Method and device for analyzing coexistence of spike and plateau behaviors of a memristive Hebbian neuron

Abstract

The application discloses an analysis method and device for coexistence discharge behavior of a memristor taboo learning neuron, and the method comprises the following steps: selection and verification of a memristor; modeling of a memristor taboo learning neuron; alternating current equilibrium state analysis; determination of a parameter domain of discharge behavior; detection of coexistence discharge behavior; and verification of coexistence discharge behavior. Compared with the prior art, the discharge behavior detection method provided by the application has high accuracy, is more comprehensive, and can effectively capture coexistence discharge activity.

Description

Technical Field

[0001] This invention relates to the field of neuronal dynamics analysis, and in particular to a method and apparatus for analyzing the co-storage electrical behavior of memristor taboo learning neurons. Background Technology

[0002] Since the discovery of memristors, various neurons have been coupled with memristors to characterize neuronal self-synapses, coupled synapses, electromagnetic radiation, and weights, thus revealing the complex firing activities of memristor neurons, such as chaos, burst oscillations, and multistability behaviors. Tabu-learning neural networks utilize tabu search algorithms to effectively search the solution space, enabling trajectories to escape local minima. As a fundamental component, research has limited understanding of the specific firing behaviors of memristor tabu-learning neurons stimulated by the combined effects of electromagnetic radiation and external input current. In reality, electromagnetic radiation and ion exchange currents in the neuronal's physiological environment induce various coexisting firing behaviors in memristor tabu-learning neurons. Detailed analysis of these coexisting firing behaviors from the perspectives of initial value dynamics and parametric dynamics reveals the firing mechanism of memristor tabu-learning neurons. Traditional methods such as one-dimensional bifurcation diagrams and Lyapunov exponent diagrams have limitations in determining the coexisting firing activities of memristor tabu-learning neurons, failing to identify the firing type of neurons in a broader parameter domain. Therefore, novel analytical methods such as two-dimensional bifurcation diagrams and two-dimensional complexity diagrams have received increasing attention. However, novel analytical methods are rarely used to systematically analyze the co-storage behavior of memristor taboo learning neurons. Summary of the Invention

[0003] The main objective of this invention is to provide a method for analyzing the co-storage electrical behavior of memristor taboo learning neurons, comprising:

[0004] S101: Select a magnetically controlled memristor and obtain a mathematical model. Solve the mathematical model using the fourth-order Runge-Kutta method and verify the characteristics of the magnetically controlled memristor through a hysteresis loop.

[0005] S102: Select a tabu learning neuron model based on the characteristics of the tabu search algorithm, and combine the tabu learning neuron model with a magnetically controlled memristor to form a memristor tabu learning neuron;

[0006] S103: Obtain equilibrium points based on the tabu learning neuron model, and classify the equilibrium points into Hopf bifurcation point type, and / or folded bifurcation point type, and / or unstable saddle point type, and / or unstable node type, and / or unstable foci type, and / or unstable saddle-foci type, and / or stable node type, and / or stable foci type.

[0007] S104: Determine the parameter domain for dynamic behavior related to the initial value of the memristor and the magnitude of the external current;

[0008] S105: Based on the parameter domain, by fixing one control parameter and changing another, various discharge behaviors with different periods can be obtained; the discharge behaviors induced by two or more initial values ​​under different control parameters are superimposed to obtain co-storage discharge behavior; different types of co-storage discharge behavior are detected by one-dimensional bifurcation diagram, maximum Lyapunov exponent spectrum, phase diagram and time series diagram;

[0009] S106: Using PSIM power electronics simulation software, a simulation circuit for memristor tabu learning neurons was built and verified through simulation.

[0010] According to another aspect of this application, an analysis device for the co-storage electrical behavior of memristor tabu learning neurons is also provided, comprising:

[0011] The memristor selection and verification module selects a magnetically controlled memristor and obtains a mathematical model. It then uses the fourth-order Runge-Kutta method to solve the mathematical model and verifies the characteristics of the magnetically controlled memristor through a hysteresis loop.

[0012] The memristor tabu learning neuron modeling module selects tabu learning neuron models based on the characteristics of the tabu search algorithm, and combines the tabu learning neuron models with magnetically controlled memristors to form memristor tabu learning neurons.

[0013] The communication equilibrium analysis module obtains equilibrium points based on the taboo learning neuron model and classifies the equilibrium points into Hopf bifurcation point type, and / or folded bifurcation point type, and / or unstable saddle point type, and / or unstable node type, and / or unstable foci type, and / or unstable saddle-foci type, and / or stable node type, and / or stable foci type.

[0014] The parameter domain determination module for discharge behavior determines the parameter domain of the dynamic behavior that is related to the initial value of the memristor and the amplitude of the external current.

[0015] The co-storage discharge behavior detection module, based on the parameter domain, can obtain various discharge behaviors with different periods by fixing one control parameter and changing another. The co-storage discharge behavior is obtained by superimposing two or more sets of initial value-induced discharge behaviors under different control parameters. Different types of co-storage discharge behaviors are detected by one-dimensional bifurcation diagram, maximum Lyapunov exponent spectrum, phase diagram and time series diagram.

[0016] The verification module for co-storage electrical behavior uses PSIM power electronics simulation software to build a simulation circuit for memristor tabu learning neurons and perform simulation verification.

[0017] According to another aspect of this application, an electronic device is also provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the steps of any of the above methods when executing the computer program.

[0018] According to another aspect of this application, a computer-readable storage medium is also provided, which stores a computer program, characterized in that the computer program, when executed by a processor, implements the steps of any of the above methods.

[0019] Compared with the prior art, the discharge behavior detection method provided by the present invention has high accuracy, is more comprehensive, and effectively captures co-storage discharge activity. Attached Figure Description

[0020] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention, making other features, objects, and beneficial effects of the invention more apparent. The illustrative embodiments of the invention illustrated in the drawings and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0021] Figure 1 It is a process for modeling, detecting and verifying memristor-prohibited learning neurons;

[0022] Figure 2 This is a verification diagram of memristor characteristic 1);

[0023] Figure 3 These are verification diagrams for memristor characteristics 2) and 3).

[0024] Figure 4 This is a classification diagram of the equilibrium state of memristor-resistant learning neurons;

[0025] Figure 5 It is a two-dimensional attractor basin related to the initial value of memristors;

[0026] Figure 6 It is a two-dimensional bifurcation graph induced by parameters with initial values ​​of (0,0,3);

[0027] Figure 7 It is a two-dimensional bifurcation graph induced by parameters with initial values ​​of (0,0,-3);

[0028] Figure 8 It is a superposition of a one-dimensional bifurcation graph and the maximum Lyapunov exponent spectrum;

[0029] Figure 9 It is a phase diagram showing the co-storage and energization behavior;

[0030] Figure 10 This is a time series diagram of the co-storage discharging behavior;

[0031] Figure 11 This is the circuit diagram of a magnetically controlled memristor;

[0032] Figure 12 This is the circuit diagram of a memristor forbidden learning neuron;

[0033] Figure 13 The results are PSIM simulations of the discharge behavior with an initial value of (0,0,3).

[0034] Figure 14 The results are PSIM simulations of the discharge behavior with initial values ​​of (0,0,-3). Detailed Implementation

[0035] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0036] The purpose of this invention is to overcome the shortcomings of existing research by using memristors to simulate electromagnetic radiation phenomena and coupling them with tabu learning neurons. A more comprehensive framework for modeling, analyzing, and verifying the co-storage current behavior of memristor tabu learning neurons is proposed. Taking tabu learning neurons as an example, firstly, a mathematical model of a magnetically controlled memristor is selected, and its characteristics are simulated through numerical simulation. A memristor tabu learning neuron model with electromagnetic radiation and external current is established by coupling the magnetically controlled memristor with the membrane potential of the tabu learning neuron. The stability of the AC equilibrium state is analyzed based on the eigenvalue type at the equilibrium point. The co-storage current behavior related to the initial values ​​and parameters of the memristor neuron is determined and verified using methods such as two-dimensional plane attraction basins, one-dimensional bifurcation diagrams, superimposed Lyapunov exponent spectra, timing diagrams, and phase diagrams. A circuit simulation diagram and circuit model are constructed based on the mathematical model of the memristor tabu learning neuron. Finally, the co-storage current behavior of the memristor tabu learning neuron is verified in PSIM circuit simulation software.

[0037] according to Figure 1 The memristor tabu learning neuron modeling, firing behavior detection, and verification process of the present invention includes the following specific implementations:

[0038] 1. Selection and verification of memristors: Select a magnetically controlled memristor and obtain its mathematical model. Solve the memristor model using the fourth-order Runge-Kutta method and verify the memristor characteristics using hysteresis loops.

[0039] (1.1) Ideal Model of Magnetically Controlled Memristor

[0040] A multistable magnetically controlled memristor is chosen to simulate the electromagnetic radiation actually received by a memristor-forbidden neuron. Its mathematical model is as follows:

[0041]

[0042] Among them, The magnetic flux of a magnetically controlled memristor is expressed in Weber (Wb). This is the VCR equation for a memristor, which conforms to Ohm's law; It is the current flowing through the memristor, measured in amperes (A). It is the voltage across the memristor, measured in volts (V). It is the memory index, in Siemens (S).

