CN105517019B - Using the method for integrated regression system detection LTE network performance - Google Patents
Using the method for integrated regression system detection LTE network performance Download PDFInfo
- Publication number
- CN105517019B CN105517019B CN201510947246.4A CN201510947246A CN105517019B CN 105517019 B CN105517019 B CN 105517019B CN 201510947246 A CN201510947246 A CN 201510947246A CN 105517019 B CN105517019 B CN 105517019B
- Authority
- CN
- China
- Prior art keywords
- regression
- model
- training set
- wine
- rice
- 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.)
- Active
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W24/00—Supervisory, monitoring or testing arrangements
- H04W24/02—Arrangements for optimising operational condition
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention provides a kind of method using integrated regression system detection LTE network performance, the following steps are included: the collection of (1) sample data set: the model that (2) pass through foundation, time delay of eating dishes without rice or wine is predicted using index is received, and the sample in training set is predicted in specific subset respectively using regression algorithm；(3) error of the error amount of each sample and regression algorithm in training set is derived；(4) error in regression algorithm is weighted by processing using analysis method, integrated regressive prediction model is constituted by weighted regression combination；(5) the integrated regression algorithm in step (4) is applied into test set, detects the accuracy of the model prediction relationship obtained based on training set.It is promoted by using the operation of the method for integrated regression system detection LTE network performance, it can analyze and be inferred to the time delay of eating dishes without rice or wine of user class, to make mobile operator identify it is higher eat dishes without rice or wine time delay the problem of cell, and then by optimization problem cell improve LTE network service quality.
Description
Technical field
The present invention relates to a kind of methods for detecting LTE network performance, more particularly, to a kind of using integrated regression system inspection
The method for surveying LTE network performance.
Background technique
In recent years, with the fast development of LTE wireless network, the generation and acquisition of communication data also flourish therewith
Come.In addition to the storage and management to this data flow, a biggish challenge is how to utilize this data preferably communication for service net
Network.Therefore, network performance and user experience quality (QoE) are assessed by converting data to corresponding network index to be become most
Whole target.By analysis, due to the time delay of eating dishes without rice or wine of the larger value will affect the network quality of reception and increase network interferences, this to
Family sensory experience has relatively straightforward influence, therefore the time delay that will eat dishes without rice or wine is as a core index.Under normal circumstances, when eating dishes without rice or wine
Prolonging is to be taken by soft harvest, but software and hardware input cost is higher, does not have all-round popularization condition.
For mobile communication business, most important time delay is end-to-end time delay, i.e. the receipts for having built up connection
Both ends are sent out, data packet is generated from transmitting terminal, the time delay being properly received to receiving end.According to business model difference, end-to-end time delay
One way time delay and backhaul time delay can be divided into, wherein one way time delay refers to that data packet is generated from transmitting terminal and correctly reaches by wireless network
The time delay of another receiving end, backhaul time delay refer to that data packet receives data packet to destination server from transmitting terminal generation and returns
Corresponding data packet is until transmitting terminal has correctly received the time delay of reply data packet.
Existing mobile communication is mainly interpersonal communication, with the miniaturization and intelligence of hardware device, not
The high speed between mobile communication more " people and objects " and " object and object " come connects application.Machine communication (Machine Type
Communication, MTC) service application range is very extensive, such as portable medical, car networking, smart home, Industry Control, ring
Border monitoring etc. will will push MTC system application explosive growth, and large number of equipment will access network, realize that really " all things on earth is mutual
Connection " brings boundless vital force for mobile communication.Meanwhile extensive MTC system application range can also bring new skill to mobile communication
Art challenge, such as real-time cloud computing, virtual reality, game on line, tele-medicine, intelligent transportation, smart grid, long-range control in real time
The business such as system are more sensitive to time delay, propose higher demand to time delay.
Therefore, it is necessary to develop a kind of detection method, the time delay of eating dishes without rice or wine of user class is analyzed and is inferred to, to make mobile fortune
Battalion quotient identify it is higher eat dishes without rice or wine time delay the problem of cell, and then by optimization problem cell improve LTE network service quality.
Summary of the invention
The object of the present invention is to provide a kind of methods with integrated regression system detection LTE network performance to solve to make to move
Operator identify it is higher eat dishes without rice or wine time delay the problem of cell, and then LTE network service quality is improved by optimization problem cell
Problem.
The technical solution of the invention is as follows:
Using the method for integrated regression system detection LTE network performance, comprising the following steps:
(1) collection of sample data set: reception achievement data and network each stage of LTE network time delay of eating dishes without rice or wine are searched
Collection forms sample data set, and the sample data set is divided into training set and test set；
Wherein, training set is used to find to receive the projected relationship between index and time delay of eating dishes without rice or wine；Test set is for detecting base
In the accuracy for the model prediction relationship that training set is obtained；
(2) in the operating process of the training set, by the model of foundation, time delay of eating dishes without rice or wine is predicted using index is received,
The sample in training set is predicted in specific subset respectively using regression algorithm；
(3) predicted value obtained by comparing model in above-mentioned steps (2) and time delay value of really eating dishes without rice or wine, derive training
Concentrate the error amount of each sample and the error of regression algorithm；
Wherein, for regression algorithm J, the column error amount in training set is then referred to as error J；
(4) error in regression algorithm is weighted by processing using analysis method, is combined by weighted regression and constitutes collection
At regressive prediction model；
(5) the integrated regression algorithm in step (4) is applied into test set, it is pre- detects the model obtained based on training set
The accuracy of survey relationship.
