CN109241603A - A kind of fixed intelligence platform gear of riding divides and power approximating method - Google Patents
A kind of fixed intelligence platform gear of riding divides and power approximating method Download PDFInfo
- Publication number
- CN109241603A CN109241603A CN201810989398.4A CN201810989398A CN109241603A CN 109241603 A CN109241603 A CN 109241603A CN 201810989398 A CN201810989398 A CN 201810989398A CN 109241603 A CN109241603 A CN 109241603A
- Authority
- CN
- China
- Prior art keywords
- data
- platform
- power
- pwm wave
- cluster
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01D—MEASURING NOT SPECIALLY ADAPTED FOR A SPECIFIC VARIABLE; ARRANGEMENTS FOR MEASURING TWO OR MORE VARIABLES NOT COVERED IN A SINGLE OTHER SUBCLASS; TARIFF METERING APPARATUS; MEASURING OR TESTING NOT OTHERWISE PROVIDED FOR
- G01D21/00—Measuring or testing not otherwise provided for
- G01D21/02—Measuring two or more variables by means not covered by a single other subclass
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M13/00—Testing of machine parts
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/60—Other road transportation technologies with climate change mitigation effect
- Y02T10/72—Electric energy management in electromobility
Abstract
A kind of fixed intelligence platform gear of riding divides and power approximating method, is adjusted first to the ride PWM wave percentage of platform control circuit board of fixed intelligence；Then, by power test machine to intelligence ride platform difference PWM wave percentage when velocity amplitude and corresponding performance number tested and recorded, it needs the initial temperature to platform of riding to measure and record before test every time, controls the difference of initial temperature in default range.Secondly, the K-means algorithm in the unsupervised learning of machine learning is recycled to cluster the velocity amplitude, corresponding performance number, the temperature value that record after PWM wave percentage adjusted, test, abnormality processing first is carried out to data before cluster, and standardization, then K-means algorithm is recycled to be clustered, the weight obtained further according to cluster, obtain the classification number of the classification of PWM wave, that is gear number, local weighted polynomial regression finally is carried out to data further according to weight and calculates power speed curve, the final output power for platform of riding is obtained by curve.
Description
Technical field
The present invention relates to a kind of fixed intelligence ride platform gear divide and power approximating method, and in particular to are as follows: pass through
Fixed intelligence is ridden the PWM wave percentage of platform, is ridden the speed and function of platform in conjunction with Speed-power test machine to fixed intelligence
Rate is tested and is recorded, and the platform internal temperature values of riding before each test are measured and remembered using technical grade thermometer
Record；Reuse K-means clustering algorithm to fixed intelligence ride platform carry out gear division, according to K-means clustering algorithm
As a result the calculating of PWM wave percentage weight is carried out, is finally ridden the output of platform using the method for polynomial regression to fixed intelligence
Power is fitted.
Background technique
With the development of society, the improvement of people's living standards, more and more people begin to focus on how to improve oneself
Quality of the life, movement of riding is gradually by the favor of more and more people.Riding can not only lose weight, but also can make stature
It is well-balanced.It is the movement for needing a large amount of oxygen due to riding, the effect of strengthening cardiac function, preventing hypertension can be played.But
It rides and is frequently subjected to the influence of the situations such as weather, more and more people is made to be intended to ride using cycling platform progress interior
It takes exercise, can not only play the role of body-building can also avoid influence of the weather reason to outdoor movement of riding.
Currently, indoor body-building person, when platform is ridden in selection, more and more people tend to platform of riding using fixed intelligence
For building body training is carried out, platform is easy to carry because fixed intelligence is ridden, small in size, light weight, and fixed intelligence is ridden
Row platform does not need to ride as direct-drive type when platform is installed and removes the rear tyre of bicycle；But directly bicycle can be fixed on
Fixed intelligence ride platform bracket on, it is easy for installation, and due to fixed intelligence ride platform power be by speed turn
Power realization is changed, without using expensive power meter, fixed intelligence is greatly reduced and rides the production cost of platform, together
When so that the ride price of platform of fixed intelligence is also far below direct-drive type and is intelligently ridden the market price of platform, so that interior is ridden body builder
More favor is in carrying out body building of riding using fixed intelligence platform of riding.