[0043] (1.2) Characteristic verification of magnetically controlled memristor

[0044] Based on the ideal magnetically controlled memristor model in (1.1), a sinusoidal AC voltage signal is given. The input voltage of the memristor is used to verify three typical characteristics of the memristor. Given the initial value of the memristor... Then, the classical fourth-order Runge-Kutta method was used to solve the differential equations of the memristor, and the numerical solution of the memristor state was obtained. .Will Substitution The voltage of the memristor is obtained. and current The relationship. 1) Maintain frequency Unchanged, Changed amplitude If the values ​​are 1 V, 1.5 V, and 2 V respectively, then the memristor... The relationship exhibits a contracting tight hysteresis loop passing through the origin, and the hysteresis loop at this point is as follows: Figure 2 As shown; 2) Keep amplitude Unchanged, Changed frequency For Hz values ​​of 0.4 Hz, 0.8 Hz, and 1.6 Hz, the sidelobe area of ​​the tight hysteresis loop of the memristor increases with... 3) Decrease as it increases; The verification shows that the tight hysteresis loop of the memristor will eventually contract to a straight line passing through the origin, at which point the hysteresis loop is as follows: Figure 3 As shown.

[0045] 2. Memristor tabu learning neuron modeling: Select a suitable tabu learning neuron model based on the characteristics of the tabu search algorithm. In order to characterize the electromagnetic radiation effect of the tabu learning neuron, combine it with the magnetically controlled memristor in (1.1) to form a memristor tabu learning neuron. Step (2) is the basis of this invention.

[0046] (2.1) Mathematical Model of Taboo Learning Neuron

[0047] Traditional tabu learning neurons possess complex bounded nonlinear activation functions, making it difficult to detect their firing behavior through theoretical analysis and circuit design. A simplified linear model of a single tabu learning neuron, evolved from the Hopfield neural network, is commonly used, and its mathematical expression is as follows:

[0048]

[0049] in, This represents the membrane potential of the neuron. The learning state of a forbidden learning neuron; , , , It is a positive control parameter; The self-connection synaptic weights of neurons; It is a simple activation function that stimulates action potentials to induce neuronal firing. This is an external current stimulation signal.

[0050] (2.2) Mathematical Model of Memristor Taboo Learning Neuron

[0051] To investigate the complex firing behavior of forbidden learning neurons under the combined influence of electromagnetic radiation and applied current, the magnetically controlled memristor from (1.1) is used to simulate the electromagnetic radiation experienced by the forbidden learning neurons in (2.1). The memristor acts on the neuron in the form of electrical stimulation, which is related to the membrane potential. By coupling the neurons, the mathematical model of the memristor tabu learning neuron is obtained as follows:

[0052]

[0053] To facilitate the introduction of the invention, let , and Therefore, the aforementioned memristor taboo learning neurons can be transformed into:

[0054]

[0055] in, The intensity of electromagnetic radiation; external current is defined as... ,and and If the amplitude and frequency of the external current are respectively, then the memristor tabu learning neuron is a three-dimensional non-autonomous nonlinear system. Under the combined influence of the external current and the electromagnetic radiation current, the memristor neuron can be induced to produce rich firing behaviors.

[0056] It is worth noting the activation function exist The discontinuity at a certain point leads to a singular derivative; therefore, in stability analysis and calculation of the Li's exponent, a different approach is often used. replace ,and .

[0057] 3. Equilibrium analysis: Set the right-hand equation of the memristor forbidden learning neuron in (2.2) to 0 to obtain the equilibrium point. Moreover, external current Over time Changes with change, therefore It is the equilibrium point of exchange, which can be represented as ,as follows

[0058]

[0059] in, The following conditions must be met:

[0060]

[0061] Communication balance point Jacobian matrix at the location for:

[0062]

[0063] in, , and Characteristic equation for:

[0064]

[0065] in:

[0066]

[0067] Based on the characteristic roots of the characteristic equation Can The system is classified into eight types: Hopf bifurcation point (HBP), folded bifurcation point (FBP), unstable saddle point (USP), unstable node (UNP), unstable coking point (UNF), unstable saddle-coking point (USF), stable node (SNP), and stable coking point (SNF). The system parameters are selected as follows: , , , , , and When time And when the step size is 0.0001 s, Figure 4 Success over time Equilibrium state of changing membrane potential A detailed classification was performed. Different types of equilibrium points can induce different firing behaviors in memristor taboo learning neurons.

[0068] 4. Determine the parameter domain of the discharge behavior: This includes the kinetic behavior parameter domain related to the initial value of the memristor and the amplitude of the external current. From the parameter domain, the different discharge behaviors of the memristor taboo learning neurons can be clearly identified.