Further, the reception index for achievement data being received in the step (1) includes Reference Signal Received Power RSRP,
Reference Signal Received Quality RSRQ, Signal to Interference plus Noise Ratio SINR, the Physical Resource Block on Physical Uplink Shared Channel
Physical Resource Block PDSCH-PRB on PUSCH-PRB and Physical Downlink Shared Channel.
Further, use eight kinds of different regression algorithms respectively in specific subset in training set in step (2)
Sample predicted, be linear regression respectively, second order polynomial regression, three rank multinomials return, ridge regression, and LASSO is returned,
Elastic is returned, and GAM is returned and MARS is returned；
Wherein, the linear regression provides following formula (1):
E (y)=β0+β1x1+...+βdxd(1), E (y) represents predicted value in formula, and y, which is represented, assumes time delay of eating dishes without rice or wine,
x1..., xdIt represents and receives index；In the model, response variable y Gaussian distributed can be counted directly with least square method
It calculates and obtains corresponding fitting coefficient β0..., βd；
The second order polynomial regression algorithm first calculates primary, the Quadratic Orthogonal multinomial of each index set, to obtain
2D form carries out models fitting with 2D+1 parameters；
The three rank multinomials regression algorithm, in the selection of variable, the number of orthogonal polynomial is chosen from 1 time to 3 time,
To reduce the constraint of prediction model；
The ridge regression algorithm provides following formula (2):
Algorithm is by increasing penalty coefficient in formula, to factor beta0..., βdIt is limited
System, becomes contraction, to find the minimum variance of least squares estimator, wherein parameter t is according to E.Cule and M.De Iorio
The theory proposed based on control forecasting value variance is chosen automatically, and k represents k-th of coefficient.The range of k from 1,2,3 ...,
d；
The LASSO regression algorithm provides following formula (3):
Constraint function in formula limits the sum of absolute value of regression coefficient, removes
Constant coefficient, t value are chosen automatically；
The Elastic regression algorithm provides following formula (4):
Constraint function in formula is to the ridge regression algorithm
The linear combination of regularization is carried out with the constraint function in the LASSO regression algorithm, wherein α 1/2, t value are to choose automatically
's；
The GAM regression algorithm provides following formula (5):
G (E (y))=β0+f1(x1)+...+fd(xd) (5), g indicates generalized linear Copula, f in formula1..., fd
Indicate the non-linear relationship between input variable, β0For constant term, x1..., xdIt is to receive the data obtained in index from five；
The MARS regression algorithm provides following formula (6):
X → max (0, x-C) or x → max (0, C-x)；C ∈ R (6), in MARS recurrence, recurrence is synthesized to one
The linear combination of a hinge function.
Further, in the step (2), the calculating on training set is completed by successive Regression, by looking for
To suitable weight.
Further, in step (4), weight is inferred using column error obtained in step (3), utilizes weight optimization
The method of model gives following formula (7):
Wherein W1...W8By minimizing weighted least-squares method and limitation
Total weight calculation obtains, wherein εjIndicate the training error value with j-th of model on training set, wjIndicate its j-th of model
The weight of training error value on training set.
Further, during the step (4) infer weight, following formula (8) are given for constraint condition:
Wherein, wjIndicate training error value of its j-th of model on training set
Weight.
Further, in the step (5), by predicted value obtained in test set and known time delay of really eating dishes without rice or wine
Value is compared, and is given formula (9) and obtained bit error rate εtest:
Wherein, yiIndicate i-th sample in test set
Time delay of eating dishes without rice or wine true value,Indicate the latency prediction value of eating dishes without rice or wine of i-th of sample in test set, ntestIndicate test sample amount,
εtestIt is then model mean absolute error on training set；
In order to confirm the stability and accuracy that integrate regressive prediction model obtained in the step (4), need to carry out
A variety of different comparisons obtain a bit error rate ε firstly, the identical bit error rate is executed on training settrainIf εtestWith
εtrainBetween difference it is smaller, then show over-fitting do not occur, it means that the stability of integrated regressive prediction model；Secondly,
The predicted value of every kind of regression algorithm is calculated on test set, it is made respectively to obtain a bit error rate, by by these bit error rates
The ε obtained with integrated recurrencetestIt is compared, then can check the accuracy of integrated regressive prediction model.
The beneficial effects of the present invention are: through the invention in the method using integrated regression system detection LTE network performance
Integrated regression system, prediction is effectively derived using cell RF index and is eated dishes without rice or wine time delay, it is not only stable to integrate regressive prediction model
And can accurately predict time delay of eating dishes without rice or wine, it can carry out automatically calculate and model selection, therefore model have it is stronger adaptive
Ying Xing；It is proved by example, the theory of the assessment models is accurately that for the angle of user, integrated regression system is easier to
In obtaining and analysis index of correlation data, thus the prediction result for the time delay that obtains eating dishes without rice or wine.It is detected by using integrated regression system
The operation of the method for LTE network performance is promoted, and can analyze and be inferred to the time delay of eating dishes without rice or wine of user class, to make mobile operator
Identify it is higher eat dishes without rice or wine time delay the problem of cell, and then by optimization problem cell improve LTE network service quality.
Detailed description of the invention
Fig. 1 is operation workflow explanation of the embodiment of the present invention using the method for integrated regression system detection LTE network performance
Schematic diagram；
Fig. 2 gives the simulated data sets comparative illustration schematic diagram of the present invention integrated regression system and classical regression algorithm.
Specific embodiment
The preferred embodiment that the invention will now be described in detail with reference to the accompanying drawings.
Embodiment provides a kind of method using integrated regression system detection LTE network performance, the implementation of the detection method
Including (1) data source: receiving cell RF performance indicator data set；(2) it is associated with: each element in cell RF index and data set
It eats dishes without rice or wine the association of time delay value；(3) predict: by cell RF index, newly value estimates time delay of eating dishes without rice or wine.Other realities in addition to first item
Apply content: sample data: sample data is allocated according to training set and test set；Modeling processing: data transporting something containerized is counted
Regression algorithm is handled, and every kind of regression algorithm weight is calculated, in conjunction with regression algorithm and weight, to obtain Aksu River prediction letter
Number；Model verifying: verifying model result on fc-specific test FC collection, then knows the accuracy of predicted value.
Embodiment
Using the method for integrated regression system detection LTE network performance, as shown in Figure 1, comprising:
(1) collection of sample data set: reception achievement data and network each stage of LTE network time delay of eating dishes without rice or wine are searched
Collection forms sample data set, and the sample data set is divided into training set and test set；
Wherein, training set is used to find to receive the projected relationship between index and time delay of eating dishes without rice or wine；Test set is for detecting base
In the accuracy for the model prediction relationship that training set is obtained；
(2) in the operating process of the training set, by the model of foundation, time delay of eating dishes without rice or wine is predicted using index is received,
The sample in training set is predicted in specific subset respectively using regression algorithm；
(3) predicted value obtained by comparing model in above-mentioned steps (2) and time delay value of really eating dishes without rice or wine, derive training
Concentrate the error amount of each sample and the error of regression algorithm；
Wherein, for regression algorithm J, the column error amount in training set is then referred to as error J；
(4) error in regression algorithm is weighted by processing using analysis method, is combined by weighted regression and constitutes collection
At regressive prediction model；
(5) the integrated regression algorithm in step (4) is applied into test set, it is pre- detects the model obtained based on training set
The accuracy of survey relationship.
Further, the reception index for achievement data being received in the step (1) includes Reference Signal Received Power RSRP,
Reference Signal Received Quality RSRQ, Signal to Interference plus Noise Ratio SINR, the Physical Resource Block on Physical Uplink Shared Channel
Physical Resource Block PDSCH-PRB on PUSCH-PRB and Physical Downlink Shared Channel.
Further, use eight kinds of different regression algorithms respectively in specific subset in training set in step (2)
Sample predicted, be linear regression respectively, second order polynomial regression, three rank multinomials return, ridge regression, and LASSO is returned,
Elastic is returned, and GAM is returned and MARS is returned；
Wherein, the linear regression provides following formula (1):
E (y)=β0+β1x1+...+βdxd(1), E (y) represents predicted value in formula, and y, which is represented, assumes time delay of eating dishes without rice or wine,
x1..., xdIt represents and receives index；In the model, response variable y Gaussian distributed can be counted directly with least square method
It calculates and obtains corresponding fitting coefficient β0..., βd；
The second order polynomial regression algorithm first calculates primary, the Quadratic Orthogonal multinomial of each index set, to obtain
2D form carries out models fitting with 2D+1 parameters；
The three rank multinomials regression algorithm, in the selection of variable, the number of orthogonal polynomial is chosen from 1 time to 3 time,
To reduce the constraint of prediction model；
The ridge regression algorithm provides following formula (2):
Algorithm is by increasing penalty coefficient in formula, to factor beta0..., βdIt is limited
System, becomes contraction, to find the minimum variance of least squares estimator, wherein parameter t is according to E.Cule and M.De Iorio
The theory proposed based on control forecasting value variance is chosen automatically, and k represents k-th of coefficient, the range of k from 1,2,3 ...,
d；
The LASSO regression algorithm provides following formula (3):
Constraint function in formula limits the sum of absolute value of regression coefficient, removes
Constant coefficient, t value are chosen automatically；
The Elastic regression algorithm provides following formula (4):
Constraint function in formula is to the ridge regression algorithm
The linear combination of regularization is carried out with the constraint function in the LASSO regression algorithm, wherein α 1/2, t value are to choose automatically
's；
The GAM regression algorithm provides following formula (5):
G (E (y))=β0+f1(x1)+...+fd(xd) (5), g indicates generalized linear Copula, f in formula1..., fd
Indicate the non-linear relationship between input variable, β0For constant term, x1..., xdIt is to receive the data obtained in index from five；
The MARS regression algorithm provides following formula (6):
X → max (0, x-C) or x → max (0, C-x)；C ∈ R (6), in MARS recurrence, recurrence is synthesized to one
The linear combination of a hinge function.