It rides platform industry for current intelligent body-building, Yu Xin Yi, Chen Wei propose a kind of bicycle body-building device stepless-adjustment
Save magnetic control means (Yu Xin Yi, Chen Wei bicycle body-building device step-less adjustment magnetic control means: China, 204447224 [P] .2015-
07-08), the step-less adjustment and internal mechanical structure and driving method that bicycle body-building device gear controls are described in detail,
How gear is not divided and how power obtains and study；The multi-functional body-sensing that Zhu Yi, Cao Yicong are proposed is synchronous
Health-care bicycle entertainment systems (Zhu Yi, the one multi-functional body-sensing synchronous self vehicle body-building recreation system of acute hearing of Cao: China,
204073263 [P] .2015-01-07), it realizes body-building and is integrated with amusement, but it only describes bicycle and computer
It is connected and realizes entertainment effect；Kong Fanbin, Yu Feng, Meng, which enable, just have been proposed a kind of electromagnetism the super-silent intelligent power of resistance is added to ride platform
(Kong Fanbin, Yu Feng, Meng enable a kind of electromagnetism of rigid that the super-silent intelligent power of resistance is added to ride platform: China, 106890444A.2017-
06-27), this is that a kind of intelligence of direct-drive type is ridden platform, and installation is inconvenient, and built-in power meter, and price is more expensive, and
And it does not also study the division methods for platform gear of riding.
Ride in platform industry in intelligent body-building, for how fixed intelligence ride introduce in platform machine learning without prison
Speed and power when educational inspector's learning method is ridden under controlling different PWM wave percentages carry out K-means cluster and realize gear
It divides；It obtains being fixed in each class relative to the weight of center of gravity and curve-fitting method after clustering in conjunction with K-means
Formula is intelligently ridden the synthesis of platform power, thus make as far as possible gear divide rationally, output power error reduces show as far as possible
It obtains most important.
Summary of the invention
The present invention overcomes the disadvantage in existing method, propose a kind of fixed intelligence ride platform gear divide and power fitting
Method, specific method flow chart are as shown in Figure 1.
First according to technical grade temperature measurer, Speed-power test machine to the fixed intelligence in the case of different PWM wave percentages
Initial temperature value, running speed and the performance number of platform of capable of riding is tested and is acquired, by collected temperature, speed,
Power and corresponding PWM wave percentage are recorded；Due to test in external disturbance and data record when fault meeting
Record out it is some be apparently higher than or lower than normal value data, at this time, it may be necessary to these data carry out abnormality processing；Due to note
There are power features, velocity characteristic, temperature profile, the guiding principle amount of each feature and the difference of numberical range in the data of record, then needs
The data after abnormality processing are standardized again, are just distributed very much so that the distribution of each feature is all near the mark；
Then it reuses K-means algorithm to cluster the data Jing Guo standardization, since there are more than one set in every one kind
Data, but fixed intelligence is ridden, platform is practical ride during a velocity amplitude can only correspond to a performance number, and cluster
There are a weights between the data in every one kind and such center of gravity distance afterwards will be each after cluster according to the difference of weight
Multi-group data is weighted in class, so that the multi-group data in one kind is become one group of data, is at this moment clustered and is calculated by K-means
The class number that method obtains is that fixed intelligence is ridden the gear number that platform finally needs to divide；One group can all be corresponded to due to every grade
Speed and power data, the speed to each grade and power data fit speed-function using the method for polynomial regression at this time
Rate curve, and obtain power-speed fitting formula, the velocity amplitude of the fitting formula are that fixed intelligence is ridden actually the riding of platform
Scanning frequency degree, the performance number of the fitting formula are the corresponding performance number of the speed (the i.e. power of user at this time in practical ride
Output valve).
A kind of fixed intelligence platform gear of riding divides and power approximating method, the specific steps are as follows:
Step 1: obtaining fixed intelligence and ride platform PWM wave percentage, velocity amplitude, performance number, initial temperature Value Data；
The formula that is fixed intelligently ride platform gear divide and power fitting before, need platform control of riding to fixed intelligence
The PWM wave percentage of wiring board processed is adjusted；It is different that different PWM wave percentage simulates climb and fall resistance when riding；
Speed-power test machine is recycled to ride the running velocity amplitude of platform the fixed intelligence in the case of different PWM wave percentages
It is tested and is acquired with performance number, collected velocity amplitude, performance number and corresponding PWM wave percentage are recorded；
Before carrying out speed and power test according to different PWM wave percentage every time, need using technical grade temperature
Measuring instrument measures fixed intelligence and rides the initial temperature of platform, it is ensured that fixed intelligence is ridden the initial temperature of platform before measuring every time
Difference records the initial temperature value size before measurement every time in default range；
In test, the ride PWM wave percentage of platform control circuit board of fixed intelligence is adjusted every time, PWM wave hundred
Divide ratio that can adjust from 0.0% to 100.0%；The PWM wave amplitude of each testing and debugging, which can according to need, to be configured；And
In order to ensure the reliability of data, the platform that needs to ride to the fixed intelligence in the case of every kind of PWM wave percentage does multiple test,
And record the corresponding initial temperature value of corresponding PWM wave percentage, velocity amplitude and power Value Data；
Step 2: outlier processing is carried out to the initial data obtained by step 1；
When due to the measurement error and data record of Speed-power test machine and technical grade temperature measuring set itself
External disturbance present in record fault and test makes to be apparently higher than or there are some lower than normal in the data of step 1 acquisition
The power features of value；If do not handled these data, it will influence the reliability of cluster；The processing of exceptional value is used
Average value is corrected；
Step 3: the data obtained by step 2 are standardized；
Since existing temperature profile has velocity characteristic in ride platform speed and power data again, between different characteristic dimension and
Numberical range is different；If quantity difference is excessive between different characteristic, the work that the small feature of numerical value is played in cluster will lead to
With becoming smaller；To solve the above-mentioned problems, it needs first to be standardized data, so that the distribution of each feature is all near the mark just
State distribution；The data normalization method used is as follows:
Wherein xiRepresent ith feature, μiRepresent the mean value of this feature, σiThe standard deviation of this feature is represented, m represents data
Total number, xijJ-th strip data ith feature is represented, T represents the transposition of matrix；
Step 4: being clustered with K-means algorithm to by pretreated data；
K-means algorithm is the process of the mobile class central point of a repetition, the central point of class, also referred to as center of gravity
(centroids), it is moved to the mean place it includes member, so that having the cluster of the data formation of similitude one by one, then
Repartition its internal members；K is given hyper parameter, indicates the quantity of class；K-means can distribute sample to difference automatically
Class, but cannot determine to divide several classes actually；K must be a positive integer smaller than training set sample number；
The parameter of K-means is the position of centre of gravity of class and the position of its internal observation value；The optimal solution of K-means parameter is
Using cost function minimization as target；K-means loss function J is defined as follows:
Wherein, k represents the class number to be divided into, CiRepresent i-th of cluster, μiThe center of gravity of i-th of cluster is represented, T represents turning for matrix
It sets；The iterative algorithm for realizing that loss function minimizes is as follows:
(1) k center of gravity is randomly selected；
(2) all data are traversed, each data are divided into nearest cluster；
(3) average value of each cluster is calculated, and as new center of gravity；
(4) step (2) and step (3) are repeated, until significant change no longer occurs for the center of gravity of this k cluster；
Step 5: the PWM wave of each gear is determined with local weighted method；
It is clustered in resulting each gear by step 4, the corresponding PWM wave of each data is different, it is thus necessary to determine that one only
One PWM wave could control platform of riding；Obviously, the PWM wave of each class is by the PWM wave weighting summation of data in such
Gained, the size of weight and data and such center of gravity are square related apart from size；Determine that the formula of class PWM wave is as follows:
Wherein, PWMiIt is the PWM wave of the i-th class, CiRepresent i-th of cluster, pjIt is the PWM wave of j-th of data in the i-th class, WiIt is
Weight, ρjRepresentative is the distance coefficient of j-th of data and center of gravity in the i-th class, μiRepresent the center of gravity of i-th of cluster；
Step 6: calculating power speed curve with local weighted polynomial regression；
Obviously, the difference of power speed curve corresponding to obtained each gear, and the pass of highway speeds are clustered as step 4
System is not a curve；In order to more accurate fitting power rate curve, local weighted polynomial regression is used；Use quadravalence
Polynomial regression fit power speed curve；It is as follows to construct 4 rank multinomial features:
X=[x0,x1,x2,x3,x4] (8)
The wherein multinomial that X is made of x, x representation speed；
It is as follows to define loss function:
Wherein yj, X respectively represent the corresponding power of j-th of data and speed multinomial in i-th grade, θ is returning for curve
Return coefficient, T represents the transposition of matrix；
Allow Li(θ) seeks local derviation to θ, and enablesObtain following optimum regression coefficient matrix:
Finally, i-th grade of power speed curve is expressed as follows:
I.e. final speed-power matched curve, then the final output work of platform of riding is obtained by speed-power matched curve
Rate.
Advantages of the present invention: a kind of fixed intelligence of the present invention platform gear of riding divides and power approximating method,
It rides the PWM wave percentage of platform control panel by adjusting fixed intelligence, and rides in conjunction with power test machine to fixed intelligence
The speed and power of platform are measured and are recorded, then introduce the method k-means algorithm of the unsupervised learning of machine learning, with
PWM wave percentage, speed, power, initial temperature are clustered as characteristic value, according to the classification number obtained after cluster, are carried out
The ride gear of platform of fixed intelligence divides, and divides to obtain the fixed intelligence reasonable gear of platform of riding, further according to K-
The weight of means clustering algorithm arrived, it is local weighted to data progress, so that it is determined that the PWM wave percentage value of each gear,
And the corresponding output power size of platform of intelligently riding under each gear friction speed, then by this time speed and performance number utilize
The method of polynomial regression fits power-rate curve, to obtain power-velocity fitting equation, obtains fixed intelligence
It rides output power value of the platform in riding when different stalls friction speed, and such method can determine after experiment test
Such gear division methods are very reasonable, and the power data after fitting is more accurate.And such method can allow no power to pass
The fixed platform of riding of sensor realizes the transformation of speed and performance number, improves fixed intelligence and rides the cost performance of platform.
Detailed description of the invention
Fig. 1 flow chart of the method for the present invention.
The fixed intelligence of Fig. 2, which is ridden, platform and rides platform PWM wave control circuit board pattern diagram.
Test and acquisition schematic diagram of Fig. 3 Speed-power test machine to velocity amplitude, performance number.
The temperature acquisition schematic diagram of Fig. 4 technical grade temperature measuring set.
Fig. 5 K-means algorithm flow chart
The PWM wave of each gear of Fig. 6
1~17 grade of Fig. 7 of speed-power curve
Specific embodiment
Technical solution of the present invention is further illustrated with reference to the accompanying drawing.
One kind of the invention be based on fixed intelligence ride platform gear divide and power combining methods, detailed process is as follows:
It is freely that fixed intelligence that intelligent technology limited provides platform of riding carries out Speed-power test using Yiwu,
The fixed intelligence rides platform and platform control circuit board of riding as shown in Fig. 2, the left side Fig. 2 is that fixed intelligence is ridden platform, Fig. 2's
Right-hand component is that fixed intelligence is ridden platform PWM wave control circuit board.The formula that is fixed intelligently ride platform gear divide and
Before power combing, by modify fixed intelligence in Fig. 2 ride platform PWM wave control panel PWM wave percentage it is fixed to change
Intelligence is ridden the drag size of platform, and the percentage of PWM wave changes 0.2% every time come in simulating and actually riding when making to test every time
Climb and fall drag effect.The platform that needs to ride fixed intelligence when test is removed from fixed bracket, and is installed into speed-
Power test machine makes the ride roller portion of platform of the drive disk of Speed-power test machine and fixed intelligence fit closely and prevent
It skids, as shown in figure 3, ride the main part of platform in the dotted line frame in the lower left corner to remove the fixed intelligence after fixed bracket,
The drive disk of the idler wheel of the part and Speed-power test machine fits closely and prevents from skidding, and two dotted line frames in the upper right corner are
The visualization interface of Speed-power test machine, interface information have the Speed-power curve graph after each run, and instantaneous speed
The display window of degree and performance number.