[0069] (4.1) Parameter domain of memristor initial value induced discharge behavior

[0070] Memristor tabu learning neurons are highly sensitive to initial values. The parameters of the memristor tabu learning neuron are set as follows: , , , , , , , and the initial value of the system is ( ,0, (Using memristor initial values) and initial value of membrane potential To control parameters and select and The range of variation is [-10, 10]. The initial value is decomposed into a combination of a series of control parameters. For each set of parameters, the numerical integral solution of the memristor taboo neuron is obtained using the fourth-order Runge-Kutta method. The number of cycles is found using the local maximum judgment method and the cycle number judgment method, thereby drawing a two-dimensional attraction basin and determining the parameter domains for different firing behaviors.

[0071] The local maximum determination method is described as follows:

[0072] Given a set dimensional membrane potential time series To ignore the effects of transient behavior, we take the integer value at the midpoint. .definition , and ,Will , and Compare, if and , Then at this moment A local maximum, i.e., a bifurcation point, is denoted as: .

[0073] The description of the cycle number determination method is as follows:

[0074] Based on the maximum value judgment method, assuming the membrane potential... If the local maximum is an m-dimensional vector, then choose... As data used to calculate the number of cycles, This represents the number of cycles to be calculated. At this point, Average score A matrix with 5 rows and 5 columns :

[0075]

[0076] Calculate matrix Average value of each row and expanded to A matrix of averages in rows of 5 columns :

[0077]

[0078] To obtain the number of periods, different number of periods are calculated. Down variance ,Right now: .when When, then the time series The number of periods is When the time series diverges, the definition is... .

[0079] The numerical solution of the memristor taboo learning neuron is analyzed using the local maximum value judgment method and the cycle number judgment method to find the cycle number, thereby drawing a two-dimensional attraction basin related to the initial memristor value, such as... Figure 5 As shown. Figure 5 The parameter domains of three different firing behaviors of memristor taboo learning neurons—cycle 1, cycle 8, and divergent—were clearly defined.

[0080] (4.2) Parameter domain of parameter-induced discharge behavior

[0081] Changes in parameters within memristor-prohibited learning neurons can also induce various firing behaviors, including selection of electromagnetic radiation intensity. and the amplitude of external input current As the primary control parameter, when and At the same time, under different control parameters Under the control of [the specific method], the numerical integral solution of the memristor taboo neuron was obtained using the fourth-order Runge-Kutta method. The local maximum and cycle number methods were used to analyze the numerical integral solution, find the cycle number, and thus draw a two-dimensional bifurcation diagram, which was marked with different colors. Figure 6 and Figure 7The two-dimensional bifurcation diagrams induced by parameters with initial values ​​of (0,0,3) and (0,0,-3) are shown respectively. The parameter domain of the parameter-induced firing behavior of memristor taboo learning neurons is clearly revealed by the color annotation.

[0082] 5. Detection of co-storage electrical behavior: based on Figure 5 , Figure 6 or Figure 7 The defined parameter domain is illustrated using the parameter domain of parameter-induced co-storage electrical behavior as an example, namely... Figure 6 and Figure 7 Fixed electromagnetic radiation intensity With external current intensity For bifurcation parameters and membrane potential Using the fourth-order Runge-Kutta method to solve equation (2.2) as the analysis object, various discharge behaviors with different periods can be obtained. Different control parameters... The co-storage discharge behavior is obtained by superimposing the discharge behaviors induced by the two initial conditions. Different types of co-storage discharge behavior can be detected by using one-dimensional bifurcation diagrams, maximum Lyapunov exponent spectra, phase diagrams, and time series diagrams.

[0083] (5.1) Superposition of one-dimensional bifurcation diagram and maximum Lyapunov exponent spectrum

[0084] external current intensity Under the control of [the system], a series of bifurcation points can be obtained by using the maximum value judgment method in section (4.1). Plot the bifurcation parameters with initial values ​​of (0,0,3) and (0,0,-3) at the bifurcation point. A scatter plot showing the changes is a one-dimensional bifurcation plot, such as... Figure 8 The image below illustrates this. The distribution of points in the bifurcation diagram distinguishes between periodic and chaotic behaviors. Periodic behavior exhibits bifurcation diagrams composed of single or multiple line segments, while chaotic behavior shows a chaotic and densely packed array of points.

[0085] The maximum Lyapunov exponent is often used as a quantitative analysis method to determine whether a system is chaotic or hyperchaotic. For a first-order nonlinear system... , For state variables, It is a Jacobian matrix. The eigenvalues ​​can be used to determine whether adjacent points in the system elongate or compress. The average elongation and compression rates of each point are then calculated; this average is called the Lyapunov exponent. The typical Wolf method is used to calculate the memristor tabu learning neuron's response to external current intensity with initial values ​​of (0,0,3) and (0,0,-3). The spectrum of the largest Lyapunov exponent changes, such as Figure 8 As shown in the figure above, if the maximum Lyapunov exponent, which varies with the control parameters, is positive, the system is chaotic; if it is negative, the system is stable.

