing of claim 8, characterized in that: the mutual capacitance correction circle radius is smaller than the average distance between the three error nodes and the ideal data point and is larger than half of the average distance.
10. The deep learning simulation consolidation method based on cloud computing of claim 9, characterized in that: the actual operation of the data trimming correction in S5 is: and taking the central point of the most intersected part of the mutual-capacitance correction circles as actual data of the next operation of the learning object to finish fine adjustment and correction of the data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010029453.2A CN111275182B (en) | 2020-01-13 | 2020-01-13 | Deep learning simulation consolidation method based on cloud computing |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010029453.2A CN111275182B (en) | 2020-01-13 | 2020-01-13 | Deep learning simulation consolidation method based on cloud computing |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111275182A true CN111275182A (en) | 2020-06-12 |
| CN111275182B CN111275182B (en) | 2022-05-31 |
Family
ID=71003026
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010029453.2A Active CN111275182B (en) | 2020-01-13 | 2020-01-13 | Deep learning simulation consolidation method based on cloud computing |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111275182B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114492875A (en) * | 2022-02-17 | 2022-05-13 | 海澜智云(上海)数据科技有限公司 | Wisdom energy management system based on industry internet |
| CN114490036A (en) * | 2021-12-28 | 2022-05-13 | 西北工业大学 | Extensible distributed redundancy unmanned aerial vehicle intelligent flight control computer |
| CN115906295A (en) * | 2023-03-09 | 2023-04-04 | 湖南大学 | Unmanned aerial vehicle health monitoring method and device based on digital twins |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107766889A (en) * | 2017-10-26 | 2018-03-06 | 济南浪潮高新科技投资发展有限公司 | A kind of the deep learning computing system and method for the fusion of high in the clouds edge calculations |
| CN108327841A (en) * | 2018-01-26 | 2018-07-27 | 浙江大学 | A kind of unmanned bicycle of self-balancing and its control method |
| CN109299812A (en) * | 2018-08-23 | 2019-02-01 | 河海大学 | A kind of Forecasting Flood method based on deep learning model and KNN real time correction |
| CN110320883A (en) * | 2018-03-28 | 2019-10-11 | 上海汽车集团股份有限公司 | A kind of Vehicular automatic driving control method and device based on nitrification enhancement |
| CN110472738A (en) * | 2019-08-16 | 2019-11-19 | 北京理工大学 | A kind of unmanned boat Real Time Obstacle Avoiding algorithm based on deeply study |
-
2020
- 2020-01-13 CN CN202010029453.2A patent/CN111275182B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107766889A (en) * | 2017-10-26 | 2018-03-06 | 济南浪潮高新科技投资发展有限公司 | A kind of the deep learning computing system and method for the fusion of high in the clouds edge calculations |
| CN108327841A (en) * | 2018-01-26 | 2018-07-27 | 浙江大学 | A kind of unmanned bicycle of self-balancing and its control method |
| CN110320883A (en) * | 2018-03-28 | 2019-10-11 | 上海汽车集团股份有限公司 | A kind of Vehicular automatic driving control method and device based on nitrification enhancement |
| CN109299812A (en) * | 2018-08-23 | 2019-02-01 | 河海大学 | A kind of Forecasting Flood method based on deep learning model and KNN real time correction |
| CN110472738A (en) * | 2019-08-16 | 2019-11-19 | 北京理工大学 | A kind of unmanned boat Real Time Obstacle Avoiding algorithm based on deeply study |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114490036A (en) * | 2021-12-28 | 2022-05-13 | 西北工业大学 | Extensible distributed redundancy unmanned aerial vehicle intelligent flight control computer |
| CN114492875A (en) * | 2022-02-17 | 2022-05-13 | 海澜智云(上海)数据科技有限公司 | Wisdom energy management system based on industry internet |
| CN115906295A (en) * | 2023-03-09 | 2023-04-04 | 湖南大学 | Unmanned aerial vehicle health monitoring method and device based on digital twins |
| CN115906295B (en) * | 2023-03-09 | 2023-05-12 | 湖南大学 | Unmanned aerial vehicle health monitoring method and device based on digital twinning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111275182B (en) | 2022-05-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111275182B (en) | Deep learning simulation consolidation method based on cloud computing | |
| CN107451607B (en) | A kind of personal identification method of the typical character based on deep learning | |
| CN105872477A (en) | Video monitoring method and system | |
| CN110211097A (en) | Crack image detection method based on fast R-CNN parameter migration | |
| CN107817798A (en) | A kind of farm machinery barrier-avoiding method based on deep learning system | |
| CN109800689A (en) | A kind of method for tracking target based on space-time characteristic fusion study | |
| CN103324937A (en) | Method and device for labeling targets | |
| CN114067143B (en) | Vehicle re-identification method based on double sub-networks | |
| CN107705324A (en) | A kind of video object detection method based on machine learning | |
| CN105046197A (en) | Multi-template pedestrian detection method based on cluster | |
| CN113074959B (en) | Automatic driving system test analysis method | |
| CN107316315A (en) | A kind of object recognition and detection method based on template matches | |
| CN108305283A (en) | Human bodys' response method and device based on depth camera and basic form | |
| CN103196430A (en) | Mapping navigation method and system based on flight path and visual information of unmanned aerial vehicle | |
| CN108682022B (en) | Visual tracking method and system based on anti-migration network | |
| CN103714077A (en) | Method and device for retrieving objects and method and device for verifying retrieval | |
| CN103105924B (en) | Man-machine interaction method and device | |
| CN104778465A (en) | Target tracking method based on feature point matching | |
| CN102200787A (en) | Robot behaviour multi-level integrated learning method and robot behaviour multi-level integrated learning system | |
| CN115617217B (en) | Vehicle state display method, device, equipment and readable storage medium | |
| CN104850120A (en) | Wheel type mobile robot navigation method based on IHDR self-learning frame | |
| CN110298330A (en) | A kind of detection of transmission line polling robot monocular and localization method | |
| CN103389728A (en) | Test simulation system and test simulation method of safety air bag controller | |
| Dong et al. | A novel loop closure detection method using line features | |
| CN114861761A (en) | Loop detection method based on twin network characteristics and geometric verification |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |