CN110428043A - Computer model parameter adaptive optimization method based on particle swarm algorithm - Google Patents

Computer model parameter adaptive optimization method based on particle swarm algorithm Download PDF

Info

Publication number
CN110428043A
CN110428043A CN201910622679.0A CN201910622679A CN110428043A CN 110428043 A CN110428043 A CN 110428043A CN 201910622679 A CN201910622679 A CN 201910622679A CN 110428043 A CN110428043 A CN 110428043A
Authority
CN
China
Prior art keywords
particle
parameter
value
averaged
computer model
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
Application number
CN201910622679.0A
Other languages
Chinese (zh)
Other versions
CN110428043B (en
Inventor
谢平
袁航
陈晓玲
张昌梦
金子强
程生翠
张园园
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Yanshan University
Original Assignee
Yanshan University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Yanshan University filed Critical Yanshan University
Priority to CN201910622679.0A priority Critical patent/CN110428043B/en
Publication of CN110428043A publication Critical patent/CN110428043A/en
Application granted granted Critical
Publication of CN110428043B publication Critical patent/CN110428043B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • G06N3/004—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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • 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
    • G06N3/048—Activation functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)

Abstract

The invention discloses a kind of Computer model parameter adaptive optimization method based on particle swarm algorithm, its step includes: S1: initialization population, particle swarm algorithm basic parameter is set, the parameter combination search range of each particle is set, Computer model basic parameter is set；S2: calculating the fitness value of each particle in primary group, initializes particle individual extreme value and global extremum according to particle fitness value；S3: particle rapidity and position are updated；S4: new particle fitness value, more new particle individual extreme value and global extremum are calculated；S5: judging whether to meet maximum number of iterations, be, exports global optimum's particle, otherwise return step S3；Step S6: the global optimum's particle exported according to step S5 obtains the best parameter group of brain electricity frequency range.The present invention provides a kind of Computer model parameter adjusting methods of convenient and efficient, increase parameter identification accuracy, shorten regulating time.

Description

Computer model parameter adaptive optimization method based on particle swarm algorithm
Technical field
The present invention relates to nervous system field, especially a kind of Computer model parameter based on particle swarm algorithm is adaptive Answer optimization method.
Background technique
With the development of nervous system Simulation and Modeling Technology, establishing EEG signals neuron models becomes a kind of research brain The important channel that electric signal is generated, transmits and handled.EEG signals neural model can be divided into two classes, and one kind is on microcosmic level Neural model is calculated, this class model is difficult to determine neuron models parameter, and needs to expend a large amount of computing resource；It is another kind of to be Lumped parameter model is thin to particular types on macro-level such as Computer model (neural mass model, NMM) The neuron pool overall permanence of born of the same parents' composition models, and has not only had simple form mathematically but also can preferably reflect nervous physiology Meaning on, therefore be used widely.
Computer model can emulate various forms of EEG signals, such as human normal or special with certain diseases EEG signals, local field potentials of sign etc..Correlative study person constantly expands Computer model, richer to generate Rich EEG signals.Jansen&Rit simulates spontaneous background EEG signals, class alpha signal and vision induced fixed using neuron pool Position；Wendling application Computer model emulates epileptic EEG Signal；David&Friston uses binary channels coupled neural member Group model simulates the narrowband EEG signals of the frequency range from δ to γ.In the emulation signal of above-mentioned Computer model, researchers It is to take examination to gather the mode of parameter to generate echo signal, this mode needs largely to adjust workload, it is difficult to shorter The EEG signals with certain feature or specific frequency are obtained in time.
Currently, existing research, which is based on optimization algorithm, probes into Computer model Parameters variation, to simulate inhibition or improve certain A little characteristic signals analyze the kinetic characteristics etc. of Computer model.University On The Mountain Of Swallows Liu Xian is based on the algebra estimation technique, has studied one The novel closed loop feedback control strategy of kind is to eliminate the epilepsy shape spike in neural group model；Shandong University Geng Shujuan with The key parameter of Wendling model is that the non-linear Bifurcation Analysis of branch parameter has studied the kinetic characteristics of the model.But Existing research is adjusted both for the neuron pool parameter optimization with characteristic signal greatly and analysis, for the specific frequency of conventional brain electricity The research of the analysis of segment signal and parameter optimization adjustment is relatively fewer.
Therefore, lack the method adjusted to the analysis of conventional brain electricity special frequency channel signal and parameter optimization at present, need to draw Entering some optimization algorithms progress parameter adaptive optimizations seems necessary.
Particle swarm optimization algorithm, which has, to be calculated short time, fast convergence rate, solves complex nonlinear optimization problem flexibly, easily In realize the advantages that, solve complex nonlinear optimization problem in be widely used.But particle swarm algorithm is used at present It is also rarely found in the research application of Computer model parameter adaptive optimization.
Summary of the invention
In order to make up the defect and overcome the deficiencies in the prior art of the above-mentioned prior art, the present invention provides one kind to be based on grain The Computer model parameter adaptive optimization method of swarm optimization.This method is conducive to quick obtaining and is given birth to by Computer model At brain electricity specific frequency waveform, reduce and manually adjust time of parameter, be conducive to study brain electrical feature frequency range and human brain Movable relationship, and can be used for analog portion genius morbi signal and the EEG signals after being stimulated.
It is to achieve the above object, of the invention that propose a kind of Computer model parameter based on particle swarm algorithm adaptive Optimization method is answered, step includes:
Step S1: particle swarm algorithm basic parameter is arranged in initialization population, and the parameter combination search of each particle is arranged Computer model basic parameter is arranged in range；
Step S2: the fitness value of each particle in the initialization population is calculated, according to the initialization population Middle particle fitness value initializes particle individual extreme value and global extremum, specific steps are as follows:
Step S2-1: the Computer model differential equation is established:
Wherein, HeIt is averaged cynapse gain, H for excitabilityiIt is averaged cynapse gain, τ for inhibitioneFor film averaging time constant, τiFor dendron averaging time constant；C1=α1C, C2=α2C, C3=C4=α3C=α4·C；P indicates external input Gauss White noise, p=mean+randn*sigma, mean=220, sigma=20, randn are random number；y0It is flat for cone neurone Equal film potential, that is, cone neurone subgroup output；y1It is averaged the impromptu nerve member subgroup of putting forth energy of film potential for excitability intrerneuron Output, y2It is averaged the output of film potential, that is, inhibitory neuron subgroup for inhibitory interneuron；y3, y4, y5Respectively y0, y1, y2Derivative；Sigm () is nonlinear function, specially nonlinear function moduleV is neuron The average film potential of group, vmaxFor the maximum excitation value of neuron pool, v0To reach postsynaptic electricity when 50% maximum excitation value Position, r are that the function existsSlope；
According to the four of each particle parameters: excitability is averaged cynapse gain He, inhibition is averaged cynapse gain Hi, film it is flat Equal timeconstantτe, dendron averaging time constant τi, using 4 rank Runge-Kutta (Runge-Kutta method) methods to neuron pool The model differential equation is solved, and obtains y for each particle1And y2, according to y1And y2It generates relative to the imitative of each particle True EEG signals waveform are as follows: eeg=y1-y2, wherein eeg is that cone neurone postsynaptic membrane is averaged film potential, representative model Output, for emulating EEG signals, y1For excitability intrerneuron be averaged film potential it is impromptu put forth energy nerve member subgroup output, y2 It is averaged the output of film potential, that is, inhibitory neuron subgroup for inhibitory interneuron；
Step S2-2: it is straight to remove 0 frequency to carry out 1Hz high-pass filtering to the artificial brain electric signal waveform relative to the particle Signal is flowed, Fourier transformation then is carried out to filtered signal, obtains width peak value and its corresponding Frequency point in spectrogram fc；
Step S2-3: each particle fitness value calculation formula are as follows: fitness=| fc-fg|, wherein fcFor currently by frequency spectrum Scheme the width peak value respective frequencies obtained, fgFor the target frequency of setting；
Step S2-4: it repeats step S2-1 to S2-3 and calculates the particle fitness value for obtaining each particle in population, just Beginningization particle individual extreme value and global extremum；
Step S3: the speed of particle and position in population are updated, new population is formed；
Each particle updates calculation formula used in particle rapidity and position in population are as follows:
In formula,It is the particle f speed that d is tieed up in+1 iteration of kth；It is particle i d in kth time iteration The speed of dimension, c1, c2It is the first Studying factors and the second Studying factors respectively, the first Studying factors adjust global optimum's flight Maximum step-length, the second Studying factors adjust the maximum step-length of the optimal flight of individual；rand1And rand2It is random between [0,1] Number；It is the position of particle i d dimension in kth time iteration,It is the position that d is tieed up in+1 iteration of kth particle i；It is the position of particle i individual extreme point of d dimension in kth time iteration；It is entire population in kth time iteration In d dimension global extreme point position；wkIt is inertia weight, calculating formula is as follows:
In formula, wsFor initial inertia weight, weTo iterate to inertia weight when maximum times, k is current iteration number, KmaxFor maximum number of iterations.
After the completion of the location updating of particles all in population, new population is formd.
Step S4: using the method for calculating particle fitness value in the step S2, each grain is calculated to new population Sub- fitness value；Further according to each particle fitness value more new particle individual extreme value and global extremum, specially if particle The fitness value individual extreme value current better than the particle, thenIt is arranged to the position of the particle, and more new individual extreme value； It is optimal in the individual extreme value of such as all particles to be better than current global extremum, thenIt is arranged to the position of the particle, And update global extremum；
Step S5: judging whether to meet maximum number of iterations, be, exports global optimum's particle, otherwise return step S3；
Step S6: the global optimum's particle exported according to step S5 obtains the best parameter group of brain electricity corresponding band, then The artificial brain electricity frequency range waveform of this frequency range is generated according to the best parameter group of brain electricity corresponding band.
Preferably, step further include: brain electricity includes 5 frequency ranges δ, θ, α, lowβ、highβAnd γ, to brain electricity 5 frequency ranges point It She Zhi not target frequency fgValue, according to corresponding target frequency fgStep S1~S6 is executed, finally obtains electric 5 frequencies of brain respectively The artificial brain electricity frequency range waveform of section.
Preferably, step includes: the target frequency f of electric 5 frequency ranges of braingIt is set to: δ=3Hz, θ=5Hz, α= 12Hz, lowβ=17Hz, highβ=25Hz, γ=41Hz.
Preferably, the step S1 is specifically included:
Step S11: setting particle swarm algorithm basic parameter, the particle swarm algorithm basic parameter include the first Studying factors C1 and the second Studying factors c2, particle dimension d, maximum number of iterations Kmax, population scale sizepop, population particle search range Bound, initial weight weWith weight w when reaching greatest iterations；
Step S1-2: parameter combination corresponding to each particle includes: that excitability is averaged cynapse gain He, inhibition it is average Cynapse gain Hi, film averaging time constant τeWith dendron averaging time constant τi；The parameter combination search model of each particle is set It encloses；
Step S1-3: average Synaptic junction the is set on the basic parameter of Computer model, including excitability feedback loop One average probability α1With the second average probability α2；Average Synaptic junction third average probability α on inhibition feedback loop3With Siping City Equal probability α4；The parameter v of nonlinear function Sigm is setmax、v0And r, nonlinear functionMiddle vmax For the maximum excitation value of neuron pool, v0To reach postsynaptic potential when 50% maximum excitation value, r is that the function existsSlope；External input white Gaussian noise p=mean+randn*sigma, mean=220, sigma=are set 20, randn be random number；Maximum cynapse between subgroup is averaged linking number C.
Preferably, in the step S1-2 further include:
Parameter combination search range corresponding to each particle are as follows: excitability is averaged cynapse gain HeFor 2.6~9.75mV, Inhibition is averaged cynapse gain HiFor 17.6~110mV, film averaging time constant τeFor 2~150ms, dendron averaging time constant τiFor 2~150ms.
Compared with prior art, the beneficial effects of the present invention are:
Invention increases the accuracys of optimized parameter identification, and the parameter regulation time is greatly shortened, effectively prevents people Work carries out trying to gather huge workload and poor target effect needed for parameter.
Detailed description of the invention
Fig. 1 is flow diagram of the invention；
Fig. 2 is the calculating schematic diagram of particle fitness value；
Fig. 3 is the simplified model figure of neuron pool.
Specific embodiment
With reference to the accompanying drawing and specific embodiment the present invention will be further described.
Brain electricity includes 5 frequency ranges δ, θ, a, lowβ、highβAnd γ, wherein lowβFor β frequency range slow wave, highβFor β frequency range Fast wave, δ, θ, a, lowβ、highβWith the target frequency f of γgDifference, in the present embodiment the target frequency f of electric 5 frequency ranges of braing It is set to: δ=3Hz, θ=5Hz, α=12Hz, lowβ=17Hz, highβ=25Hz, γ=41Hz, for 5 frequencies of brain electricity Section executes step S1~S6 respectively, to obtain the best parameter group of each frequency range respectively.
It is the process of the Computer model parameter adaptive optimization method the present invention is based on particle swarm algorithm as shown in Figure 1 Figure, comprising the following steps:
Step S1: particle swarm algorithm basic parameter is arranged in initialization population, and the parameter combination search of each particle is arranged Computer model basic parameter is arranged in range；
Step S1-1: setting particle swarm algorithm basic parameter, particle swarm algorithm basic parameter include the first Studying factors c1 With the second Studying factors c2, particle dimension d, maximum number of iterations Kmax, population scale sizepop, population particle search range Bound, initial weight weWith weight w when reaching greatest iterations；
Step S1-2: being arranged the parameter combination search range of each particle, and parameter combination corresponding to each particle includes: Excitability is averaged cynapse gain He, inhibition is averaged cynapse gain Hi, film averaging time constant τeWith dendron averaging time constant τi；Wherein, excitability is averaged cynapse gain HeSearch range be set as 2.6~9.75mV；Inhibition is averaged cynapse gain Hi's Search range is set as 17.6~110mV；Film averaging time constant τeSearch range be set as 2~150ms；Dendron mean time Between constant, τiSearch range be set as 2~150ms；
Step S1-3: average Synaptic junction the is set on the basic parameter of Computer model, including excitability feedback loop One average probability α1With the second average probability α2, a1=1, α2=0.8；Average Synaptic junction third is average on inhibition feedback loop Probability α3With the 4th average probability α4, α3==0.25, α4=0.25；Nonlinear function Sigm parameter are as follows: v0=6mV, vmax= 5s-1, r=0.56mV-1；External input white Gaussian noise p=mean+randn*sigma, mean=220, sigma=20, Randn is random number；Maximum cynapse between subgroup is averaged linking number C=135；
Step S2: calculating the fitness value of each particle in initialization population, suitable according to particle in initialization population Angle value is answered, particle individual extreme value and global extremum are initialized.
Fig. 2 is to calculate particle fitness value schematic diagram, the calculating of the fitness value of particle specifically includes the following steps:
Step S2-1: Fig. 3 is the simplified model figure of neuron pool, wherein y0, y1, y2For dynamic linear function module (M1, M2, M3) output, state variable y3, y4, y5Respectively y0, y1, y2Derivative, nonlinear function The average film potential v of neuron pool is converted to the averag density for the action potential that neuron is excited, vmaxNeuron has been determined The maximum excitation value of group, v0To reach postsynaptic potential when 50% maximum excitation value, r is that the function existsIt is oblique Rate；Two class dynamic linear function modules include excitability he(t) and inhibition hi(t):
Wherein, HeIt is averaged cynapse gain, H for excitabilityiIt is averaged cynapse gain, τ for inhibitioneFor film averaging time constant, τiIt is the time for dendron averaging time constant, t, u (t) is unit jump function, its value takes 0 when its value takes 1, t≤0 when t >=0.
According to the prior art, to h (t)=H τ te-tτU (t) asks Laplace (Laplce) transformation that dynamic linear can be obtained The transfer function of function module:Can must be described by carrying out Laplace inverse transformation to transfer function by three The differential equation of dynamic linear function module (M1, M2, M3):Wherein, x (t) For the input of each module, then each module can indicate are as follows:By the shape introduced State variable z (t) (y3, y4, y5) each module can be indicated again are as follows:Therefore, In The simplified model of Fig. 3 neuron pool is represented by the following Computer model differential equation in the present embodiment:
Wherein, wherein HeIt is averaged cynapse gain, H for excitabilityiIt is averaged cynapse gain, τ for inhibitioneFor film average time Constant, τiFor dendron averaging time constant；C1=α1C, C2=α2C, C3=C4=α3C=α4·C；P indicates external input White Gaussian noise；y0It is averaged the output of film potential, that is, cone neurone subgroup for cone neurone；y1It is flat for excitability intrerneuron Equal film potential, that is, excitatory neuron subgroup output, y2It is averaged film potential, that is, inhibitory neuron for inhibitory interneuron Group's output；y3, y4, y5Respectively y0, y1, y2Derivative；Sigm () is nonlinear function, specially nonlinear function moduleV is the average film potential of neuron pool, vmaxFor the maximum excitation value of neuron pool, v0To reach Postsynaptic potential when 50% maximum excitation value, r are that the function existsSlope；