[0086] The co-storage discharging behavior can be determined by superimposing the one-dimensional bifurcation graphs with initial values ​​of (0,0,3) and (0,0,-3) with the maximum Lyapunov exponent spectrum, such as... Figure 8 As shown. By Figure 8 As shown in the bifurcation diagram below, the single-cycle stable discharge corresponding to the initial value (0,0,-3) coexists with the multi-cycle limiting cycle, quasi-periodic, or chaotic attractor mixed discharge modes corresponding to the initial value (0,0,3), forming a coexisting bistable discharge behavior. Similarly, from... Figure 8 As shown in the above figure, (0,0,-3) corresponds to a smooth curve where the maximum Lyapunov exponent is less than 0, while (0,0,3) corresponds to an oscillating curve showing the change in the maximum Lyapunov exponent, indicating that neurons exhibit coexisting bistable firing behaviors. Further analysis reveals... At that time, due to and Therefore, at this time, the memristor forbidden learning neuron produces chaotic and single-cycle coexisting bistable discharge behavior.

[0087] (5.2) Phase diagrams and time series diagrams characterize co-storage discharging behavior

[0088] The membrane potential was obtained by solving the memristor tabu learning neuron model (2.2) using the fourth-order Runge-Kutta method. Learning status and memristor flux The numerical solution. By using ( , The phase trajectories of the plane are plotted as phase diagrams, such as... Figure 9 As shown, the trajectory of memristor neurons in phase space can be intuitively displayed. Analysis Figure 9 It can be seen that the trajectory with initial value (0,0,-3) is a closed curve, so the memristor tabu learning neuron fires periodically; the trajectory with initial value (0,0,3) is never closed and never intersects, so the memristor tabu learning neuron fires chaotically.

[0089] To reflect the firing behavior of memristor forbidden learning neurons over time Plot the numerical solution of the membrane potential under varying conditions. With time Time series plots between, such as Figure 10 As shown. Figure 10 In the above figure, the initial value is (0,0,3) and the discharge state curve is chaotic, which means that the memristor forbidden learning neuron is undergoing chaotic discharge. Figure 10In the figure below, the initial value is (0,0,-3). The curve shows periodic fluctuations, which means that the memristor forbidden learning neuron is firing periodically. Figure 9 and Figure 10 The phase diagrams and time series diagrams vividly characterize the co-storage behavior of memristor taboo learning neurons.

[0090] 6. Verification of co-storage electrical behavior: Using PSIM power electronics simulation software, a simulation circuit of the memristor tabu learning neuron was built for simulation verification. 1) First, based on the mathematical models in (1.1) and (2.2), the circuit schematic diagrams of the magnetically controlled memristor and the memristor tabu learning neuron were designed, such as... Figure 11 and Figure 12 As shown; 2) Derive the circuit equations of the magnetically controlled memristor and the memristor tabu learning neuron based on the circuit schematic; the circuit equation of the magnetically controlled memristor is:

[0091]

[0092] in, represents the output voltage of the integrator, and represents the internal variable of the memristor simulator; For memory derivative value; is the integration time constant.

[0093] The circuit equation for a memristor-forbidden learning neuron is:

[0094]

[0095] 3) Select circuit components based on the system parameters in (1.1) and (2.2) to... Taking the coexistence of chaos and single-cycle discharge as an example, circuit simulation is performed, with the time constant taken as... According to Circuit parameter conversion is performed to obtain , , , , and It should be noted that operational amplifiers and The saturation voltage is The remaining amplifiers use Voltage power supply.

[0096] 4) Set the simulation parameters in the PSIM software, run the circuit, and display the co-current discharge behavior through Simview, such as... Figure 13 and Figure 14 As shown. Figure 13 and Figure 14 Verified Figure 9The co-storage electrical behavior demonstrates that the co-storage electrical behavior of memristor tabu learning neurons does indeed exist, thus confirming that the method for modeling, detecting, and verifying the firing behavior of memristor tabu learning neurons using the present invention meets logical and practical requirements.

[0097] This application also provides an analysis device for the co-storage electrical behavior of memristor tabu learning neurons, comprising:

[0098] The memristor selection and verification module selects a magnetically controlled memristor and obtains a mathematical model. It then uses the fourth-order Runge-Kutta method to solve the mathematical model and verifies the characteristics of the magnetically controlled memristor through a hysteresis loop.

[0099] The memristor tabu learning neuron modeling module selects tabu learning neuron models based on the characteristics of the tabu search algorithm, and combines the tabu learning neuron models with magnetically controlled memristors to form memristor tabu learning neurons.

[0100] The communication equilibrium analysis module obtains equilibrium points based on the taboo learning neuron model and classifies the equilibrium points into Hopf bifurcation point type, and / or folded bifurcation point type, and / or unstable saddle point type, and / or unstable node type, and / or unstable foci type, and / or unstable saddle-foci type, and / or stable node type, and / or stable foci type.

[0101] The parameter domain determination module for discharge behavior determines the parameter domain of the dynamic behavior that is related to the initial value of the memristor and the amplitude of the external current.

[0102] The co-storage discharge behavior detection module, based on the parameter domain, can obtain various discharge behaviors with different periods by fixing one control parameter and changing another. The co-storage discharge behavior is obtained by superimposing two or more sets of initial value-induced discharge behaviors under different control parameters. Different types of co-storage discharge behaviors are detected by one-dimensional bifurcation diagram, maximum Lyapunov exponent spectrum, phase diagram and time series diagram.

[0103] The verification module for co-storage electrical behavior uses PSIM power electronics simulation software to build a simulation circuit for memristor tabu learning neurons and perform simulation verification.

[0104] This application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of any of the above methods.

[0105] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of any of the above methods.

[0106] Compared with the prior art, the discharge behavior detection method provided by the present invention has high accuracy, is more comprehensive, and effectively captures co-storage discharge activity.

[0107] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An analysis method of co-spiking behavior of a memristive Hebbian neuron, characterized by, The method comprises the following steps: S101: selecting a magnetically controlled memristor and obtaining a mathematical model, solving the mathematical model by using a fourth-order Runge-Kutta method, and verifying the characteristics of the magnetically controlled memristor through a magnetic hysteresis loop; S102: selecting a tabu learning neuron model according to the characteristics of a tabu search algorithm, and combining the tabu learning neuron model with the magnetically controlled memristor to form a memristor tabu learning neuron; S103: obtaining equilibrium points according to the tabu learning neuron model, and classifying the equilibrium points into a Hopf bifurcation point type, and / or a fold bifurcation point type, and / or an unstable saddle point type, and / or an unstable node type, and / or an unstable node focus type, and / or an unstable saddle focus type, and / or a stable node type, and / or a stable node focus type; S104: determining a parameter domain of a dynamic behavior related to a memristor initial value and related to an external current amplitude; S105: according to the parameter domain, a plurality of different periodic discharge behaviors can be obtained by fixing one control parameter and changing another control parameter; superimposing discharge behaviors induced by two or more groups of initial values under different control parameters to obtain coexistence discharge behaviors; and detecting different types of coexistence discharge behaviors through a one-dimensional bifurcation diagram, a maximum Lyapunov exponent spectrum, a phase diagram and a time series diagram; S106: using PSIM power electronic simulation software to build a simulation circuit of the memristor tabu learning neuron, and performing simulation verification; The method for selecting a magnetically controlled memristor and obtaining a mathematical model, solving the mathematical model by using a fourth-order Runge-Kutta method, and verifying the characteristics of the magnetically controlled memristor through a magnetic hysteresis loop comprises the following steps: S1011: selecting a magnetically controlled memristor with multi-stability to simulate electromagnetic radiation actually received by a memristor tabu neuron, and the mathematical model of the magnetically controlled memristor is as follows: wherein, wherein is the magnetic flux of the magnetically controlled memristor, in units of weber (Wb); is the VCR equation of the memristor, which complies with Ohm's law; is the current flowing through the memristor, in units of ampere (A); is the voltage across the memristor, in units of volt (V); is the memductance value, in units of siemens (S); S1012: Based on the mathematical model in S1011, using a given sinusoidal AC voltage signal... Using the input voltage of the memristor as the basis for verifying its three typical characteristics, given the initial value of the memristor... Then, the classical fourth-order Runge-Kutta method was used to solve the differential equations of the memristor, and the numerical solution of the memristor state was obtained. ,Will Substitution The voltage of the memristor is obtained. and current Relationship, maintain frequency Unchanged, Changed amplitude If the values ​​are 1 V, 1.5 V, and 2 V respectively, then the memristor... The relationship exhibits a contracting tight hysteresis loop passing through the origin; secondly, it maintains... amplitude Unchanged, Changed frequency The sidelobe area of ​​the memristor's tight hysteresis loop increases with 0.4 Hz, 0.8 Hz, and 1.6 Hz, respectively. It decreases as it increases; then, by further increasing... The study verified that the tight hysteresis loop of the memristor eventually contracts to a straight line passing through the origin, thus verifying the three typical characteristics of the memristor.