In addition, needing to guarantee every time when in order to reduce the test of every time ride to fixed intelligence platform speed and performance number
When test fixed intelligence ride platform initial temperature value difference be no more than 10 degrees Celsius.And initial temperature value is recorded, is made
Feature when for K-means cluster, to fixed intelligence ride platform inside temperature measurement using industry as shown in Figure 4
Grade temperature measuring set measures.
When complete every time fixed intelligence ride platform initial temperature value measurement after, starting Speed-power test machine into
Fixed intelligence is ridden platform speed and performance number in the case of the different PWM wave percentages of row, and record 10km/h, 15km/h,
20km/h、25km/h、30km/h、35km/h、40km/h、45km/h、50km/h、55km/h、60km/h、65km/h、70km/h、
Corresponding performance number size when 75km/h, 80km/h.
Check by above-mentioned steps obtain data in the presence or absence of numerical value be more than bound abnormal point, each feature it is upper
Lower limit value is shown in Table 1, and (friction speed value corresponds to the bound and PWM wave percentage bound, initial temperature of performance number or more
Limit).Abnormal point if it exists then by calculating the mean value under identical speed and identical PWM wave, and replaces abnormal point with mean value.
The upper lower limit value of each feature of table 1
The mean value and standard deviation of each feature are calculated according to formula (2) and (3), as shown in table 2 (after the test of friction speed value
Corresponding performance number mean value and standard deviation, PWM wave percentage mean value and standard deviation, initial temperature mean value and standard deviation),
Each feature in data is standardized with formula (1) again.
Wherein xiRepresent ith feature, μiRepresent the mean value of this feature, σiRepresent the standard deviation of this feature.
The mean value and standard deviation of each feature of table 2
| Feature | Mean value | Standard deviation |
| 10km/h | 71.9 (watts) | 17.4 (watts) |
| 15km/h | 117 (watts) | 33.7 (watts) |
| 20km/h | (170.5 watt) | 56.1 (watts) |
| 25km/h | (229.3 watt) | 78.9 (watts) |
| 30km/h | (292.0 watt) | (102.0 watt) |
| 35km/h | (358.1 watt) | (128.5 watt) |
| 40km/h | (416.5 watt) | (148.7 watt) |
| 45km/h | (480.9 watt) | (171.7 watt) |
| 50km/h | (545.2 watt) | (197.0 watt) |
| 55km/h | (611.2 watt) | (220.3 watt) |
| 60km/h | (678.3 watt) | (242.8 watt) |
| 65km/h | (747.1 watt) | (266.9 watt) |
| 70km/h | (813.9 watt) | (289.5 watt) |
| 75km/h | (894.9 watt) | (306.2 watt) |
| 80km/h | (965.9 watt) | (333.3 watt) |
| PWM | 50.0% | 28.8% |
| Initial temperature | 14.92℃ | 2.2℃ |
Then classified with K-means unsupervised learning algorithm to data.It is herein 0.0%- by PWM wave range
100.0% 500 groups of data are divided into 17 classes.It is as follows to define loss function:
Wherein, k=17 represents the class number to be divided into, CiRepresent i-th of cluster, μiRepresent the center of gravity of i-th of cluster.Realize loss
The iterative algorithm of function minimization is as shown in Figure 5.
(1) 17 centers of gravity are randomly selected.
(2) all data are traversed, each data are divided into nearest cluster.
(3) average value of each cluster is calculated, and as new center of gravity.
It repeats (2) and (3), until significant change no longer occurs for the center of gravity of this 17 clusters.
The standard that each class is divided as gear later determines the PWM wave of each gear with local weighted method.Determine class
The formula of PWM wave is as follows:
Wherein, PWMiIt is the PWM wave of the i-th class, pjIt is the PWM wave of j-th of data in the i-th class, WiIt is weight, ρjRepresentative is
In i-th class j-th data and center of gravity distance coefficient.The barycentric coodinates of each cluster are as shown in table 3, and gear divides figure such as Fig. 6
It is shown.