According to the four of each particle parameters: excitability is averaged cynapse gain He, inhibition is averaged cynapse gain Hi, film it is flat Equal timeconstantτe, dendron averaging time constant τi, using 4 rank Runge-Kutta (Runge-Kutta method) methods to neuron pool The model differential equation is solved, and obtains y for each particle1And y2, according to y1And y2It generates relative to the imitative of each particle True EEG signals waveform are as follows: eeg=y1-y2, wherein eeg is that cone neurone postsynaptic membrane is averaged film potential, representative model Output, for emulating EEG signals, y1For excitability intrerneuron be averaged film potential it is impromptu put forth energy nerve member subgroup output, y2 It is averaged the output of film potential, that is, inhibitory neuron subgroup for inhibitory interneuron；
Step S2-2: 1Hz high pass is carried out to the artificial brain electric signal waveform relative to each particle that step S2-1 is generated Filtering to remove 0 frequency direct current signal, then to filtered signal carry out Fourier transformation, obtain spectrogram in width peak value and Its corresponding Frequency point fc；
Step S2-3: each particle fitness value calculation formula are as follows: fitness=| fc-fg|, wherein fcFor currently by frequency spectrum Scheme the width peak value respective frequencies obtained, fgFor the target frequency of setting；
Step S2-4: it repeats step S2-1 to S2-3 and calculates the particle fitness value for obtaining each particle in population, just Beginningization particle individual extreme value and global extremum.
Step S3: the speed of particle and position in population are updated, new population is formed.
Each particle updates calculation formula used in particle rapidity and position in population are as follows:
In formula,It is the particle i speed that d is tieed up in+1 iteration of kth；It is particle i d in kth time iteration The speed of dimension, c1, c2It is the first Studying factors and the second Studying factors respectively, the first Studying factors adjust global optimum's flight Maximum step-length, the second Studying factors adjust the maximum step-length of the optimal flight of individual；rand1And rand2It is random between [0,1] Number；It is the position of particle i d dimension in kth time iteration,It is the position that d is tieed up in+1 iteration of kth particle i；It is the position of particle i individual extreme point of d dimension in kth time iteration；It is entire population in kth time iteration In d dimension global extreme point position；wkIt is inertia weight, calculating formula is as follows:
In formula, wsFor initial inertia weight, weTo iterate to inertia weight when maximum times, k is current iteration number, KmaxFor maximum number of iterations.
After the completion of the location updating of particles all in population, new population is formd.
Step S4: using the method for calculating particle fitness value in step S2, it is suitable that each particle is calculated to new population Answer angle value.Further according to each particle fitness value more new particle individual extreme value and global extremum, specially if the adaptation of particle The angle value individual extreme value current better than the particle, thenIt is arranged to the position of the particle, and more new individual extreme value；As institute Have it is optimal better than current global extremum in the individual extreme value of particle, thenIt is arranged to the position of the particle, and more New global extremum.
Using in step S2 calculate particle fitness value method, specifically: be using the Computer model differential equation Formula (1) solves the artificial brain electric signal waveform eeg=y for obtaining each particle1-y2；To the emulation EEG signals of each particle Waveform carries out 1Hz high-pass filtering to remove 0 frequency direct current signal, then carries out Fourier transformation to filtered signal, obtains frequency Width peak value and its corresponding Frequency point f in spectrogramc；Use formula fitness=| fc-fg| calculate each particle fitness value.
Step S5: judging whether to meet maximum number of iterations, be, exports global optimum's particle, otherwise return step S3；
Step S6: the global optimum's particle exported according to step S5 obtains the best parameter group of brain electricity corresponding band, then The artificial brain electricity frequency range waveform of this frequency range is generated according to the best parameter group of brain electricity corresponding band.
After brain 5 frequency ranges of electricity execute step S1~S6 respectively, then the brain electricity frequency range of the emulation of electric 5 frequency ranges of brain can be generated Waveform.
Table 1 is a specific embodiment using the application, in the present embodiment, Studying factors c is arranged1== 1.49 c2=1.49, maximum number of iterations Kmax=20, population scale sizepop=100, particle dimension d=4, initial weight we =0.9, reach weight w when maximum number of iterationss=0.4, using the neuron pool mould based on particle swarm algorithm of the application Shape parameter adaptive optimization method finally obtains the optimized parameter group that Computer model generates artificial brain electricity target frequency waveform It closes as shown in table 1, and artificial brain electricity actual frequency and the progress of artificial brain electricity target frequency to the best parameter group using table 1 Comparison.
Table 1
A kind of Computer model parameter adaptive optimization method based on particle swarm algorithm of the present invention, by particle group optimizing Algorithm is applied in the adjusting of Computer model parameter optimization, and to obtain the electric 5 special frequency channel typical waveforms of brain, the present invention is one Determine to provide a kind of method of convenient and efficient in degree for the optimization of Computer model parameter adaptive.As can be seen from Table 1, make It effectively prevents manually being tried so that actual frequency is very close with target frequency with the parameter that method of the invention obtains Gather huge workload and poor target effect needed for parameter.
The preferred embodiment of the patent is described in detail above, but the present invention is not limited to above-mentioned embodiment party Formula within the knowledge of one of ordinary skill in the art can also be under the premise of not departing from this patent objective It makes a variety of changes.