Further, in the step (2), the calculating on training set is completed by successive Regression, by looking for
To suitable weight.
Further, in step (4), weight is inferred using column error obtained in step (3), utilizes weight optimization
The method of model gives following formula (7):
Wherein W1...W8By minimizing weighted least-squares method and limitation
Total weight calculation obtains, wherein εjIndicate the training error value with j-th of model on training set, wjIndicate its j-th of model
The weight of training error value on training set.
Further, during the step (4) infer weight, following formula (8) are given for constraint condition:
Wherein, wjIndicate training error value of its j-th of model on training set
Weight.
Further, in the step (5), by predicted value obtained in test set and known time delay of really eating dishes without rice or wine
Value is compared, and is given formula (9) and obtained bit error rate εtest:
Wherein, yiIndicate the sky of i-th of sample in test set
Mouth time delay true value,Indicate the latency prediction value of eating dishes without rice or wine of i-th of sample in test set, ntestIndicate test sample amount, εtest
It is then model mean absolute error on training set；
In order to confirm the stability and accuracy that integrate regressive prediction model obtained in the step (4), need to carry out
A variety of different comparisons obtain a bit error rate ε firstly, the identical bit error rate is executed on training settrainIf εtestWith
εtrainBetween difference it is smaller, then show over-fitting do not occur, it means that the stability of integrated regressive prediction model；Secondly,
The predicted value of every kind of regression algorithm is calculated on test set, it is made respectively to obtain a bit error rate, by by these bit error rates
The ε obtained with integrated recurrencetestIt is compared, then can check the accuracy of integrated regressive prediction model.
Fig. 1 is given embodiment and is shown using the operation workflow explanation of the method for integrated regression system detection LTE network performance
It is intended to.In more detail, it is the biography by being explained in detail below which, which integrates the purpose of the method for regression system detection LTE network performance,
System receives index prediction and eats dishes without rice or wine the approximation of time delay, but more critical eated dishes without rice or wine between time delay and conventional receiver index
It shows slight non-linear, has no direct correlation.Therefore, algorithm model cannot be simply by linear regression to receiving
Index is handled.The system (the integrated regressive prediction model in the present invention) is when predicting big data, by a variety of times
Reduction method is integrated in final weight estimation, can increase system robustness, and obtains accurate predicted value.
Legacy system can be broadly described by flow chart shown in FIG. 1.It completes to receive achievement data and net first
Network each stage eats dishes without rice or wine the collection of time delay.This sample data set is divided into training set and test set by model.Wherein 70% data are
Training set, 30% data be test set.
Secondly, training set is used to find to receive the projected relationship between index and time delay of eating dishes without rice or wine.In the operation of training set
Cheng Zhong predicts time delay of eating dishes without rice or wine using index is received by the model of foundation.Specifically, being exactly to use eight kinds of different recurrence
Algorithm respectively predicts the sample in training set in specific subset.(by staying a cross-validation method, next
It will be explained in paragraph).
The predicted value obtained by comparing model and time delay value of really eating dishes without rice or wine, to derive each sample in training set
Error amount and every kind of regression algorithm error.For regression algorithm J, the column error amount in training set is then referred to as " error
J”。
Later, take a kind of analysis method that regression algorithm is weighted processing (referred to as least square according to error 1~8
Method will be described in detail in next paragraph).Algorithm after being combined by weighted regression just constitutes our integrated recurrence
Prediction model.
Finally, integrated regression algorithm is applied to test set.It is closed with its detection based on the model prediction that training set is obtained
The accuracy of system, and then improve the forecasting accuracy of entire model.According to mathematical model proof rule, model result can instructed
Practice collection and verifying collection is accurately fitted, it is considered that the case where overfitting is not present, then the model is accurate.
The present embodiment is eated dishes without rice or wine in addition to description other than time delay, is further related to other five and is used to measure the quality of reception and MPS process
Reception index.There is certain positive and negative correlativity between time delay of eating dishes without rice or wine in them.This five indexs are specifically: with reference to letter
Number power (RSRP) is received, Reference Signal Received Quality (RSRQ), Signal to Interference plus Noise Ratio (SINR), physical uplink is shared
Physical Resource Block (PUSCH-PRB) on channel and the Physical Resource Block (PDSCH-PRB) on Physical Downlink Shared Channel.RSRP
It indicates on the frequency band of measurement, carries the linear average of the power of the resource unit of cell own reference signal, i.e., average letter
Number intensity.The power for receiving signal is measured by the average value of some reference signals, therefore by it may know that user is
It is no to be easily accessed the cell.It is the ratio of received signal strength and total bandwidth power that RSRQ is corresponding, wherein total bandwidth power packet
Include serving cell power, jamming power and noise.It is to receive letter by the level of quantizing noise and signal interference to assess
Number quality.SINR is the ratio between signal power and jamming power, noise summation.PUSCH-PRB (PDSCH-PRB) indicates to divide
The PRB quantity of dispensing up channel (down channel), user are an important references in the numerical value that local resource consumes.
