CN109308518A - A kind of monitoring system and its smoothing parameter optimization method based on probabilistic neural network - Google Patents
A kind of monitoring system and its smoothing parameter optimization method based on probabilistic neural network Download PDFInfo
- Publication number
- CN109308518A CN109308518A CN201811065716.4A CN201811065716A CN109308518A CN 109308518 A CN109308518 A CN 109308518A CN 201811065716 A CN201811065716 A CN 201811065716A CN 109308518 A CN109308518 A CN 109308518A
- Authority
- CN
- China
- Prior art keywords
- axis
- layer
- particle
- layer module
- submodule
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S19/00—Satellite radio beacon positioning systems; Determining position, velocity or attitude using signals transmitted by such systems
- G01S19/01—Satellite radio beacon positioning systems transmitting time-stamped messages, e.g. GPS [Global Positioning System], GLONASS [Global Orbiting Navigation Satellite System] or GALILEO
- G01S19/13—Receivers
- G01S19/23—Testing, monitoring, correcting or calibrating of receiver elements
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Computer Networks & Wireless Communication (AREA)
- Position Fixing By Use Of Radio Waves (AREA)
Abstract
The present invention relates to a kind of monitoring system and its smoothing parameter optimization method based on probabilistic neural network, belong to artificial intelligence and technical field of satellite navigation.The monitoring system includes input layer, mode layer, layer of summing, compares layer and output layer module；The parameter optimization method includes: 1) initiation parameter；2) particle rapidity and position are initialized；3) particle rapidity and position are updated；4) particle fitness is calculated；5) the history optimal value of more new particle；6) global optimum of more new particle；7) judge whether to meet monitoring system requirements, if really skipping to 10)；Otherwise it skips to 8)；8) judge whether all particles are disposed, if skipping to 9), otherwise skip to 4)；9) judge whether 3) reach maximum number of iterations otherwise skips to if skipping to 10)；10) fitness function of particle is exported.Vertical direction minimum detectable failure of the present invention can be controlled in 35m；Omission factor and false alarm rate can be effectively controlled.
Description
Technical field
The present invention relates to a kind of monitoring system and its smoothing parameter optimization method based on probabilistic neural network belong to artificial
Intelligence and technical field of satellite navigation.
Background technique
When being positioned using satellite navigation system, it is necessary to which assessment receives the integrity of data, to ensure positioning result
Reliability.Receiver autonomous integrity monitoring (RAIM) is a kind of satellite integrity inspection method, is independently executed by receiver,
Real-time guard is provided for user.When satellite navigation system breaks down or when can not provide necessary positioning accuracy, RAIM meeting and
When remind user, so as to avoid an accident.The pseudorange error according to caused by satellite failure, which may cause, positions wrong feature, is based on
Traditional RAIM of hypothesis testing detects satellite failure using the noise and false alarm probability of measurement, wherein with least-square residuals
Method, odd_even adjudgement rule and multi-risk System assume that disaggregation separation (MHSS) method is most representative.But the monitoring of these three methods is smart
It spends low, can only be used in ocean section (position error be less than 200m).
From 2012 to 2016 year, the research being made of Federal Aviation management board (FAA) and European Space Agency (ESA)
The advanced RAIM algorithm of group development, the algorithm are possible to meet LPV-200 that (aircraft is at 200 feet away from ground of vertical direction
The navigation performance of height) rank navigation needs.But this method in the case where more constellation alignment by union, can only meet
The navigation performance demand of LPV-200, and the availability in its global range is also in the stage of discussion.
Recent years, the RAIM algorithm based on Bayes's detection theory experienced fast development.Pesonen proposes one kind
Using the RAIM method of bayesian theory detection single fault satellite, and structural framing is developed, by satellite navigation integrity problem
It is converted into the Posterior probability distribution of positional parameter.Zhang Qianqian proposes a kind of more satellites event based on Bayes Hypothesis Test frame
The RAIM method for hindering detection determines satellite failure by calculating the probability of dangerous misleading information.Regrettably, these algorithms need
A large amount of computing resource is wanted, and some hyper parameters in sampling model must be exported from historical information, work as amount of history information
When insufficient, the uncertainty of hyper parameter will affect the quality of satellite integrity monitoring result.
Although above-mentioned existing RAIM technology has the effect of satellite integrity monitoring, however, navigating in precision approach
(LPV-200) in the case of, there is also much rooms for the improvement of traditional RAIM method performance.The purpose of the present invention is be dedicated to solving
The technological deficiency of traditional RAIM proposes that one kind can be applied in precision approach navigation (LPV-200) in conjunction with artificial intelligence technology
Receiver autonomous integrity monitoring method.
Summary of the invention
It is an object of the invention to solve in precision approach navigation (LPV-200), receiver can not be in GPS single system solely
Vertical the problem of completing the analysis of satellite integrity, proposes a kind of monitoring system based on probabilistic neural network and its smoothing parameter is excellent
Change method.
A kind of monitoring system and its smoothing parameter optimization method based on probabilistic neural network includes probabilistic neural network prison
Examining system and a kind of smoothing parameter optimization method based on particle group optimizing, the method are to improve the monitoring systematicness
It can be to meet the smoothing parameter optimization methods of LPV-200 navigation needs；
Wherein, probabilistic neural network monitoring system includes input layer module, mode layer module, summation layer module, compares layer
Module and output layer module；
Wherein, in addition to output layer module, input layer module, mode layer module, summation layer module, compare layer module all and be it is single
To transmission；
Wherein, input layer module includes input layer X-axis submodule, input layer Y-axis submodule and input layer Z axis submodule；
Mode layer module includes mode layer X-axis submodule, mode layer Y-axis submodule and mode layer Z axis submodule；Summation layer module include
Summation layer X-axis submodule, summation layer Y-axis submodule and summation layer Z axis submodule.
The connection relationship of each module is as follows in the monitoring system:
Input layer module is connect with mode layer module；Mode layer module is connect with summation layer module；Sum layer module with than
It is connected compared with layer module；Compare layer module to connect with output layer module.
The function of each module is in the monitoring system:
The function of input layer module is the standard deviation for calculating one group of position data of receiver collection；Mode layer mould
The function of block is the value for calculating activation primitive；Summation layer functions of modules is to calculate input data to belong to the average general of failure classes
Rate and the average probability for belonging to fault-free class；The function of comparing layer module is carried out to the probability of malfunction from summation three axis of layer
Compare calculating；The function of output layer module is will to compare X-axis in layer, Y-axis and three axis output valve of Z axis to do Boolean calculation, and provide
Final detection result；.
Wherein, shown in input layer module calculation formula following (1), (2) and (3):
Wherein, xi, yiAnd ziRespectively indicate i-th of input quantity of X-axis, Y-axis and Z axis, X2It indicates to mode layer X-axis submodule
The data of block input, Y2Indicate the data inputted to mode layer Y-axis submodule, Z2Indicate the number inputted to mode layer Z axis submodule
According to n indicates the quantity of neuron in submodule, and the neuronal quantity of X-axis, Y-axis and Z axis submodule is equal；
The data of summation layer output are expressed as follows shown in (4), (5), (6), (7), (8) and (9) with Gaussian function:
Wherein, X31、Y31、Z31Respectively indicate the number that fault-free node exports into summation layer X-axis, Y-axis, Z axis submodule
According to X32、Y32、Z32Indicate the data that malfunctioning node exports into summation layer X-axis, Y-axis, Z axis submodule, LXi、LYiAnd LZiIt represents
I-th of training sample of non-failure conditions, LXFi、LYFiAnd LZFiRepresent i-th of training sample of faulty situation, s indicate without
Fault sample quantity, w indicate fault sample quantity, LXi、LYi、LZi、LXFi、LYFiAnd LZFiCan by following formula (10), (11),
(12), (13), (14) and (15) are calculated；
σ in formulax、σyAnd σzThe standard deviation of X-axis, Y-axis and the distribution of Z axis Satellite position error is respectively indicated, k indicates expansion
Coefficient, d indicate sample sample size, CmIndicate correction factor, the neuronal quantity in mode layer is equal to the sum of training sample
Amount；
The operation of summation layer module can be calculated by following formula (16), (17), (18), (19), (20) and (21):
fx1=X31/s (16)
fx2=X32/w (17)
fy1=Y31/s (18)
fy2=Y32/w (19)
fz1=Z31/s (20)
fz2=Z32/w (21)
Wherein, fx1、fy1And fz1Respectively indicate X-axis, Y-axis and the trouble-free probability of Z axis locator value, fx2、fy2And fz2Respectively
Indicate X-axis, Y-axis and the faulty probability of Z axis locator value, the neuronal quantity in layer of summing is 6；
The operation for comparing layer module is realized by formula (22), (23) and (22):
X in formula4, Y4, Z4The data inputted to output layer are respectively indicated, the quantity for comparing layer neuron is 3；
The operation of output layer module is will to compare X-axis in layer, Y-axis and three axis output valve of Z axis to do Boolean calculation, operation method
As shown in following formula (25):
R=X4∧Y4∧Z4 (25)
Wherein, R indicates output as a result, R=1 indicates that satellite used for positioning includes failure, and R=0 indicates satellite positioning
Normally, there was only 1 neuron in output layer；
The Monitoring Performance of monitoring system is mainly influenced by following two parts:
1. training sample LXi、LYi、LXFi、LYFiAnd LZFiThe quality of data；
2. smoothing parameter λ；
The monitoring system in the modeling process of above layers, training sample using formula (10), (11), (12),
(13), (14) and (15) are calculated, and smoothing parameter λ carries out optimizing using particle group optimizing method, i.e., a kind of to be based on population
The smoothing parameter optimization method of optimization, comprising the following steps:
Step 1, initiation parameter；
Wherein, initiation parameter specifically includes given inertial parameter ω, constant C1And C2, random number R1And R2Initial value, give
Determine iteration sum ttotalWith the numerical value of total number of particles p；
Step 2, initialization particle rapidity and position；Specifically: in given range, the first of each particle is set at random
Beginning speed v and position P, and t=1 is set by number of iterations；
Step 3 updates particle rapidity and position, specifically includes following sub-step:
Population i is set i=1 by step 3.1；
Step 3.2 calculates the speed and positional value of particle i in the case of the t times iteration using following formula (26) and (27),
vi,t=ω × vi,t-1+C1×R1×(Pbesti-Pi,t-1)+C2×R2×(Gbest-Pi,t-1) (26)
Pi,t=Pi,t-1+vi,t (27)
.v in formulai,tIndicate particle rapidity of i-th of particle in the t times iteration, Pi,tIndicate i-th of particle at the t times
Particle position when iteration, ω are inertial factor, C1And C2It is constant, R1And R2It is the random number in given section, PbestiIt is
The history optimal value of i-th of particle, Gbest are the global optimums of all particles；
Step 3.3i=i+1, if i≤p jump procedure 3.2, otherwise, and i=1, jump procedure 4；
Step 4 calculates particle fitness, and the fitness F (P of i-th of particle is specifically calculated using following formula (28)i,t),
Wherein, Ffa(Pi) and Fmd(Pi) it is piecewise function:
F(Pi,t) indicate particle Pi,tFitness function, Qfa(Pi,t) indicate with particle Pi,tProbability as smoothing parameter
Neural network monitors system under non-failure conditions, and the number of false alarm, Q occurs in detection resultmd(Pi,t) it is that 35m pseudorange error goes out
In the case where 6 seconds existing, there is the number of missing inspection in fault-finding result；
The history optimal value of step 5, more new particle；By F (Pi,t) and F (Pbesti) compare, if F (Pi,t) > F
(Pbesti), perform the following operations F (Pbesti)=F (Pi,t), and go to step 6；Otherwise, 8 are gone to step；
The global optimum of step 6, more new particle, by F (Pi,t) compared with F (Gbest), if F (Pi,t) > F
(Gbest), F (Gbest)=F (P is performed the following operationsi,t), go to step 7；Otherwise, 8 are gone to step；
Whether the particle that step 7, judgement search meets system requirements；F (Gbest)=1 is judged, if it is really jumping to
Step 10；If it is vacation, 8 are gone to step；
Step 8, more new particle carry out the operation of next particle, specifically: judge i≤p, if it is true, i=i+1, and
Go to step 4；If it is vacation, 9 are gone to step；
Step 9 updates iteration, carries out next iteration operation, specifically: judge t≤ttotal, if it is true, t=t+1,
And go to step 3；If it is vacation, 10 are gone to step；
Step 10 terminates operation, exports Pi,t。
Beneficial effect
A kind of monitoring system and its smoothing parameter optimization method based on probabilistic neural network, and it is existing receiver-autonomous complete
Good property monitoring method is compared, and is had the following beneficial effects:
1. vertical direction minimum detectable failure can be controlled in 35m (one of navigation needs of LPV-200)；
2. by using omission factor and false alarm rate that proposed method can be effectively controlled based on PSO smoothing parameter optimization method
(omission factor < 4 × 10-6, false alarm rate < 10-7, one of navigation needs of LPV-200)；Only ARAIM (advanced at present
Receiver autonomous integrity monitoring) method is expected to assign in the localizing environment of Galileo+GPS
To the demand.
Detailed description of the invention
Fig. 1 is a kind of structural schematic diagram of monitoring system based on probabilistic neural network；
Fig. 2 is parameter in a kind of monitoring system and its smoothing parameter optimization method and embodiment based on probabilistic neural network
σ with epoch variation diagram；
Fig. 3 be a kind of monitoring system based on probabilistic neural network and its smoothing parameter optimization method embodiment Satellite therefore
Hinder testing result figure；
Wherein, each node on behalf neuron in Fig. 1.
Specific embodiment
The present invention will be further described with reference to the accompanying drawings and examples and detailed description.
Embodiment 1
The present embodiment illustrates the structural representation that Fig. 1 of the present invention is a kind of monitoring system based on probabilistic neural network
Figure.
This example describes the method and steps when being embodied under using GPS single system positioning scenarios.
Step 1 utilizes receiver broadcast ephemeris and pseudo-range information, and data used in the present embodiment are collected in 2015 11
The moon 10 included the ephemeris and received pseudo-range information of GPS system broadcast in data；
Step 2 calculates every visible satellite measurement according to the elevation angle of ephemeris computation visible satellite, and according to elevation angle
The error of pseudorange is distributed, and in this example, the pseudorange error distribution of GPS system calculates the mathematical model of reference table 1；
The pseudorange error model of 1 GPS system of table
*: the elevation angle of α expression visible satellite
Step 3 calculates separately the standard deviation sigma of tri- axis positioning accuracy error of x, y and z according to Convolution Formulax, σyAnd σz, such as scheme
Shown in 2, Fig. 2 describes the changing value of σ in 5000 epoch, and the longitudinal axis indicates σ value, and horizontal axis indicates epoch；
Step 4 monitors system using probabilistic neural network involved in patent, constructs the reception based on probabilistic neural network
Machine autonomous integrity monitoring system；
Neuronal quantity in step 4.1 input layer module for X-axis fault diagnosis is set as 12, uses visible satellite
After starting positioning, continuously reads the locator value in 12 epoch in X-direction and be input in input layer, utilize formula
The input data X of mode layer is calculated2, the modeling method of Y-axis and Z axis input layer is identical as X-axis；
Neuronal quantity in step 4.2 mode layer module for X-axis fault diagnosis is set as 20, wherein 10 nerves
Member is for calculating X31, other neurons are for calculating X32, utilize formula
With
The input data X of summation layer is calculated31And X32, wherein parameter LXiAnd LXFiBy formula
With
It acquiring, the initial value of parameter can refer to table 2,
The initial value of 2 mode layer parameter of table
| s | 10 |
| w | 20 |
| d | 30 |
| Cm | 1 |
| k | 9 |
The final value of smoothing parameter λ will be used and be obtained based on PSO smoothing parameter optimization method, at this time set its initial value
It is 0.1, the mode layer modeling method of Y-axis and Z axis is identical as X-axis；
Neuronal quantity in step 4.3 summation layer module for X-axis fault diagnosis is set as 2, utilizes formula
fx1=x31/ s and fx2=x32/w
Calculating parameter fx1And fx2；
The neuronal quantity that step 4.4 compares layer module is set as 3, compares layer for the data f in two neuronsx1With
fx2It is compared, works as fx1> fx2, summation layer output 0；fx1≤fx2, summation layer output 1, Y-axis and Z axis summation layer modeling method with
X-axis is identical；
The neuronal quantity of step 4.5 output layer module is set as 1, calculates X-axis, Y-axis and Z axis using Boolean calculation and exists
The data that layer of summing exports, and result is exported, when output is 1, expression is current, and with the satellite positioned, there are failures, conversely,
Fault-free；
Step 5 is carried out excellent using smoothing parameter in the monitoring system established based on POS smoothing parameter optimization method to step 4
Change, optimization aim is to meet the needs of appropriate level navigation is to monitoring method false alarm rate and omission factor, and wherein the initial value of parameter is such as
Shown in table 3, below by taking X-axis as an example, the optimal enforcement example of smoothing parameter is introduced, the optimization method of Y-axis and Z axis is identical as X-axis；
The parameter value of 3 smoothing parameter optimization method of table
| Parameter type | Preset value |
| ω | 1 |
| C1 | 2 |
| C2 | 2 |
| R1 | (0,1] random number in section |
| R2 | (0,1] random number in section |
Step 5.1 brings 3 parameter of table into corresponding equation；
Step 5.2 total number of particles p is set as 100, in the section [0.01-0.3], is randomly provided just for each particle
Beginning position, the initial velocity of all particles are set as 0.001, and iteration sum t is arrangedtotal=1000；
Step 5.3 updates particle rapidity and position；Wherein PbestiIt is set as 0.1 with the initial value of Gbest, other parameters take
Value is as shown in table 3；
Population i is set i=1 by step 5.3.1；
Step 5.3.2 generates R at random1And R2Numerical value, utilize formula (23) and (24) to calculate grain in the case of the t times iteration
The speed and positional value of sub- i；
Step 5.3.3i=i+1, if i≤p jump procedure 5.3.2, otherwise, and i=1, jump procedure 5.4；
Step 5.4 calculates particle fitness；
1. parameter QfaCalculation method, the P that step 5.3 is calculatedi,tBring the probability of step 4 building into as smoothing parameter λ
Neural network monitors system, at this point, being distributed respectively from following three
dx~N (0, σx)
dy~N (0, σy)
dz~N (0, σz)
3 groups of samples of middle extraction, 12 data of every group of sample are input to probabilistic neural network monitoring system as input data
In detected, and record output as a result, repeat this operation 3 × 106It is secondary, QfaFor recording the number that detection result is failure；
2. parameter QmdCalculation method, the P that step 5.3 is calculatedi,tBring the probability of step 4 building into as smoothing parameter λ
Neural network monitors system, at this point, obeying following three distributions:
dx~N (0, σx)
dy~N (0, σy)
dz~N (0, σz)
It is middle to extract X-axis respectively, 3 groups of samples corresponding to Y-axis and Z axis, 12 data of every group of sample, in rear 6 data,
35m position error is added, is input in probabilistic neural network monitoring system and is detected as input data, and records output knot
Fruit repeats this operation 1 × 107It is secondary, QmdIt is trouble-free number for recording detection result；
Obtain parameter QfaAnd QmdAfter value, the fitness of particle is calculated using following formula,
The history optimal value of step 5.5 more new particle, by F (Pi,t) and F (Pbesti) compare, if F (Pi,t) > F
(Pbesti), perform the following operations F (Pbesti)=F (Pi,t), and go to step 5.6；Otherwise, 5.8 are gone to step；
The global optimum of step 5.6 more new particle, by F (Pi,t) compared with F (Gbest), if F (Pi,t) > F
(Gbest), F (Gbest)=F (P is performed the following operationsi,t), and go to step 5.7；Otherwise, 5.8 are gone to step；
Whether the particle that searches of step 5.7 judgement meets system requirements, judges F (Gbest)=1, if it is really jumping
To step 5.10；If it is vacation, 5.8 are gone to step；
Step 5.8 judges i≤p, if it is true, i=i+1, and gos to step 5.4；If it is vacation, go to step
5.9；
Step 5.9 judges t≤ttotal, if it is true, t=t+1, and go to step 5.3, if it is vacation, jump to
Step 5.10；
Step 5.10 exports λ=Pi,t, complete the building of monitoring system；
Three fault messages are added in step 6 in the pseudorange of satellite, and the fault message of addition is as shown in table 2；
2 fault message table of table
| Visible satellite asterisk | G1 | G17 | G30 |
| The fault value of addition | 50m | 40m | 30m |
| Epoch | 100 | 500 | 1000 |
| Trouble duration | 6s | 6s | 6s |
Step 7 Fig. 3 describe fault-finding as a result, the longitudinal axis indicate detection system output valve, wherein 0 indicate fault-free, 1
Indicate faulty；Horizontal axis indicates epoch, it is clear that the epoch-making moment of failure is added from Fig. 3, monitoring system can be quasi-
Really detect that there are failures for satellite used for positioning.
The above is presently preferred embodiments of the present invention, and it is public that the present invention should not be limited to embodiment and attached drawing institute
The content opened.It is all not depart from the lower equivalent or modification completed of spirit disclosed in this invention, both fall within the model that the present invention protects
It encloses.
Claims (2)
1. probabilistic neural network monitors system, it is characterised in that: including input layer module, mode layer module, summation layer module, ratio
Compared with layer module and output layer module；
Wherein, in addition to output layer module, input layer module, mode layer module, summation layer module, to compare layer module all be unidirectionally to pass
It is defeated；
Wherein, input layer module includes input layer X-axis submodule, input layer Y-axis submodule and input layer Z axis submodule；Mode
Layer module includes mode layer X-axis submodule, mode layer Y-axis submodule and mode layer Z axis submodule；Layer module of summing includes summation
Layer X-axis submodule, summation layer Y-axis submodule and summation layer Z axis submodule；
The connection relationship of each module is as follows in the monitoring system:
Input layer module is connect with mode layer module；Mode layer module is connect with summation layer module；Summation layer module layer compared with
Module connection；Compare layer module to connect with output layer module；
The function of each module is in the monitoring system:
The function of input layer module is the standard deviation for calculating one group of position data of receiver collection；Mode layer module
Function is the value for calculating activation primitive；Summation layer functions of modules be calculate input data belong to failure classes average probability and
Belong to the average probability of fault-free class；The function of comparing layer module is compared to the probability of malfunction from summation three axis of layer
It calculates；The function of output layer module is will to compare X-axis in layer, Y-axis and three axis output valve of Z axis to do Boolean calculation, and provide final
Testing result；
Wherein, shown in input layer module calculation formula following (1), (2) and (3):
Wherein, xi, yiAnd ziRespectively indicate i-th of input quantity of X-axis, Y-axis and Z axis, X2It indicates to input to mode layer X-axis submodule
Data, Y2Indicate the data inputted to mode layer Y-axis submodule, Z2Indicate the data inputted to mode layer Z axis submodule, n table
Show the quantity of neuron in submodule, the neuronal quantity of X-axis, Y-axis and Z axis submodule is equal；
The data of summation layer output are expressed as follows shown in (4), (5), (6), (7), (8) and (9) with Gaussian function:
Wherein, X31、Y31、Z31Respectively indicate the data that fault-free node exports into summation layer X-axis, Y-axis, Z axis submodule, X32、
Y32、Z32Indicate the data that malfunctioning node exports into summation layer X-axis, Y-axis, Z axis submodule, LXi、LYiAnd LZiRepresent fault-free
I-th of training sample of situation, LXFi、LYFiAnd LZFiI-th of training sample of faulty situation is represented, s indicates fault-free sample
This quantity, w indicate fault sample quantity, LXi、LYi、LZi、LXFi、LYFiAnd LZFiCan by following formula (10), (11), (12),
(13), (14) and (15) are calculated；
σ in formulax、σyAnd σzThe standard deviation of X-axis, Y-axis and the distribution of Z axis Satellite position error is respectively indicated, k indicates the coefficient of expansion,
D indicates sample sample size, CmIndicate correction factor, the neuronal quantity in mode layer is equal to the total quantity of training sample；
The operation of summation layer module can be calculated by following formula (16), (17), (18), (19), (20) and (21):
fx1=X31/s (16)
fx2=X32/w (17)
fy1=Y31/s (18)
fy2=Y32/w (19)
fz1=Z31/s (20)
fz2=Z32/w (21)
Wherein, fx1、fy1And fz1Respectively indicate X-axis, Y-axis and the trouble-free probability of Z axis locator value, fx2、fy2And fz2Respectively indicate X
Axis, Y-axis and the faulty probability of Z axis locator value, sum layer in neuronal quantity be 6；
The operation for comparing layer module is realized by formula (22), (23) and (22):
X in formula4, Y4, Z4The data inputted to output layer are respectively indicated, the quantity for comparing layer neuron is 3；
The operation of output layer module is will to compare X-axis in layer, Y-axis and three axis output valve of Z axis to do Boolean calculation, and operation method is as follows
Shown in formula (25):
R=X4∧Y4∧Z4 (25)
Wherein, R indicates output as a result, R=1 indicates that satellite used for positioning includes failure, and R=0 indicates that satellite positioning is normal,
There was only 1 neuron in output layer；
The Monitoring Performance of monitoring system is mainly influenced by following two parts:
1. training sample LXi、LYi、LXFi、LYFiAnd LZFiThe quality of data；
2. smoothing parameter λ；
The monitoring system in the modeling process of above layers, training sample using formula (10), (11), (12), (13),
(14) it is calculated with (15), smoothing parameter λ carries out optimizing using particle group optimizing method.
2. a kind of smoothing parameter optimization method based on particle group optimizing, method includes the following steps:
Step 1, initiation parameter；
Wherein, initiation parameter specifically includes given inertial parameter ω, constant C1And C2, random number R1And R2Initial value, give repeatedly
Generation sum ttotalWith the numerical value of total number of particles p；
Step 2, initialization particle rapidity and position；Specifically: in given range, the initial speed of each particle is set at random
V and position P is spent, and sets t=1 for number of iterations；
Step 3 updates particle rapidity and position, specifically includes following sub-step:
Population i is set i=1 by step 3.1；
Step 3.2 calculates the speed and positional value of particle i in the case of the t times iteration using following formula (26) and (27),
vi,t=ω × vi,t-1+C1×R1×(Pbesti-Pi,t-1)+C2×R2×(Gbest-Pi,t-1) (26)
Pi,t=Pi,t-1+vi,t (27)
.v in formulai,tIndicate particle rapidity of i-th of particle in the t times iteration, Pi,tIndicate i-th of particle in the t times iteration
When particle position, ω is inertial factor, C1And C2It is constant, R1And R2It is the random number in given section, PbestiIt is i-th
The history optimal value of particle, Gbest are the global optimums of all particles；
Step 3.3i=i+1, if i≤p jump procedure 3.2, otherwise, and i=1, jump procedure 4；
Step 4 calculates particle fitness, and the fitness F (P of i-th of particle is specifically calculated using following formula (28)i,t),
Wherein, Ffa(Pi) and Fmd(Pi) it is piecewise function:
F(Pi,t) indicate particle Pi,tFitness function, Qfa(Pi,t) indicate with particle Pi,tProbabilistic neural as smoothing parameter
For network monitoring system under non-failure conditions, there is the number of false alarm, Q in detection resultmd(Pi,t) it is that 35m pseudorange error occurs 6 seconds
In the case where, there is the number of missing inspection in fault-finding result；
The history optimal value of step 5, more new particle；By F (Pi,t) and F (Pbesti) compare, if F (Pi,t) > F (Pbesti),
Perform the following operations F (Pbesti)=F (Pi,t), and go to step 6；Otherwise, 8 are gone to step；
The global optimum of step 6, more new particle, by F (Pi,t) compared with F (Gbest), if F (Pi,t) > F (Gbest), it holds
Following operation F (Gbest)=F (P of rowi,t), go to step 7；Otherwise, 8 are gone to step；
Whether the particle that step 7, judgement search meets system requirements；F (Gbest)=1 is judged, if it is really going to step
10；If it is vacation, 8 are gone to step；
Step 8, more new particle carry out the operation of next particle, specifically: judge i≤p, if it is true, i=i+1, and jumps
To step 4；If it is vacation, 9 are gone to step；
Step 9 updates iteration, carries out next iteration operation, specifically: judge t≤ttotal, if it is true, t=t+1, and jump
Go to step 3；If it is vacation, 10 are gone to step；
Step 10 terminates operation, exports Pi,t。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811065716.4A CN109308518B (en) | 2018-09-13 | 2018-09-13 | Monitoring system based on probabilistic neural network and smoothing parameter optimization method thereof |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811065716.4A CN109308518B (en) | 2018-09-13 | 2018-09-13 | Monitoring system based on probabilistic neural network and smoothing parameter optimization method thereof |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109308518A true CN109308518A (en) | 2019-02-05 |
| CN109308518B CN109308518B (en) | 2021-09-10 |
Family
ID=65224573
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811065716.4A Active CN109308518B (en) | 2018-09-13 | 2018-09-13 | Monitoring system based on probabilistic neural network and smoothing parameter optimization method thereof |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109308518B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110850844A (en) * | 2019-11-11 | 2020-02-28 | 清华大学深圳国际研究生院 | Calculation method for generalized minimum detectable fault of multi-dimensional actuator |
| CN111708054A (en) * | 2020-06-16 | 2020-09-25 | 沈阳航空航天大学 | ARAIM vertical protection level optimization method based on particle swarm optimization algorithm |
| CN112200104A (en) * | 2020-10-15 | 2021-01-08 | 重庆科技学院 | Chemical engineering fault diagnosis method based on novel Bayesian framework for enhanced principal component analysis |
| CN112330064A (en) * | 2020-11-26 | 2021-02-05 | 中国石油大学(华东) | New drilling workload prediction method based on ensemble learning |
| CN112989734A (en) * | 2021-02-25 | 2021-06-18 | 中国人民解放军海军航空大学 | Equipment analog circuit fault diagnosis method based on probabilistic neural network |
| CN113447813A (en) * | 2020-09-03 | 2021-09-28 | 鲁能集团有限公司 | Fault diagnosis method and equipment for offshore wind generating set |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104361393A (en) * | 2014-09-06 | 2015-02-18 | 华北电力大学 | Method for using improved neural network model based on particle swarm optimization for data prediction |
| CN104502922A (en) * | 2014-12-09 | 2015-04-08 | 沈阳航空航天大学 | Autonomous integrity monitoring method for neural network assisted particle filter GPS (global positioning system) receiver |
| US20160109579A1 (en) * | 2014-10-16 | 2016-04-21 | Gmv Aerospace And Defence, S.A. | Device and method for computing an error bound of a kalman filter based gnss position solution |
| CN106920008A (en) * | 2017-02-28 | 2017-07-04 | 山东大学 | A kind of wind power forecasting method based on Modified particle swarm optimization BP neural network |
-
2018
- 2018-09-13 CN CN201811065716.4A patent/CN109308518B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104361393A (en) * | 2014-09-06 | 2015-02-18 | 华北电力大学 | Method for using improved neural network model based on particle swarm optimization for data prediction |
| US20160109579A1 (en) * | 2014-10-16 | 2016-04-21 | Gmv Aerospace And Defence, S.A. | Device and method for computing an error bound of a kalman filter based gnss position solution |
| CN104502922A (en) * | 2014-12-09 | 2015-04-08 | 沈阳航空航天大学 | Autonomous integrity monitoring method for neural network assisted particle filter GPS (global positioning system) receiver |
| CN106920008A (en) * | 2017-02-28 | 2017-07-04 | 山东大学 | A kind of wind power forecasting method based on Modified particle swarm optimization BP neural network |
Non-Patent Citations (1)
| Title |
|---|
| 彭雅奇 等: "MIEKF优化算法在GPS定位测速中的应用", 《计算机仿真》 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110850844A (en) * | 2019-11-11 | 2020-02-28 | 清华大学深圳国际研究生院 | Calculation method for generalized minimum detectable fault of multi-dimensional actuator |
| CN111708054A (en) * | 2020-06-16 | 2020-09-25 | 沈阳航空航天大学 | ARAIM vertical protection level optimization method based on particle swarm optimization algorithm |
| CN111708054B (en) * | 2020-06-16 | 2023-06-30 | 沈阳航空航天大学 | ARAIM vertical protection level optimization method based on particle swarm optimization algorithm |
| CN113447813A (en) * | 2020-09-03 | 2021-09-28 | 鲁能集团有限公司 | Fault diagnosis method and equipment for offshore wind generating set |
| CN112200104A (en) * | 2020-10-15 | 2021-01-08 | 重庆科技学院 | Chemical engineering fault diagnosis method based on novel Bayesian framework for enhanced principal component analysis |
| CN112200104B (en) * | 2020-10-15 | 2022-06-24 | 重庆科技学院 | Chemical engineering fault diagnosis method based on novel Bayesian framework for enhanced principal component analysis |
| CN112330064A (en) * | 2020-11-26 | 2021-02-05 | 中国石油大学(华东) | New drilling workload prediction method based on ensemble learning |
| CN112989734A (en) * | 2021-02-25 | 2021-06-18 | 中国人民解放军海军航空大学 | Equipment analog circuit fault diagnosis method based on probabilistic neural network |
| CN112989734B (en) * | 2021-02-25 | 2022-05-03 | 中国人民解放军海军航空大学 | Equipment analog circuit fault diagnosis method based on probabilistic neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109308518B (en) | 2021-09-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109308518A (en) | A kind of monitoring system and its smoothing parameter optimization method based on probabilistic neural network | |
| CN109490916B (en) | GNSS receiver autonomous integrity monitoring method | |
| CN111060133B (en) | Integrated navigation integrity monitoring method for urban complex environment | |
| CN110068840B (en) | ARAIM fault detection method based on pseudo-range measurement characteristic value extraction | |
| CN109359270B (en) | Threshold model establishing method for integrity risk monitoring of Beidou foundation enhancement system | |
| CN100582811C (en) | Method for monitoring GNSS receiver autonomous integrity based on multi-satellite failure recognition | |
| EP1965219A1 (en) | Method and system for predicting the performance of satellite navigation systems | |
| CN110161543B (en) | Partial gross error tolerance self-adaptive filtering method based on chi-square test | |
| CN107064961A (en) | The method and device tested satellite navigation system integrity monitoring performance | |
| CN109061683B (en) | H-ARAIM system for optimizing horizontal protection level | |
| CN112230247B (en) | GNSS integrity monitoring method used in urban complex environment | |
| Pesonen | A framework for Bayesian receiver autonomous integrity monitoring in urban navigation | |
| CN104199051A (en) | Method for detecting and identifying satellite navigation RAIM (Receiver Autonomous Integrity Monitoring) multi-satellite faults | |
| CN115265594B (en) | Multi-source PNT information elastic fusion navigation multi-level autonomous integrity monitoring method and system | |
| CN104331602A (en) | GNSS (global navigation satellite system) completeness detection rate estimation method based on Markov chain | |
| de Oliveira et al. | Recent advances in sensor integrity monitoring methods—A review | |
| CN115420284A (en) | Fault detection and identification method for combined navigation system | |
| CN115291255A (en) | Distributed GNSS anomaly monitoring method suitable for vehicle-mounted end | |
| CN115561782B (en) | Satellite fault detection method in integrated navigation based on odd-even vector projection | |
| Zheng et al. | A dynamic-data-driven method for improving the performance of receiver autonomous integrity monitoring | |
| CN111913192B (en) | GBAS integrity risk allocation method based on key star | |
| Luo et al. | A CIPSO-FCM-based RAIM algorithm for the GPS/BDS integrated navigation system | |
| CN116859417B (en) | Integrity monitoring method for Beidou PPP-RTK/MEMS | |
| CN117031502B (en) | Real-time PPP-RTK satellite Zhong Gui correction product integrity monitoring method | |
| CN116736339B (en) | Beidou autonomous monitoring and early warning method for control of forbidden navigation areas |
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 |