2. The method of claim 1, wherein the method is performed by a computer system. The method for selecting a tabu learning neuron model according to the characteristics of a tabu search algorithm, and combining the tabu learning neuron model with the magnetically controlled memristor to form a memristor tabu learning neuron comprises the following steps: S1021: determining a tabu learning single neuron linear model, and the mathematical expression is as follows: wherein, is the membrane potential of a neuron; is the learning state of a forbidden learning neuron; , , , is a positive control parameter; is the self-connection synapse weight of a neuron; is a simple activation function; is an external current stimulation signal; S1022: The electromagnetic radiation received by the forbidden learning neuron in step S1021 is simulated by the magnetically controlled memristor in step S1011; the memristor acts on the neuron in the form of electrical stimulation, which is related to the membrane potential. By coupling the neurons, the mathematical model of the memristor tabu learning neuron is obtained as follows: When , and , the above mentioned memristive Hebbian learning neuron is transformed into: where is the intensity of the electromagnetic radiation; the external current is , and and are the amplitude and frequency of the external current, respectively; the activation function is discontinuous at , which leads to its derivative being singular, and in the stability analysis and computation of the Lyapunov exponents, is used instead of , and .

3. The method of claim 2, wherein the method is performed by a computer system. The method for obtaining equilibrium points according to the tabu learning neuron model, and classifying the equilibrium points into a Hopf bifurcation point type, and / or a fold bifurcation point type, and / or an unstable saddle point type, and / or an unstable node type, and / or an unstable node focus type, and / or an unstable saddle focus type, and / or a stable node type, and / or a stable node focus type comprises the following steps: S1031: When the right end equation of the memristive taboo learning neuron in S1022 is 0, the equilibrium point is solved external current changes over time, changes over time, is the alternating equilibrium point, expressed as as follows wherein satisfies: AC equilibrium point the Jacobian matrix at the equilibrium point is: wherein , and , the characteristic equation is: Wherein: Based on the characteristic roots of the characteristic equation Can The system is classified into eight types: Hopf bifurcation point (HBP), folded bifurcation point (FBP), unstable saddle point (USP), unstable node (UNP), unstable coking point (UNF), unstable saddle-coking point (USF), stable node (SNP), and stable coking point (SNF). The selected system parameters are... , , , , , and ,time And the step size is 0.0001 s.

4. The method of claim 3, wherein the method is performed by a computer system. The method for determining a parameter domain of a dynamic behavior related to a memristor initial value and related to an external current amplitude comprises the following steps: S1041: When the memristive taboo learning neuron parameters are , , , , , , , the system initial value is ( ,0, ), the memristor initial value and the membrane potential initial value are control parameters, and the variation ranges of and are both [-10, 10], the initial value is decomposed into a series of combinations of control parameters , under each group of parameters, the numerical integral solution of the memristive taboo neuron is obtained by using the fourth-order Runge-Kutta method, the periodic number is found by using the local maximum value judgment method and the period number judgment method, and then a two-dimensional attractor basin is drawn to realize the determination of the parameter domain of different discharge behaviors. The local maximum value judgment method is as follows: Given a set of The membrane potential time series of a ventricle In order to ignore the influence of transient behavior, take the integer , define , and Compare , and If and , , the at this moment is a local maximum, that is, a bifurcation point, denoted as ; The cycle number judgment method is as follows: On the basis of the maximum value judging method, assuming that the local maximum value of the membrane potential is an m-dimensional vector, then the data is selected as the data of the number of calculation periods, wherein represents the number of calculation periods required, at this time is evenly divided into a matrix of rows and 5 columns : matrix of calculations average value of each row and extended to matrix of average values of row 5 columns : To get the number of periods, calculate the different number of periods The variance of the lower That is: When , then the time series has a number of periods When the time series diverges, ;​ S1042: intensity of electromagnetic radiation and the amplitude of the external input current As the main control parameter, when and Under the regulation of different control parameters , the numerical integral solution of the memristive taboo neuron is obtained by using the fourth-order Runge-Kutta method, the numerical integral solution is analyzed by using the local maximum value judgment method and the period number judgment method, and the period number is found, so as to draw a two-dimensional bifurcation diagram.

5. The method of claim 4, wherein the method is performed by a computer system. According to the parameter domain, a plurality of different periodic discharge behaviors can be obtained by fixing one control parameter and changing another control parameter; superimposing discharge behaviors induced by two or more groups of initial values under different control parameters to obtain coexistence discharge behaviors; The method for detecting different types of coexistence discharge behaviors through a one-dimensional bifurcation diagram, a maximum Lyapunov exponent spectrum, a phase diagram and a time series diagram comprises the following steps: S1051: Under external current intensity Under the control of [the system], the bifurcation point obtained by the maximum value judgment method in step S1041 is [the point]. Plot the bifurcation parameters with initial values ​​of (0,0,3) and (0,0,-3). The changing scatter plot is a one-dimensional bifurcation plot. Different periodic and chaotic behaviors are distinguished based on the point distribution in this one-dimensional bifurcation plot. The typical Wolf method is used to calculate the changes in external current intensity of the memristor tabu learning neuron with initial values ​​of (0,0,3) and (0,0,-3). The spectrum of the largest changing Lyapunov exponents; The coexisting discharge behavior is determined by superimposing the one-dimensional bifurcation diagram and the maximum Lyapunov exponent spectrum with initial values of (0, 0, 3) and (0, 0, -3); S1052: the numerical solution of the membrane potential of the memristive Hebbian learning neuron model in step S1022 is obtained by using the fourth-order Runge-Kutta method , learning state and the numerical solution of the memristive magnetic flux , the running track of the memristive neuron in the phase space is shown by using the phase trajectory of the phase trajectory of the plane , ).

6. The method of claim 5, wherein the method is performed by a computer system. The simulation circuit of the memristor tabu learning neuron is built by using PSIM power electronic simulation software, and simulation verification is performed, including: S1061: circuit diagrams of the magnetically controlled memristor and the memristor tabu learning neuron are designed according to the mathematical model in steps S1011 and S1022; S1062: circuit equations of the magnetically controlled memristor and the memristor tabu learning neuron are derived according to the circuit diagrams, the circuit equation of the magnetically controlled memristor is: wherein is the output voltage of the integrator and represents an internal variable of the memristor emulator; is the flux value; is the integration time constant, The circuit equation of the memristor tabu learning neuron is: S1063: Selecting circuit elements according to the system parameters in steps S1011 and S1022, and when coexisting chaos and single-period discharge circuit simulation when the time constant is , the circuit parameter conversion is carried out according to , , , , , and , the saturation voltage of the operational amplifier and is , and the remaining amplifiers are powered by voltage, S1064: simulation parameters in the PSIM software are set, the circuit is run, and the coexisting discharge behavior is displayed through Simview. 7.An analysis device of coexistence of spiking behavior of a memristive Hebbian neuron, characterized in that, Including: The selection and verification module of the memristor selects the magnetically controlled memristor and obtains a mathematical model, solves the mathematical model by using the fourth-order Runge-Kutta method, and verifies the characteristics of the magnetically controlled memristor through a hysteresis loop; The memristor tabu learning neuron modeling module selects a tabu learning neuron model according to the characteristics of the tabu search algorithm, and the tabu learning neuron model is combined with the magnetically controlled memristor to form a memristor tabu learning neuron; The AC equilibrium state analysis module obtains equilibrium points according to the tabu learning neuron model, and divides the equilibrium points into Hopf bifurcation point types, and / or fold bifurcation point types, and / or unstable saddle point types, and / or unstable node types, and / or unstable node focus types, and / or unstable saddle focus types, and / or stable node types, and / or stable node focus types; The parameter domain determination module of the discharge behavior determines the parameter domain of the dynamic behavior related to the initial value of the memristor and related to the external current amplitude; The detection module of the coexisting discharge behavior obtains a plurality of discharge behaviors with different periods by fixing one control parameter and changing another control parameter according to the parameter domain; and superimposes discharge behaviors induced by two or more groups of initial values under different control parameters to obtain the coexisting discharge behavior; Different types of coexisting discharge behaviors are detected through one-dimensional bifurcation diagrams, maximum Lyapunov exponent spectra, phase diagrams, and time series diagrams; The verification module of the coexisting discharge behavior builds a simulation circuit of the memristor tabu learning neuron by using PSIM power electronic simulation software, and performs simulation verification; The selection and verification module of the magnetically controlled memristor includes: S1011: a magnetically controlled memristor with multiple steady states is selected to simulate the electromagnetic radiation actually received by the memristor tabu neuron, and a mathematical model of the magnetically controlled memristor is: wherein, wherein Φ is the magnetic flux of the magnetic controlled memristor, unit: weber (Wb); is the VCR equation of the memristor, which conforms to Ohm's law; is the current flowing through the memristor, unit: ampere (A); is the voltage across the memristor, unit: volt (V); is the memduct value, unit: siemens (S); S1012: According to the mathematical model in S1011, the AC voltage signal of the given sine is verified as the input voltage of the memristor The three typical characteristics of the memristor are verified as the input voltage of the memristor Then, the classical fourth-order Runge-Kutta method is used to solve the differential equation of the memristor to obtain the numerical solution of the state of the memristor , the relationship between the voltage and the current of the memristor is obtained by substituting into , the frequency of is kept unchanged, and the amplitude of is changed to 1 V, 1.5 V and 2 V respectively, the relationship of the memristor presents a tight hysteresis loop that shrinks over the origin; secondly, the amplitude of is kept unchanged, and the frequency of is changed to 0.4 Hz, 0.8 Hz and 1.6 Hz respectively, the sidelobe area of the tight hysteresis loop of the memristor decreases with the increase of ; then, by further increasing , it is verified that the tight hysteresis loop of the memristor eventually shrinks to a straight line over the origin, verifying the three typical characteristics of the memristor.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the steps of the method of any one of claims 1 to 6. The processor executes the computer program to implement the steps of the method of any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 8. The computer program, which when executed by the processor, implements the steps of the method as claimed in any one of claims 1 to 6.

Images