Sample data is divided into training set and test set, when the training set based on sample data carries out regression training, needs elder generation
Understand eight kinds of different regression algorithms involved in this system.As described above, eight kinds of regression algorithms have used different systems
Meter method and statistic algorithm, it is ensured that the accuracy of integrated regression system.Eight kinds of regression algorithms are as follows: linear regression,
Second order polynomial regression, three rank multinomials return, and ridge regression, LASSO is returned, and Elastic is returned, and GAM is returned and MARS is returned.
In classical linear regression, it is assumed that time delay of eating dishes without rice or wine (hereinafter referred to as y) and receive index (referred to as x1..., xd) between
Relationship be linear.Formula 1 indicates relationship between the two, and wherein E (y) represents predicted value.In the model, response becomes
Y Gaussian distributed is measured, can directly be calculated with least square method and obtain corresponding fitting coefficient β0..., βd.Take this time
Reduction method, final predicted value will be influenced by all reception indexs.
E (y)=β0+β1x1+...+βdxd (1)
Second order polynomial regression algorithm is closer to classical linear regression.But it is not that direct select receives index
x1..., xd, but primary, the Quadratic Orthogonal multinomial of each index set are first calculated, to obtain 2D form.This has been done to change
After change, models fitting is carried out by 2D+1 (in the form of the 2D of index and constant term) item parameter, rather than the D+1 in classical linear model
Item parameter.This fit approach operation is upper as least square method before.
Three rank multinomial regression algorithms, model is similar to second order polynomial, but in the selection of variable, orthogonal polynomial
Number is chosen from 1 time to 3 time, to reduce the constraint of prediction model.
Ridge regression algorithm described to formula 1 it is similar, but unlike, the algorithm by increase penalty coefficient, to coefficient
β0..., βdIt is limited and (becomes and shrink).Precisely, which is limited by the quadratic sum to coefficient, to find
The minimum variance (being similar to classical linear regression) of least squares estimator, as shown in equation 2.Wherein parameter t is according to E.Cule
It is chosen automatically with the M.De Iorio theory proposed based on control forecasting value variance.
Some are similar with ridge regression for LASSO recurrence, but constraint function therein is different.The constraint function limit that LASSO is returned
The sum of absolute value of regression coefficient (removing constant coefficient) is made, as shown in formula 3.Theoretical analysis shows that this is more for coefficient
Strong constraint, or even certain coefficients exactly equal to zero can be generated.Therefore, which is contracted by using another method
Small coefficient, to realize the purpose that index set is simplified.In ridge regression, t value is chosen automatically.
Elastic recurrence is another by being compromised in ridge regression and LASSO return come the algorithm of coefficient of reduction.
Here constraint function is the linear combination of the constraint function progress regularization in returning to ridge regression and LASSO, such as 4 institute of formula
Show.Wherein, 1/2 α, t value are chosen automatically.
In GAM recurrence, eat dishes without rice or wine time delay value y and index input variable x1..., xdBetween relationship it is as shown in formula 5,
Middle hypothesis y comes from ED~* class.On the left side of formula, g indicates generalized linear Copula.F on the right of formula1..., fdTable
Show the non-linear relationship between input variable, β0For constant term.Function f1It can be considered as a kind of nonparametric backfitting algorithm.
The algorithm carries out approximate description by way of iteration, using cubic spline function (cube that at least there is certain index).?
Herein, y is indicated to eat dishes without rice or wine time delay and is assumed Gaussian distributed (belonging to exponential family), g=id, x1..., xdIt is to be connect from five
Receive the data obtained in index.
G (E (y)=β0+f1(x1)+...+fd(xd) (5)
In MARS recurrence, the linear combination for being synthesized to a hinge function is returned.Hinge function, such as institute in formula 6
Definition, be by the whole nonlinear model of itself non-linear release.As a whole, entire space is divided into each response
Polynomial subspace.In order to separate these spaces and execute recurrence operation, need to complete two steps.Firstly, passing through reduction
The error of residual sum of squares (RSS), forward calculation goes out new basic function and hinge function in an iterative manner.Secondly, again by deleting backward
Except minimum contribution item is modified model.Final step is then to reduce over-fitting.
X → max (0, x-C) or x → max (0, C-x)；C∈R (6)
Calculating on training set is completed by successive Regression.By finding suitable weight, so that every kind be returned
It is grouped into the integrated recurrence of synthesis.Next this process will be illustrated.
In each regression algorithm J, to element i each on training set using a cross-validation method is stayed, to be predicted
ValueIn order to obtain this predicted value, then all elements other than i are carried out with the operation of regression algorithm j.So obtain one
It is a to be obtained by element iAnticipation function.When eating dishes without rice or wine, Yanzhong all elements are all known, variancesIt then can be by being calculated.Therefore, for each regression algorithm J, column vector error is then defined as: εj:
=(εij)'i(wherein ' represent transposition).
Next step is to infer weight using these column errors.w1...w8By minimize weighted least-squares method and
Total weight calculation is limited to obtain.Wherein, formula 7 defines the method using weight optimization model, and formula 8 defines constraint item
Part.Here | | | |2Represent Euclid norm.