Claims (5)

1. a kind of Computer model parameter adaptive optimization method based on particle swarm algorithm, which is characterized in that its step packet It includes:
Step S1: particle swarm algorithm basic parameter is arranged in initialization population, and the parameter combination search model of each particle is arranged It encloses, Computer model basic parameter is set；
Step S2: the fitness value of each particle in the initialization population is calculated, according to grain in the initialization population Sub- fitness value initializes particle individual extreme value and global extremum, specific steps are as follows:
Step S2-1: the Computer model differential equation is established:
Wherein, HeIt is averaged cynapse gain, H for excitabilityiIt is averaged cynapse gain, τ for inhibitioneFor film averaging time constant, τiFor Dendron averaging time constant；C1=α1C, C2=α2C, C3=C4=α3C=α4·C；P indicates external input Gauss white noise Sound, p=mean+randn*sigma, mean=220, sigma=20, randn are random number；y0For cone neurone average film Current potential, that is, cone neurone subgroup output；y1It is defeated for the excitability intrerneuron impromptu nerve member subgroup of putting forth energy of film potential that is averaged Out, y2It is averaged the output of film potential, that is, inhibitory neuron subgroup for inhibitory interneuron；y3, y4, y5Respectively y0, y1, y2 Derivative；Sigm () is nonlinear function, specially nonlinear function moduleV is neuron pool Average film potential, vmaxFor the maximum excitation value of neuron pool, v0To reach postsynaptic potential when 50% maximum excitation value, r Exist for the functionSlope；
According to the four of each particle parameters: excitability is averaged cynapse gain He, inhibition is averaged cynapse gain Hi, film mean time Between constant, τe, dendron averaging time constant τi, the Computer model differential equation is asked using 4 rank Runge-Kutta methods Solution, obtains y for each particle1And y2, according to y1And y2Generate the artificial brain electric signal waveform relative to each particle are as follows: Eeg=y1-y2, wherein eeg is that cone neurone postsynaptic membrane be averaged film potential, and the output of representative model is electric for artificial brain Signal, y1For excitability intrerneuron be averaged film potential it is impromptu put forth energy nerve member subgroup output, y2For inhibitory interneuron Average film potential, that is, inhibitory neuron subgroup output；
Step S2-2: 1Hz high-pass filtering is carried out to the artificial brain electric signal waveform relative to the particle to remove 0 frequency direct current letter Number, Fourier transformation then is carried out to filtered signal, obtains width peak value and its corresponding Frequency point f in spectrogramc；
Step S2-3: each particle fitness value calculation formula are as follows: fitness=| fc-fg|, wherein fcCurrently to be obtained by spectrogram The width peak value respective frequencies taken, fgFor the target frequency of setting；
Step S2-4: it repeats step S2-1 to S2-3 and calculates the particle fitness value for obtaining each particle in population, initialization Particle individual extreme value and global extremum；
Step S3: the speed of particle and position in population are updated, new population is formed；
Each particle updates calculation formula used in particle rapidity and position in population are as follows:
In formula,It is the particle i speed that d is tieed up in+1 iteration of kth；It is the speed of particle i d dimension in kth time iteration Degree, c1, c2It is the first Studying factors and the second Studying factors respectively, the first Studying factors adjust the maximum step of global optimum's flight Long, the second Studying factors adjust the maximum step-length of the optimal flight of individual；rand1And rand2It is the random number between [0,1]； It is the position of particle i d dimension in kth time iteration,It is the position that d is tieed up in+1 iteration of kth particle i；It is The position of particle i individual extreme point of d dimension in kth time iteration；It is that entire population d in kth time iteration is tieed up The position of global extreme point；wkIt is inertia weight, calculating formula is as follows:
In formula, wsFor initial inertia weight, weTo iterate to inertia weight when maximum times, k is current iteration number, KmaxFor Maximum number of iterations；
After the completion of the location updating of particles all in population, new population is formd；
Step S4: using the method for calculating particle fitness value in the step S2, it is suitable that each particle is calculated to new population Answer angle value；Further according to each particle fitness value more new particle individual extreme value and global extremum, specially if the adaptation of particle The angle value individual extreme value current better than the particle, thenIt is arranged to the position of the particle, and more new individual extreme value；As institute Have it is optimal better than current global extremum in the individual extreme value of particle, thenIt is arranged to the position of the particle, and updates Global extremum；
Step S5: judging whether to meet maximum number of iterations, be, exports global optimum's particle, otherwise return step S3；
Step S6: the global optimum's particle exported according to step S5 obtains the best parameter group of brain electricity corresponding band, further according to The best parameter group of brain electricity corresponding band generates the artificial brain electricity frequency range waveform of this frequency range.
2. a kind of Computer model parameter adaptive optimization method based on particle swarm algorithm according to claim 1, It is characterized in that, step further include:
Brain electricity includes 5 frequency ranges δ, θ, α, lowβ、highβAnd γ, target frequency f is respectively set to electric 5 frequency ranges of braingValue, root According to corresponding target frequency fgStep S1~S6 is executed, finally obtains the artificial brain electricity frequency range waveform of electric 5 frequency ranges of brain respectively.
3. a kind of Computer model parameter adaptive optimization method based on particle swarm algorithm according to claim 2, It is characterized in that, step includes:
The target frequency f of brain 5 frequency ranges of electricitygIt is set to: δ=3Hz, θ=5Hz, α=12Hz, lowβ=17Hz, highβ= 25Hz, γ=41Hz.
4. a kind of Computer model parameter adaptive optimization method based on particle swarm algorithm according to claim 1, It is characterized in that, the step S1 is specifically included:
Step S1-1: setting particle swarm algorithm basic parameter, the particle swarm algorithm basic parameter include the first Studying factors c1With Second Studying factors c2, particle dimension d, maximum number of iterations Kmax, population scale sizepop, population particle search range Bound, initial weight weWith weight w when reaching greatest iterations；
Step S1-2: parameter combination corresponding to each particle includes: that excitability is averaged cynapse gain He, inhibition is averaged cynapse Gain Hi, film averaging time constant τeWith dendron averaging time constant τi；The parameter combination search range of each particle is set；
Step S1-3: it is flat that average Synaptic junction first is set on the basic parameter of Computer model, including excitability feedback loop Equal probability α1With the second average probability α2；Average Synaptic junction third average probability α on inhibition feedback loop3It is average general with the 4th Rate α4；The parameter v of nonlinear function Sigm is setmax、v0And r, nonlinear functionMiddle vmaxFor nerve The maximum excitation value of first group, v0To reach postsynaptic potential when 50% maximum excitation value, r is that the function exists's Slope；Be arranged external input white Gaussian noise p=mean+randn*sigma, mean=220, sigma=20, randn be with Machine number；Maximum cynapse between subgroup is averaged linking number C.
5. a kind of Computer model parameter adaptive optimization method based on particle swarm algorithm according to claim 4, It is characterized in that, in the step S1-2 further include:
Parameter combination search range corresponding to each particle are as follows: excitability is averaged cynapse gain HeFor 2.6~9.75mV, inhibition Mild-natured equal cynapse gain HiFor 17.6~110mV, film averaging time constant τeFor 2~150ms, dendron averaging time constant τiIt is 2 ~150ms.
CN201910622679.0A 2019-07-11 2019-07-11 Neuron group model parameter adaptive optimization method based on particle swarm optimization Active CN110428043B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910622679.0A CN110428043B (en) 2019-07-11 2019-07-11 Neuron group model parameter adaptive optimization method based on particle swarm optimization

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910622679.0A CN110428043B (en) 2019-07-11 2019-07-11 Neuron group model parameter adaptive optimization method based on particle swarm optimization

Publications (2)

Publication Number Publication Date
CN110428043A true CN110428043A (en) 2019-11-08
CN110428043B CN110428043B (en) 2021-05-07

Family

ID=68410440

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910622679.0A Active CN110428043B (en) 2019-07-11 2019-07-11 Neuron group model parameter adaptive optimization method based on particle swarm optimization

Country Status (1)

Country Link
CN (1) CN110428043B (en)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110765654A (en) * 2019-11-14 2020-02-07 南方电网科学研究院有限责任公司 Parameter self-adaptive proxy method and device suitable for power market simulation
CN110859600A (en) * 2019-12-06 2020-03-06 深圳市德力凯医疗设备股份有限公司 Method for generating electroencephalogram signal, storage medium and electronic equipment
CN110931123A (en) * 2019-12-06 2020-03-27 深圳市德力凯医疗设备股份有限公司 Matrix brain network and construction method thereof
CN111860179A (en) * 2020-06-23 2020-10-30 浙江清华长三角研究院 Self-adaptive filtering method based on non-contact sensor
CN112395803A (en) * 2020-09-11 2021-02-23 北京工商大学 ICP-AES multimodal spectral line separation method based on particle swarm optimization
CN114521904A (en) * 2022-01-25 2022-05-24 中山大学 Electroencephalogram activity simulation method and system based on coupled neuron group

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20140009715A (en) * 2012-07-12 2014-01-23 중앙대학교 산학협력단 Apparatus and method for determining of optimal eeg channel based on pso
CN105740887A (en) * 2016-01-26 2016-07-06 杭州电子科技大学 Electroencephalogram feature classification method based on PSO-SVM (Particle Swarm Optimization-Support Vector Machine)
US20180136976A1 (en) * 2016-11-14 2018-05-17 King Abdulaziz University Temporal task scheduling in a hybrid system
CN108399451A (en) * 2018-02-05 2018-08-14 西北工业大学 A kind of Hybrid Particle Swarm Optimization of combination genetic algorithm
CN109992113A (en) * 2019-04-09 2019-07-09 燕山大学 A kind of MI-BCI system and its control method induced based on more scenes

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20140009715A (en) * 2012-07-12 2014-01-23 중앙대학교 산학협력단 Apparatus and method for determining of optimal eeg channel based on pso
CN105740887A (en) * 2016-01-26 2016-07-06 杭州电子科技大学 Electroencephalogram feature classification method based on PSO-SVM (Particle Swarm Optimization-Support Vector Machine)
US20180136976A1 (en) * 2016-11-14 2018-05-17 King Abdulaziz University Temporal task scheduling in a hybrid system
CN108399451A (en) * 2018-02-05 2018-08-14 西北工业大学 A kind of Hybrid Particle Swarm Optimization of combination genetic algorithm
CN109992113A (en) * 2019-04-09 2019-07-09 燕山大学 A kind of MI-BCI system and its control method induced based on more scenes

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
BONAN SHAN 等: "Estimate the effective connectivity in multi-coupled neural mass model using particle swarm optimization", 《STATISTICAL MECHANICS AND ITS APPLICATIONS》 *
耿淑娟: "基于分支理论的神经元群模型非线性动力学特性分析", 《中国博士学位论文全文数据库医药卫生科技辑》 *
陈迎亚: "康复运动中脑肌电融合及同步分析方法研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 *

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110765654A (en) * 2019-11-14 2020-02-07 南方电网科学研究院有限责任公司 Parameter self-adaptive proxy method and device suitable for power market simulation
CN110859600A (en) * 2019-12-06 2020-03-06 深圳市德力凯医疗设备股份有限公司 Method for generating electroencephalogram signal, storage medium and electronic equipment
CN110931123A (en) * 2019-12-06 2020-03-27 深圳市德力凯医疗设备股份有限公司 Matrix brain network and construction method thereof
WO2021109600A1 (en) * 2019-12-06 2021-06-10 深圳市德力凯医疗设备股份有限公司 Electroencephalogram signal generation method, storage medium and electronic device
CN110931123B (en) * 2019-12-06 2024-01-05 深圳市德力凯医疗设备股份有限公司 Matrix type brain network and construction method thereof
CN111860179A (en) * 2020-06-23 2020-10-30 浙江清华长三角研究院 Self-adaptive filtering method based on non-contact sensor
CN112395803A (en) * 2020-09-11 2021-02-23 北京工商大学 ICP-AES multimodal spectral line separation method based on particle swarm optimization
CN112395803B (en) * 2020-09-11 2023-10-13 北京工商大学 ICP-AES multimodal spectral line separation method based on particle swarm optimization
CN114521904A (en) * 2022-01-25 2022-05-24 中山大学 Electroencephalogram activity simulation method and system based on coupled neuron group
CN114521904B (en) * 2022-01-25 2023-09-26 中山大学 Brain electrical activity simulation method and system based on coupled neuron group

Also Published As

Publication number Publication date
CN110428043B (en) 2021-05-07

Similar Documents

Publication Publication Date Title
CN110428043A (en) Computer model parameter adaptive optimization method based on particle swarm algorithm
CN107316099A (en) Ammunition Storage Reliability Forecasting Methodology based on particle group optimizing BP neural network
CN110119590B (en) Water quality model particle filtering assimilation method based on multi-source observation data
CN103745271B (en) Forecasting method for induction thermal deposition calcium-phosphate coating process on basis of neural network
CN104794973B (en) Distributed cardiovascular system realistic model
CN105427241A (en) Distortion correction method for large-field-of-view display device
CN105894090A (en) Tide intelligent real-time forecasting method based on adaptive variation particle swarm optimization
CN106521459A (en) Optimization method of MOCVD equipment growth uniformity process parameters
Shah et al. Monte Carlo simulation of microbial population growth
CN109492766A (en) A kind of width learning method based on minimum P norm
Chen et al. Research on prediction model of tractor sound quality based on genetic algorithm
Damayanti et al. Epilepsy detection on EEG data using backpropagation, firefly algorithm and simulated annealing
CN104182854A (en) Mixed energy consumption measuring method for enterprise energy management system
CN115826743B (en) Multi-channel brain-computer signal modeling method for SSVEP brain-computer interface
CN110474323B (en) Method for measuring inertia time constant of power system
CN110450164A (en) Robot control method, device, robot and storage medium
ELIAS et al. Modelling tidal-inlet morphodynamics on medium time scales
Hartmann et al. Growth rate estimation of algae in raceway ponds: A novel approach
Kurniawan et al. Premise Parameter Optimization on Adaptive Network Based Fuzzy Inference System Using Modification Hybrid Particle Swarm Optimization and Genetic Algorithm
Sang et al. Double Inverted Pendulum control based on three-loop PID and improved BP Neural network
CN113761777A (en) Ultra-short-term photovoltaic power prediction method based on HP-OVMD
Pramanik Use of artificial neural networks for prediction of cell mass and ethanol concentration in batch fermentation using saccharomyces cerevisiae yeast
Friederich et al. Data modelling for analysis of adaptive changes in fly photoreceptors
Cui et al. Study on decoupling control system of temperature and humidity in intelligent plant factory
CN109840629A (en) Based on wavelet transformation-dendron shape neuron models photovoltaic power prediction technique

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