The barycentric coodinates of each cluster of table 3
Reuse the performance number that local weighted method calculates each velocity characteristic in data.Method is similar with the calculating of PWM wave.
After obtaining the local weighted performance number under each speed, polynomial regression fit speed-power curve is used.Construct 4 rank multinomials
Feature is as follows:
X=[x0,x1,x2,x3,x4] (19)
Wherein x is speed, the vector that X is made of the multinomial of x.
It is as follows to define loss function:
Wherein yj, X respectively represent the corresponding power of j-th of data and speed multinomial in i-th grade, θ is returning for curve
Return coefficient.
Allow Li(θ) seeks local derviation to θ, and enablesIt is not difficult to show that (coefficient matrix is shown in Table following optimum regression coefficient matrix
4):
4 coefficient matrix of table
Every grade of corresponding speed-power curve may finally be obtained, as shown in Figure 7.
Then according to speed-power matched curve, institute was right when attachment coefficient matrix can both obtain friction speed under different stalls
Should ride the final output power value of platform.
Content described in this specification embodiment is only enumerating to the way of realization of inventive concept, protection of the invention
Range should not be construed as being limited to the specific forms stated in the embodiments, and protection scope of the present invention is also and in art technology
Personnel conceive according to the present invention it is conceivable that equivalent technologies mean.
Claims (1)
1. a kind of fixed intelligence is ridden, platform gear is divided and power approximating method, the specific steps are as follows:
Step 1: obtaining fixed intelligence and ride platform PWM wave percentage, velocity amplitude, performance number, initial temperature Value Data；
The formula that is fixed intelligently ride platform gear divide and power fitting before, need platform control line of riding to fixed intelligence
The PWM wave percentage of road plate is adjusted；It is different that different PWM wave percentage simulates climb and fall resistance when riding；It is sharp again
It is ridden the running velocity amplitude of platform and function with Speed-power test machine to the fixed intelligence in the case of different PWM wave percentages
Rate value is tested and is acquired, and collected velocity amplitude, performance number and corresponding PWM wave percentage are recorded；
Before carrying out speed and power test according to different PWM wave percentage every time, need to measure using technical grade temperature
Instrument measures fixed intelligence and rides the initial temperature of platform, it is ensured that fixed intelligence is ridden the initial temperature difference of platform before measuring every time
In default range, and record the initial temperature value size before measurement every time；
In test, the ride PWM wave percentage of platform control circuit board of fixed intelligence is adjusted every time, PWM wave percentage
It can adjust from 0.0% to 100.0%；The PWM wave amplitude of each testing and debugging, which can according to need, to be configured；And in order to
The reliability for ensuring data, the platform that needs to ride to the fixed intelligence in the case of every kind of PWM wave percentage does multiple test, and remembers
Picture recording answers the corresponding initial temperature value of PWM wave percentage, velocity amplitude and power Value Data；
Step 2: outlier processing is carried out to the initial data obtained by step 1；
Due to being recorded when the measurement error and data record of Speed-power test machine and technical grade temperature measuring set itself
External disturbance present in fault and test makes to be apparently higher than or there are some lower than normal value in the data of step 1 acquisition
Power features；If do not handled these data, it will influence the reliability of cluster；The processing of exceptional value is used average
Value is corrected；
Step 3: the data obtained by step 2 are standardized；
Since existing temperature profile has velocity characteristic in ride platform speed and power data again, dimension and numerical value between different characteristic
Range is different；If quantity difference is excessive between different characteristic, it will lead to the small feature of numerical value and play the role of becoming in cluster
It is small；To solve the above-mentioned problems, it needs first to be standardized data, the normal state point so that the distribution of each feature is all near the mark
Cloth；The data normalization method used is as follows:
Wherein xiRepresent ith feature, μiRepresent the mean value of this feature, σiThe standard deviation of this feature is represented, m represents the total of data
Item number, xijJ-th strip data ith feature is represented, T represents the transposition of matrix；
Step 4: being clustered with K-means algorithm to by pretreated data；
K-means algorithm is the process of a mobile class central point of repetition, the central point of class, also referred to as center of gravity (centroids),
It is moved to the mean place it includes member, so that there is the cluster of the data formation of similitude one by one, is then repartitioned in it
Portion member；K is given hyper parameter, indicates the quantity of class；K-means can distribute sample to different classes automatically, but not
It can determine to divide several classes actually；K must be a positive integer smaller than training set sample number；
The parameter of K-means is the position of centre of gravity of class and the position of its internal observation value；The optimal solution of K-means parameter be at
This function minimization is target；K-means loss function J is defined as follows:
Wherein, k represents the class number to be divided into, CiRepresent i-th of cluster, μiThe center of gravity of i-th of cluster is represented, T represents the transposition of matrix；
The iterative algorithm for realizing that loss function minimizes is as follows:
(1) k center of gravity is randomly selected；
(2) all data are traversed, each data are divided into nearest cluster；
(3) average value of each cluster is calculated, and as new center of gravity；
(4) step (2) and step (3) are repeated, until significant change no longer occurs for the center of gravity of this k cluster；
Step 5: the PWM wave of each gear is determined with local weighted method；
It is clustered in resulting each gear by step 4, the corresponding PWM wave of each data is different, it is thus necessary to determine that one unique
PWM wave could control platform of riding；Obviously, the PWM wave of each class is by the PWM wave weighting summation institute of data in such
, the size of weight and data and such center of gravity are square related apart from size；Determine that the formula of class PWM wave is as follows:
Wherein, PWMiIt is the PWM wave of the i-th class, CiRepresent i-th of cluster, pjIt is the PWM wave of j-th of data in the i-th class, WiIt is weight,
ρjRepresentative is the distance coefficient of j-th of data and center of gravity in the i-th class, μiRepresent the center of gravity of i-th of cluster；
Step 6: calculating power speed curve with local weighted polynomial regression；
Obviously, the difference of power speed curve corresponding to obtained each gear is clustered as step 4, and the relationship of highway speeds is simultaneously
It is not a curve；In order to more accurate fitting power rate curve, local weighted polynomial regression is used；It is more with quadravalence
Item formula regression fit power speed curve；It is as follows to construct 4 rank multinomial features:
X=[x0,x1,x2,x3,x4] (8)
The wherein multinomial that X is made of x, x representation speed；
It is as follows to define loss function:
Wherein yj, X respectively represent the corresponding power of j-th of data and speed multinomial in i-th grade, θ is the recurrence system of curve
Number, T represent the transposition of matrix；
Allow Li(θ) seeks local derviation to θ, and enablesObtain following optimum regression coefficient matrix:
Finally, i-th grade of power speed curve is expressed as follows:
I.e. final speed-power matched curve, then the final output power of platform of riding is obtained by speed-power matched curve.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810989398.4A CN109241603B (en) | 2018-08-28 | 2018-08-28 | Gear dividing and power fitting method for fixed intelligent riding platform |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810989398.4A CN109241603B (en) | 2018-08-28 | 2018-08-28 | Gear dividing and power fitting method for fixed intelligent riding platform |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109241603A true CN109241603A (en) | 2019-01-18 |
| CN109241603B CN109241603B (en) | 2023-05-26 |
Family
ID=65068725
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810989398.4A Active CN109241603B (en) | 2018-08-28 | 2018-08-28 | Gear dividing and power fitting method for fixed intelligent riding platform |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109241603B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117254745A (en) * | 2023-11-17 | 2023-12-19 | 深圳市精锐昌科技有限公司 | Operation control method, system and storage medium of motor |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103920287A (en) * | 2014-05-06 | 2014-07-16 | 许昌学院 | Virtual scene network exercise bicycle |
| CN105938116A (en) * | 2016-06-20 | 2016-09-14 | 吉林大学 | Gas sensor array concentration detection method based on fuzzy division and model integration |
| CN106647272A (en) * | 2016-12-23 | 2017-05-10 | 东华大学 | Robot route planning method by employing improved convolutional neural network based on K mean value |
| CN107198860A (en) * | 2017-07-18 | 2017-09-26 | 义乌畅为智能科技有限公司 | The Intelligentized Information system of for building body platform adapter |
| US20170309093A1 (en) * | 2014-11-11 | 2017-10-26 | Chunkui FENG | Vehicle operation monitoring, overseeing, data processing and overload monitoring method and system |
-
2018
- 2018-08-28 CN CN201810989398.4A patent/CN109241603B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103920287A (en) * | 2014-05-06 | 2014-07-16 | 许昌学院 | Virtual scene network exercise bicycle |
| US20170309093A1 (en) * | 2014-11-11 | 2017-10-26 | Chunkui FENG | Vehicle operation monitoring, overseeing, data processing and overload monitoring method and system |
| CN105938116A (en) * | 2016-06-20 | 2016-09-14 | 吉林大学 | Gas sensor array concentration detection method based on fuzzy division and model integration |
| CN106647272A (en) * | 2016-12-23 | 2017-05-10 | 东华大学 | Robot route planning method by employing improved convolutional neural network based on K mean value |
| CN107198860A (en) * | 2017-07-18 | 2017-09-26 | 义乌畅为智能科技有限公司 | The Intelligentized Information system of for building body platform adapter |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117254745A (en) * | 2023-11-17 | 2023-12-19 | 深圳市精锐昌科技有限公司 | Operation control method, system and storage medium of motor |
| CN117254745B (en) * | 2023-11-17 | 2024-03-22 | 深圳市精锐昌科技有限公司 | Operation control method, system and storage medium of motor |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109241603B (en) | 2023-05-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108960426B (en) | Road slope comprehensive estimation system based on BP neural network | |
| CN113255078A (en) | Bearing fault detection method and device under unbalanced sample condition | |
| CN110111563A (en) | A kind of real-time traffic states estimation method of city expressway | |
| CN104899135B (en) | Software Defects Predict Methods and system | |
| CN103895649B (en) | A kind of driver safety driving warning method | |
| CN106021789B (en) | Railway vehicle suspension system Fault Classification and system based on fuzzy intelligence | |
| CN112418013A (en) | Complex working condition bearing fault diagnosis method based on meta-learning under small sample | |
| CN109858104A (en) | A kind of rolling bearing health evaluating and method for diagnosing faults and monitoring system | |
| CN108152050A (en) | A kind of whole-car parameters calibration method | |
| CN110641523B (en) | Subway train real-time speed monitoring method and system | |
| CN108871788A (en) | A kind of automatic transmission shift attribute test rack and its method of calibration and shift quality evaluation method | |
| CN107180534B (en) | The express highway section average speed estimation method of support vector regression fusion | |
| CN109118052A (en) | Energy efficiency evaluation method | |
| CN104677641A (en) | Measurement method for simultaneously obtaining air resistance coefficient and rolling resistance coefficient of vehicle | |
| JP7318076B1 (en) | Evaluating the impact of switching working conditions on vehicle fuel consumption | |
| CN109241603A (en) | A kind of fixed intelligence platform gear of riding divides and power approximating method | |
| CN112364706A (en) | Small sample bearing fault diagnosis method based on class imbalance | |
| CN109018184A (en) | Bicycle intelligent speed changing method, intelligent speed changing device and intelligent speed changing bicycle | |
| CN103076146B (en) | Drop test seven-degree-of-freedom vehicle model-based vehicle parameter identification method | |
| CN108846200A (en) | A kind of quasi-static Bridge Influence Line recognition methods based on iterative method | |
| CN106384507A (en) | Travel time real-time estimation method based on sparse detector | |
| CN113837071A (en) | Partial migration fault diagnosis method based on multi-scale weight selection countermeasure network | |
| CN110867075A (en) | Method for evaluating influence of road speed meter on reaction behavior of driver under rainy condition | |
| CN107491656B (en) | Pregnancy outcome influence factor evaluation method based on relative risk decision tree model | |
| Li et al. | Transformer-based meta learning method for bearing fault identification under multiple small sample conditions |
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 |