After obtaining weight, it is still necessary to understand regression function cited in every kind of prediction and every kind of regression algorithm.For reality
Existing this point, all elements on training set is all executed by regression algorithm j, to obtain anticipation function Pj.Finally, integrated
Recurrence is then defined as PjLinear weighted combination.
It obtains, then checks its result in test set once integrated recurrence is derived.In checking process, it will test
Predicted value obtained is concentrated to be compared with known time delay value of really eating dishes without rice or wine.By average error rate defined in formula 9
Bit error rate ε is calculatedtest.In order to confirm the stability and accuracy of integrated regression algorithm, need to carry out a variety of different ratios
Compared with.Firstly, the identical bit error rate is executed on training set, a bit error rate ε is obtainedtrain.If εtestAnd εtrainBetween difference
It is smaller, then show over-fitting do not occur, it means that the stability of algorithm.It is returned secondly, every kind is calculated on test set
The predicted value of reduction method makes it respectively obtain a bit error rate.Pass through the ε for obtaining these bit error rates and integrated recurrencetestIt carries out
Compare, then can check the accuracy of algorithm.
The sample that Fig. 2 gives analogue data is calculated associated weight with analytic approach and pushes away as procedure described above
Integrated regression algorithm is exported, average error rate is calculated in training set and test set.In this example, training set and test set
Error increment it is lower, show that overfitting does not occur in model.Thus, it can be known that integrated regression algorithm is calculated relative to other recurrence
For method, the result obtained is more accurate, and practicability is stronger.In addition, relative to other all algorithm (even three rank multinomials
Regression algorithm), it is also stronger to integrate the stability returned.Therefore, preferred regression algorithm is exactly integrated returns here.
In general, it can effectively be derived by integrated regression system using cell RF index and predict time delay of eating dishes without rice or wine.It is integrated
Regression model is not only stable but also can accurately predict time delay of eating dishes without rice or wine, it can carry out calculating the selection with model, therefore mould automatically
Type has stronger adaptivity.It is proved by example, the theory of the assessment models is accurate.For the angle of user,
This system is easier to obtain and analyzes index of correlation data, thus the prediction result for the time delay that obtains eating dishes without rice or wine.
Claims (7)
1. a kind of method using integrated regression system detection LTE network performance, comprising the following steps:
(1) collection shape the collection of sample data set: is carried out to reception achievement data and network each stage of LTE network time delay of eating dishes without rice or wine
It is divided into training set and test set at sample data set, and by the sample data set；
Wherein, training set is used to find to receive the projected relationship between index and time delay of eating dishes without rice or wine；Test set is for detecting based on instruction
Practice the accuracy for the model prediction relationship for collecting obtained；
(2) in the operating process of the training set, by the model of foundation, time delay of eating dishes without rice or wine is predicted using index is received, is used
Regression algorithm respectively predicts the sample in training set in specific subset；
(3) predicted value obtained by comparing model in above-mentioned steps (2) and time delay value of really eating dishes without rice or wine, are derived in training set
The error amount of each sample and the error of regression algorithm；
Wherein, for regression algorithm J, the column error amount in training set is then referred to as error J；
(4) error in regression algorithm is weighted by processing using analysis method, is made up of and integrates back weighted regression combination
Return prediction model；
(5) the integrated regression algorithm in step (4) is applied into test set, detects the model prediction obtained based on training set and closes
The accuracy of system.
2. the method as described in claim 1 using integrated regression system detection LTE network performance, which is characterized in that described
The reception index that achievement data is received in step (1) includes Reference Signal Received Power RSRP, Reference Signal Received Quality RSRQ,
Signal to Interference plus Noise Ratio SINR, the Physical Resource Block PUSCH-PRB and physical down on Physical Uplink Shared Channel are shared
Physical Resource Block PDSCH-PRB on channel.
3. the method as claimed in claim 2 using integrated regression system detection LTE network performance, which is characterized in that in step
Suddenly the sample in training set is predicted in specific subset respectively using eight kinds of different regression algorithms in (2), is respectively
Linear regression, second order polynomial regression, three rank multinomials return, ridge regression, LASSO return, Elastic return, GAM return and
MARS is returned；
Wherein, the linear regression provides following formula (1):
E (y)=β0+β1x1+...+βdxd(1), E (y) represents predicted value in formula, and y, which is represented, assumes time delay of eating dishes without rice or wine, x1..., xdGeneration
Table receives index；In the model, it is corresponding can directly to calculate acquisition with least square method for response variable y Gaussian distributed
Fitting coefficient β0..., βd；
The second order polynomial regression algorithm first calculates primary, the Quadratic Orthogonal multinomial of each index set, to obtain 2D shape
Formula carries out models fitting with 2D+1 parameters；
The three rank multinomials regression algorithm, in the selection of variable, the number of orthogonal polynomial is chosen from 1 time to 3 time, thus
Reduce the constraint of prediction model；
The ridge regression algorithm provides following formula (2):
Algorithm is by increasing penalty coefficient in formula, to factor beta0..., βdIt is limited, is become
It shrinks, to find the minimum variance of least squares estimator, wherein parameter t is based on control according to E.Cule and M.De Iorio
The theory that predicted value variance is proposed is chosen automatically, and k represents k-th of coefficient, and the range of k is from 1,2,3 ..., d；
The LASSO regression algorithm provides following formula (3):
Constraint function in formula limits the sum of absolute value of regression coefficient, removes often system
Number, t value are chosen automatically；
The Elastic regression algorithm provides following formula (4):
Constraint function in formula is to the ridge regression algorithm and institute
State the linear combination that the constraint function in LASSO regression algorithm carries out regularization, wherein α 1/2, t value are chosen automatically；
The GAM regression algorithm provides following formula (5):
G (E (y))=β0+f1(x1)+...+fd(xd) (5), g indicates generalized linear Copula, f in formula1..., fdIndicate input
Non-linear relationship between variable, β0For constant term, x1..., xdIt is to receive the data obtained in index from five；
The MARS regression algorithm provides following formula (6):
X → max (0, x-C) or x → max (0, C-x)；C ∈ R (6), in MARS recurrence, recurrence is synthesized to a hinge
The linear combination of function.
4. the method as claimed in claim 3 using integrated regression system detection LTE network performance, which is characterized in that in institute
It states in step (2), the calculating on training set is completed by successive Regression, by finding suitable weight.
5. the method as claimed in claim 4 using integrated regression system detection LTE network performance, which is characterized in that in step
Suddenly in (4), weight is inferred using column error obtained in step (3), is given using the method for weight optimization model following
Formula (7):
Wherein W1...W8By minimizing weighted least-squares method and limiting total weight
It is calculated, wherein εjIndicate the training error value with j-th of model on training set, wjIndicate its j-th of model in training
The weight of training error value on collection.
6. the method as claimed in claim 5 using integrated regression system detection LTE network performance, which is characterized in that in institute
During stating step (4) deduction weight, following formula (8) are given for constraint condition:Wherein, wjIndicate the weight of training error value of its j-th of model on training set.
7. the method as claimed in claim 5 using integrated regression system detection LTE network performance, which is characterized in that in institute
It states in step (5), predicted value obtained in test set is compared with known time delay value of really eating dishes without rice or wine, and give public affairs
Formula (9) obtains bit error rate εtest:
Wherein, yiIndicate when the eating dishes without rice or wine of i-th sample in test set
Prolong true value,Indicate the latency prediction value of eating dishes without rice or wine of i-th of sample in test set, ntestIndicate test sample amount, εtestIt is then
Model mean absolute error on training set；
In order to confirm the stability and accuracy that integrate regressive prediction model obtained in the step (4), need to carry out various
Different comparisons obtains a bit error rate ε firstly, the identical bit error rate is executed on training settrainIf εtestAnd εtrain
Between difference it is smaller, then show over-fitting do not occur, it means that the stability of integrated regressive prediction model；Secondly, surveying
The predicted value of every kind of regression algorithm is calculated on examination collection, it is made respectively to obtain a bit error rate, by by these bit error rates and collection
The ε obtained at recurrencetestIt is compared, then can check the accuracy of integrated regressive prediction model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510947246.4A CN105517019B (en) | 2015-12-17 | 2015-12-17 | Using the method for integrated regression system detection LTE network performance |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510947246.4A CN105517019B (en) | 2015-12-17 | 2015-12-17 | Using the method for integrated regression system detection LTE network performance |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105517019A CN105517019A (en) | 2016-04-20 |
| CN105517019B true CN105517019B (en) | 2019-03-08 |
Family
ID=55724582
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510947246.4A Active CN105517019B (en) | 2015-12-17 | 2015-12-17 | Using the method for integrated regression system detection LTE network performance |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105517019B (en) |
Families Citing this family (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106604290B (en) * | 2016-12-19 | 2020-02-14 | 南京华苏科技有限公司 | User perception evaluation wireless network performance method based on web browsing |
| EP3503610B1 (en) * | 2017-12-22 | 2022-03-23 | AirTies Belgium SPRL | A method and device for identifying at least a wireless link causing interferences on other wireless links |
| CN108803348B (en) * | 2018-08-03 | 2021-07-13 | 北京深度奇点科技有限公司 | PID parameter optimization method and PID parameter optimization device |
| CN109284320B (en) * | 2018-08-15 | 2021-10-26 | 上海派拉软件股份有限公司 | Automatic regression diagnosis method on big data platform |
| CN109948262A (en) * | 2019-03-22 | 2019-06-28 | 清华大学 | A kind of semiconductor devices modeling method and system using rational fraction regression model |
| CN110969370B (en) * | 2020-01-14 | 2023-05-02 | 深圳市建筑科学研究院股份有限公司 | Quality risk analysis method for building structural member |
| CN111338304A (en) * | 2020-03-02 | 2020-06-26 | 顺忠宝智能科技(深圳)有限公司 | Method for real-time prediction and information communication of production line yield by applying artificial intelligence cloud computing |
| CN112131706A (en) * | 2020-08-21 | 2020-12-25 | 上海大学 | Method for rapidly predicting melting point of low-melting-point alloy through ridge regression |
| CN113382477B (en) * | 2021-05-14 | 2022-09-27 | 北京邮电大学 | Method for modeling uplink interference between wireless network users |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101267362A (en) * | 2008-05-16 | 2008-09-17 | 亿阳信通股份有限公司 | A dynamic identification method and its device for normal fluctuation range of performance normal value |
| CN102904603A (en) * | 2011-07-28 | 2013-01-30 | 中兴通讯股份有限公司 | Searching window processing method and device |
| WO2013056435A1 (en) * | 2011-10-19 | 2013-04-25 | Telefonaktiebolaget L M Ericsson (Publ) | Method and apparatus for channel predicting |
| CN103093095A (en) * | 2013-01-14 | 2013-05-08 | 湖州师范学院 | Software failure time forecasting method based on kernel principle component regression algorithm |
| CN103514369A (en) * | 2013-09-18 | 2014-01-15 | 上海交通大学 | Regression analysis system and method based on active learning |
| CN105050125A (en) * | 2015-06-23 | 2015-11-11 | 武汉虹信通信技术有限责任公司 | Method and device for evaluating mobile data service quality oriented to user experience |
-
2015
- 2015-12-17 CN CN201510947246.4A patent/CN105517019B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101267362A (en) * | 2008-05-16 | 2008-09-17 | 亿阳信通股份有限公司 | A dynamic identification method and its device for normal fluctuation range of performance normal value |
| CN102904603A (en) * | 2011-07-28 | 2013-01-30 | 中兴通讯股份有限公司 | Searching window processing method and device |
| WO2013056435A1 (en) * | 2011-10-19 | 2013-04-25 | Telefonaktiebolaget L M Ericsson (Publ) | Method and apparatus for channel predicting |
| CN103093095A (en) * | 2013-01-14 | 2013-05-08 | 湖州师范学院 | Software failure time forecasting method based on kernel principle component regression algorithm |
| CN103514369A (en) * | 2013-09-18 | 2014-01-15 | 上海交通大学 | Regression analysis system and method based on active learning |
| CN105050125A (en) * | 2015-06-23 | 2015-11-11 | 武汉虹信通信技术有限责任公司 | Method and device for evaluating mobile data service quality oriented to user experience |
Non-Patent Citations (1)
| Title |
|---|
| 基于数据挖掘的移动网络优化与运营技术研究;闵佳;《中国优秀硕士学位论文全文数据库》;20150815;全文 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105517019A (en) | 2016-04-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105517019B (en) | Using the method for integrated regression system detection LTE network performance | |
| Hsieh et al. | Deep learning-based indoor localization using received signal strength and channel state information | |
| KR102588572B1 (en) | Motion detection based on machine learning of wireless signal characteristics | |
| CN103874118B (en) | Radio Map bearing calibrations in WiFi indoor positionings based on Bayesian regression | |
| Bisio et al. | Smart probabilistic fingerprinting for WiFi-based indoor positioning with mobile devices | |
| CN108293238A (en) | Promote the indoor positioning under access point signals change and fingerprint update | |
| CN108051779B (en) | TDOA-oriented positioning node optimization method | |
| US11681839B2 (en) | Systems and methods for modeling water quality | |
| CN109490826B (en) | Ranging and position positioning method based on radio wave field intensity RSSI | |
| Li et al. | Unsupervised learning of indoor localization based on received signal strength | |
| Miyagusuku et al. | Improving gaussian processes based mapping of wireless signals using path loss models | |
| Chaudhari et al. | Spatial interpolation of cyclostationary test statistics in cognitive radio networks: Methods and field measurements | |
| CN111698695A (en) | LTE fingerprint type positioning method based on neural network | |
| Li et al. | Adaptive multi‐task compressive sensing for localisation in wireless local area networks | |
| Tan et al. | An efficient fingerprint database construction approach based on matrix completion for indoor localization | |
| CN104821854B (en) | A kind of many primary user's multidimensional frequency spectrum sensing methods based on random set | |
| US11128391B1 (en) | System and method for predicting wireless channel path loss | |
| CN112911502B (en) | Base station longitude and latitude estimation method based on MDT technology and RSRP ranging | |
| Zhou et al. | A Robust Detection and Optimization Approach for Delayed Measurements in UWB Particle-Filter-Based Indoor Positioning | |
| CN107613466B (en) | Indoor positioning method based on fingerprint similarity under ultra-dense wireless network | |
| US20230362039A1 (en) | Neural network-based channel estimation method and communication apparatus | |
| CN105187139A (en) | Outdoor wireless received signal strength (RSS) map building method based on crowd sensing | |
| Wang et al. | A novel indoor ranging method using weighted altofrequent RSSI measurements | |
| Yayeh Munaye et al. | Hybrid deep learning‐based throughput analysis for UAV‐assisted cellular networks | |
| Naveed et al. | Smart indoor Positioning Model for Deterministic Environment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB02 | Change of applicant information | ||
| CB02 | Change of applicant information |
Address after: 210000 No. 6 Longjing Road, Chunxi Town, Gaochun District, Nanjing City, Jiangsu Province Applicant after: Nanjing Hua Su Science and Technology Ltd. Address before: 210000 No. 6 Longjing Road, Chunxi Town, Gaochun District, Nanjing City, Jiangsu Province Applicant before: Nanjing Hua Su Science and Technology Co., Ltd. |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |