Precision control method for grating turntable of broadband ultrahigh-resolution spectrometer based on PSO-RBF neural network

Abstract

According to the broadband ultrahigh-resolution spectrometer grating turntable precision control method based on the PSO-RBF neural network provided by the invention, the PSO algorithm is taken as an efficient global optimization method, the PSO algorithm and the RBF neural network are organically combined, the PSO global optimization and the online learning ability of the RBF neural network are fused, and the precision of the grating turntable of the broadband ultrahigh-resolution spectrometer is improved. An intelligent control system suitable for precise control of the grating rotary table is constructed, effective compensation of the system for nonlinear characteristics and parameter uncertainty is achieved, the positioning precision and the anti-interference capacity of the system are improved, and the requirement for ultrahigh-resolution spectral measurement under the condition of multiple diffraction is met.

Description

Technical Field

[0001] This invention relates to the field of spectrometer grating turntable control technology, and in particular to a precision control method for a wide-band ultra-high resolution spectrometer grating turntable based on a PSO-RBF neural network. Background Technology

[0002] Wideband high-resolution spectrometers, as key optical measurement devices in modern scientific research, play an irreplaceable role in fields such as astronomical observation, atmospheric environmental monitoring, materials characterization, and biomedical analysis. To simultaneously achieve wideband coverage and high-resolution detection, a scanning mechanism in the spectrometer is typically used to perform spectral scanning through horizontal rotation. To achieve ultra-high resolution at the picometer to femtometer level, the diffraction order of the grating is changed by adjusting it in the vertical direction. This requires a high-precision two-axis turntable mechanism to control the grating scanning in two dimensions: the yaw axis handles a wide-range spectral scan to achieve wideband coverage; the pitch axis performs fine-tuning of minute angles to achieve ultra-high-resolution measurements. Through coordinated two-axis motion and a multi-diffraction optical path design, a combination of wideband coverage and ultra-high resolution is achieved.

[0003] With the increasing demands for spectral resolution, the requirements for grating angle positioning accuracy are becoming increasingly stringent, typically needing to reach the arc-second level or even higher. In the ultraviolet-visible band, a positioning error of 0.5 arc-seconds can lead to a wavelength measurement deviation of 2-3 nm, significantly affecting the accuracy of spectral peak identification. This extremely high precision requirement presents multiple challenges to the control system of the grating scanning mechanism: the system exhibits significant nonlinear characteristics, including variations in friction, the influence of gravity, and transmission backlash; parameter uncertainties are significant, such as load inertia variations of approximately ±15%, friction coefficient fluctuations of ±30%, and motor parameter variations of approximately ±8%; in addition, environmental factors such as temperature variations (each 1°C can produce a positioning error of 0.5-1 arc-seconds), external vibrations, and electromagnetic interference also have a significant impact on system performance.

[0004] The nonlinear characteristics of the grating turntable system are mainly manifested in frictional nonlinearity, the influence of gravity, and transmission backlash, leading to "creep-slide" phenomena and positional uncertainty in low-speed precision positioning. Traditional control methods show significant limitations in addressing these challenges. Classical PID control is widely used due to its simplicity and stability; however, fixed-parameter PID control struggles to adapt to system nonlinearity and parameter variations, often resulting in large steady-state errors and weak anti-interference capabilities in high-precision positioning tasks. Although modern self-tuning PID technology has made some progress, it still struggles to simultaneously meet the requirements of high-precision steady-state performance and good dynamic characteristics in high-precision scenarios such as spectrometer turntables.

[0005] In recent years, intelligent control technology has provided new solutions to such problems. Combining RBF neural networks with PID controllers can improve system adaptability through online parameter adjustment; optimization algorithms can be used to improve the convergence characteristics of neural networks. However, existing research has two limitations: first, the initial parameters of the RBF-PID controller are highly dependent, making it prone to getting trapped in local optima; second, the global optimization algorithm has high computational complexity and insufficient real-time performance. Summary of the Invention

[0006] The purpose of this invention is to propose a control system for a wide-band ultra-high resolution spectrometer grating turntable by organically combining the PSO algorithm with the RBF neural network. This system effectively compensates for nonlinear characteristics and parameter uncertainties, improves positioning accuracy and anti-interference capability, and meets the requirements of ultra-high resolution spectral measurement under multiple diffraction conditions.

[0007] To achieve the above objectives, this invention proposes a precision control method for a wide-band ultra-high resolution spectrometer grating turntable based on a PSO-RBF neural network, comprising the following steps:

[0008] S1: Construct a three-input, three-output RBF neural network model, using system error, error rate of change, and desired angle as inputs, to dynamically generate three parameters for the adaptive PID controller: K p K i K d ;

[0009] S2: The particle swarm optimization algorithm is used to globally optimize the parameters of the RBF neural network: center vector, width parameter and connection weights. A dynamic inertial weight strategy is introduced to balance the global exploration and local refinement capabilities.

[0010] S3: It adopts a two-stage control strategy of "offline optimization + online adjustment". In the offline stage, the initial parameters of the RBF network are iteratively optimized through the particle swarm optimization algorithm. In the online stage, real-time control is performed based on the optimized parameters and the adaptive learning capability is maintained.

[0011] S4: The optimal PID parameters are calculated in real time through the RBF network to achieve arcsecond-level precision positioning control of the grating turntable.

[0012] Furthermore, in S1, the RBF neural network includes: an input layer, a hidden layer, and an output layer;

[0013] The hidden layer uses 5 neurons and a Gaussian function as the radial basis function, which is expressed as: φ i (|XC i |)=exp(-|XC i |2 / (2σ i 2 ));

[0014] In the formula, X is the input vector. C i Let σ be the center vector of the i-th neuron. i This is the width parameter.

[0015] Furthermore, in S2, the dynamic inertia weight strategy of the particle swarm optimization algorithm is: w = w max -(w max -w min ) / K max ×k where w max =0.9, w min =0.4, K max is the maximum number of iterations, and k is the current number of iterations.

[0016] Furthermore, in S3, the particle swarm optimization algorithm introduces a mutation operation: when the swarm optimal solution does not improve significantly after 10 consecutive iterations, a small random perturbation is applied to the global optimal position to avoid premature convergence of the algorithm.

[0017] Furthermore, in S3, the objective function of the particle swarm optimization algorithm is: Where e(t) is the systematic error, Let u(t) be the error rate of change, u(t) be the control output, and α1, α2, α3 and α4 be weighting coefficients, which correspond to the transient error, dynamic characteristics, control energy consumption and steady-state accuracy requirements of the system, respectively.

[0018] Furthermore, the mathematical expression for the adaptive PID controller is: u(t) = K p (X)e(t)+K i (X)∫e(τ)dτ+K d (X)de(t) / dt where, K p (X), K i (X) and K d (X) represents the adaptive PID parameters output by the RBF network.

[0019] Furthermore, the parameter adjustment range of the adaptive PID controller is: K p (X)=K p0 +ΔK p ·f p (X), K i (X)=K i0 +ΔK i ·f i (X), K d (X)=K d0 +ΔK d ·f d (X);

[0020] In the formula, K p0 ,K i0 ,K d0 For the initial PID parameters, ΔK p ,ΔK i, ΔK d To adjust the range, f p (X),f i (X),f d (X) is the normalized output of the RBF network, with a value range of [-1, 1].

[0021] Furthermore, the spectrometer grating stage adopts a "yaw-pitch" two-axis structure. The yaw axis is responsible for large-range spectral scanning ±30°, and the pitch axis is responsible for fine adjustment of small angles (±10°). Ultra-high resolution spectral measurement is achieved by changing the diffraction number of the grating.

[0022] Furthermore, the drive system of the grating turntable adopts a combination of DC servo motor and harmonic reducer. The pitch axis uses a harmonic reducer with a reduction ratio of n=20:1, and the position detection uses a high-precision magnetoelectric encoder with a resolution of 0.001°.

[0023] Compared with the prior art, the advantages of the present invention are:

[0024] 1. This invention is based on the Particle Swarm Optimization (PSO) algorithm as an efficient global optimization method. It organically combines the PSO algorithm with the RBF neural network, integrating the online learning capabilities of PSO global optimization and RBF neural network to construct an intelligent control system suitable for the precision control of grating turntable. This enables the system to effectively compensate for nonlinear characteristics and parameter uncertainties, improve the system's positioning accuracy and anti-interference capability, and meet the requirements of ultra-high resolution spectral measurement under multiple diffraction conditions.

[0025] 2. This invention, through its "global optimization + local adaptation" dual-layer optimization mechanism, successfully solves the problems of system nonlinearity and parameter uncertainty that are difficult to handle by traditional control methods. It provides an effective technical solution for the arcsecond-level precise positioning of the grating turntable of a wide-band ultra-high resolution spectrometer and has broad application prospects.

[0026] 3. Based on the PSO-RBF neural network of this invention, the system positioning accuracy is improved to 0.7-1.0 arcseconds, which is 76% higher than that of the traditional PID controller, meeting the picometer to femtometer resolution requirements under multiple diffraction conditions of the spectrometer; the step response rise time is shortened by 40-44%, and the settling time is significantly reduced; in high-frequency tracking tasks, the phase lag is reduced by 73.4%, and the tracking error is reduced by 71.2%; it can still maintain stable high-precision control performance under the condition of ±30% variation of system parameters, and the interference recovery time is shortened by about 50%, which greatly improves the system's adaptability; the interference suppression degree is improved by 5.76dB, and the impact of interference on steady-state accuracy is reduced by 80.3%, ensuring stable operation in the actual environment. Attached Figure Description

[0027] Figure 1 This is a block diagram of the PSO-RBF-PID control system in Embodiment 1 of the present invention;

[0028] Figure 2 This is a schematic diagram of the RBF neural network structure in Embodiment 1 of the present invention;

[0029] Figure 3 The flowchart of the PSO-RBF-PID control algorithm in Embodiment 1 of the present invention is shown below.

[0030] Figure 4 This is a performance comparison chart of different control methods in step response testing according to Embodiment 2 of the present invention, wherein... Figure 4 (a) Comparison of the responses of the three controllers to a 1.1° step signal. Figure 4 (b) is the adaptive adjustment curve of the PSO-RBF-PID controller parameters;

[0031] Figure 5 This is a performance comparison chart of different control methods in sinusoidal tracking tests in Embodiment 2 of the present invention, wherein... Figure 5 (a) Waveform tracking comparison in high-frequency mode (5Hz), Figure 5 (b) Comparison of tracking errors in high-frequency mode Figure 5 (c) Waveform tracking comparison in low-frequency mode (0.5Hz), Figure 5 (d) Comparison of tracking errors in low-frequency mode;

[0032] Figure 6 This is a comparison chart of the response characteristics of different control methods under disturbance conditions in Embodiment 2 of the present invention, wherein... Figure 6 (a) Comparison of system responses under step disturbance. Figure 6 (b) shows the comparison of error curves under step disturbance. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be further described below.

[0034] Example 1:

[0035] This invention provides a precise control method for a wide-band ultra-high resolution spectrometer grating stage based on a PSO-RBF neural network, aiming to achieve arcsecond-level precise positioning of the grating stage and meet the measurement requirements of picometer to femtometer-level spectral resolution under multiple diffraction conditions. Figure 1 As shown, the PSO-RBF-PID control system of the present invention includes a desired angle input module, an error calculation module, a PSO-optimized RBF neural network module, an adaptive PID controller module, a two-axis rotary table actuator, and a feedback measurement module.

[0036] The spectrometer's two-axis turntable employs a "yaw-pitch" configuration to achieve precise rotational positioning of the grating around the Z-axis (yaw) and X-axis (pitch). The yaw axis handles large-range spectral scanning (±30°), while the pitch axis performs fine-tuning of small angles (±10°). The coordinated movement of the two axes constitutes a precise grating positioning system. The experiment used a grating measuring 20cm × 8cm and weighing approximately 0.32kg, mounted on the pitch axis. The two-axis turntable's drive system uses a combination of a DC servo motor and a harmonic reducer, featuring low backlash and high rigidity. The pitch axis uses a harmonic reducer with a reduction ratio of n = 20:1, and the transmission relationship between the motor and the load satisfies the following equation:

[0037] θ L =θ m / (n·η),ω L =ω m / n,T L =n·η·T m ;

[0038] In the formula, θ L and θ m ω represents the angular displacement at the load end and the motor end, respectively. L and ω m T represents the angular velocity at the load end and the motor end, respectively. L and T m These are the torques at the load end and the motor end, respectively; n is the reduction ratio; and η is the transmission efficiency (approximately 0.85).

[0039] Position detection employs a high-precision magnetoelectric encoder with a resolution of 0.001°, and its measurement equation can be expressed as:

[0040] θ measured =θ actual +δ backlash +δ sensor ;

[0041] In the formula, δ backlash For transmission clearance error, δ sensor Both sensor error and sensor error affect the control accuracy of the system.

[0042] like Figure 2 As shown, the RBF neural network adopts a three-layer structure: the input layer receives the system error e(t) and the error rate of change. and the expected angle θ r (t); the hidden layer uses 5 neurons and a Gaussian function as the radial basis function; the output layer generates three adaptive PID parameters K. p K i and K d The mathematical expression for an RBF network is:

[0043] y j =Σ(w ji ·φ i (|XC i |))+b j .

[0044] In the formula, X is the input vector. C i Let φ be the center vector of the i-th hidden neuron. i For radial basis functions, w ji b represents the connection weights from the hidden layer to the output layer. j This is the threshold for the output layer.

[0045] Hidden layer neurons use a Gaussian function:

[0046] φ i (|XC i |)=exp(-|XC i |2 / (2σ i 2 )).

[0047] The mathematical expression for the PID controller based on RBF network is:

[0048] u(t) = K p (X)e(t)+K i (X)∫e(τ)dτ+K d (X)de(t) / dt.

[0049] To ensure the stability of the control system, the parameter variation range is designed as follows:

[0050] K p (X)=K p0 +ΔK p ·f p (X), K i(X)=K i0 +ΔK i ·f i (X), K d (X)=K d0 +ΔK d ·f d (X).

[0051] In the formula, K p0 =4.8, K i0 =22.5, K d0 =0.28 is the initial PID parameter, ΔK p ΔK i ΔK d To adjust the range, f p (X), f i (X), f d (X) is the normalized output of the RBF network, with a value range of [-1, 1].

[0052] The PSO algorithm is used to optimize the parameters of an RBF network, including the center vector C. i Width parameter σ i and connection weight w i The equations for updating the particle's position and velocity are:

[0053] v i (k+1) =w·v i k +c1·r1·(p i k -x i k )+c2·r2·(p g k -x i k ), x i (k+1) =x i k +v i (k+1) ,

[0054] In the formula, v i k and x i k Let p be the velocity and position vectors of the i-th particle in the k-th iteration. i k For the individual's historical optimal position, p g k The global optimal position of the group is w, the inertia weight is c1 and c2 are acceleration constants (both set to 2.0), and r1 and r2 are random numbers in the interval [0,1].

[0055] To balance the algorithm's global exploration and local refinement capabilities, a dynamic inertia weighting strategy is adopted:

[0056] w = w max -(w max -w min ) / K max ×k.

[0057] In the formula, w max =0.9, w min =0.4, K max Let k be the maximum number of iterations (set to 100), and k be the current iteration number. A mutation operation is introduced: when the population optimal solution shows no significant improvement after 10 consecutive iterations, a small random perturbation is applied to the global optimal position to prevent premature convergence of the algorithm.

[0058] The objective function for PSO optimization is designed as follows:

[0059]

[0060] In the formula, e(t) represents the systematic error. Let u(t) be the error rate of change, α1 = 0.5, α2 = 0.2, α3 = 0.1, and α4 = 0.2 be the weighting coefficients, which correspond to the transient error, dynamic characteristics, control energy consumption, and steady-state accuracy requirements of the system, respectively.

[0061] like Figure 3 As shown, the control algorithm of this invention adopts a two-stage strategy of "offline optimization + online adjustment". The offline optimization stage establishes a refined simulation platform that includes a friction model, transmission clearance, and environmental disturbances. The friction model comprises three parts: static friction, Coulomb friction, and viscous friction, and its mathematical model is as follows:

[0062] T f (ω)=T s ·sign(ω)+T c ·sign(ω)+B v ·ω.

[0063] In the formula, the parameters are T s = 0.012 N·m, T c =0.008 N·m, B v = 3.4 × 10⁻⁴ N·m·s / rad. The gravity effect model is: T g (θ)=m gl sinθ≈0.21 sinθN·m.

[0064] The transmission clearance model is a piecewise function:

[0065] θ L ={θm / n-δ,θ m >nδ,0,-nδ≤θ m ≤nδ,θ m / n+δ,θ m

[0066] <-nδ}.

[0067] Based on the above model, the PSO algorithm is used to globally optimize the RBF network parameters. The PSO algorithm parameters are designed and the particle swarm is initialized. The control performance is evaluated through test scenarios such as step response, sinusoidal tracking and anti-interference. The RBF network parameters are iteratively optimized until the convergence condition is met, and the globally optimal parameter combination is saved.

[0068] During the online control phase, the initial parameters obtained from PSO optimization are loaded, and the system error e(t) and error change rate are collected. and the expected angle θ r (t) The optimal PID parameters are calculated using an RBF network, and the control output is adjusted in real time to compensate for system nonlinearity and parameter changes. During operation, the RBF network weights are further fine-tuned to achieve online learning. In the online learning phase, the RBF network weights are updated using an improved gradient descent method:

[0069]

[0070] In the formula, β = 0.8 is the momentum coefficient, η0 = 0.05 is the initial learning rate, and λ = 1.2 is the adaptive adjustment parameter. When the error is large, the learning rate is small to avoid drastic parameter changes; when the error decreases, the learning rate increases to accelerate the convergence process.

[0071] Example 2:

[0072] To verify the control performance of this invention, step response, sinusoidal tracking, and anti-interference tests were designed and compared with traditional PID and basic RBF-PID control methods. The tests were conducted using a simulation environment built on the MATLAB / Simulink platform, with a sampling frequency of 1000Hz and a time step of 0.001s. Figure 4 As shown, two amplitude step tests, 1.0° and 1.1°, were designed. Compared with traditional methods, the PSO-RBF-PID controller of this invention has a faster response speed and higher steady-state accuracy. In the 1.0° step test, the rise time was shortened to 30.3ms (a reduction of 41.4% compared with traditional PID), and the steady-state error was reduced to 0.7 arcseconds (an improvement of approximately 76%). Figure 4 (b) Demonstrating the adaptive adjustment process of PSO-RBF-PID controller parameters: initial proportional gain K in the response phase. p The integral gain K is rapidly increased to accelerate the system response; as the error decreases, the integral gain K... iand differential gain K d Gradually adjust to suppress oscillations; when approaching steady state, K i Reduce K to decrease overshoot. d Maintain an appropriate level to provide sufficient damping.

[0073] Figure 5 The results of sinusoidal tracking tests were presented, with two typical operating conditions designed: a high-frequency mode (5Hz, amplitude 0.2°) simulating rapid scanning, and a low-frequency mode (0.5Hz, amplitude 2°) simulating precise scanning. Under the high-frequency condition, the phase lag of the PSO-RBF-PID controller decreased by 73.4% (from 99.92° to 26.56°), and the root mean square error decreased by 71.2%. Error curve analysis shows that the maximum tracking error of the PSO-RBF-PID is approximately one-quarter that of the traditional PID, and the average error is reduced by about 70%. Under the low-frequency condition, the maximum error of the PSO-RBF-PID controller is approximately 0.12°, only 55% of that of the traditional PID (0.22°); the error distribution is more uniform, and the standard deviation is reduced by about 60%, indicating that it maintains consistent control accuracy throughout the entire scanning range. The PSO-RBF-PID controller's parameter adaptive adjustment mechanism exhibits different characteristics under different frequency operating conditions. Under high-frequency operating conditions, the control parameters show large periodic adjustments, synchronized with the input signal frequency; under low-frequency operating conditions, the parameter adjustment amplitude is smaller, but it still maintains periodic changes.

[0074] Figure 6 The results of anti-interference tests were presented, with two typical test scenarios: step interference and continuous sinusoidal interference. In the step interference test, after the interference was applied, both the traditional PID and RBF-PID controllers showed significant deviations, with maximum deviations of 0.8044° and 0.6986°, respectively; while the maximum deviation of the PSO-RBF-PID controller was only 0.3643° (a reduction of approximately 54.7%), and the recovery time was approximately 100ms (a reduction of 28.6% compared to the traditional PID). In the continuous sinusoidal interference test, the PSO-RBF-PID controller showed the most significant attenuation of the 4Hz interference signal, with an interference suppression level of -25.31dB, an improvement of 5.76dB compared to the traditional PID (-19.55dB), equivalent to an interference amplitude attenuation of approximately 48.5%. Regarding the impact of disturbances on steady-state accuracy, the PSO-RBF-PID controller achieved 215.5131 arc-sec, which is 80.3% lower than the traditional PID (1093.584 arc-sec), indicating a significant improvement in its ability to suppress continuous disturbances.

[0075] In the system parameter variation test, two types of parameter disturbance tests were designed: system gain variation (±30%) and load inertia variation (±20%). Sudden parameter changes were performed under stable system operation (t=1s and t=3s) to evaluate the adaptive performance of the control algorithm. The results show that the PSO-RBF-PID controller exhibits optimal adaptive performance under all parameter variation conditions. Its integral absolute error (IAE) is reduced by an average of approximately 50% compared to the traditional PID controller, and its disturbance recovery time is only 1 / 3 to 1 / 2 of that of the traditional PID controller. Especially under the most challenging condition of a 30% gain reduction, the performance advantage of the PSO-RBF-PID is even more significant, demonstrating its substantial reduction in sensitivity to system parameters.

[0076] In practical spectral analysis applications, the PSO-RBF-PID control method of this invention has been successfully applied to the grating control system of a wide-band (300-1100nm) high-resolution spectrometer. The experiment used a 300 lines / mm grating, achieving wide-band coverage through the yaw axis (±30°), and finely adjusting the diffraction order through the pitch axis (±10°) to achieve ultra-high resolution measurements under multiple diffraction conditions. Based on the grating diffraction equation: d(sinα+sinβ)=mλ, where d is the grating constant, α is the incident angle, β is the diffraction angle, m is the diffraction order, λ is the wavelength, and θ is the grating angle. i With wavelength λ i The relationship is represented as: θ i =θ0+arcsin(mλ) i / d-sinα)-arcsin(mλ0 / d-sinα). The control system of this invention ensures the precise positioning of the grating under this nonlinear relationship, enabling the spectrometer to achieve a wavelength resolution of 0.01nm under multiple diffraction conditions, and successfully realizing ultra-high resolution spectral measurements across a wide band from ultraviolet to near-infrared.

[0077] A wideband ultra-high resolution spectrometer employing the PSO-RBF-PID control method of this invention has been successfully applied in the field of high-precision spectral analysis, significantly improving the accuracy and stability of spectral measurements. Compared with traditional control methods, this invention has significant technical advantages: PSO global optimization ensures that the initial parameters of the RBF network are in a near-global optimal state, avoiding getting trapped in local optima; the real-time adaptive mechanism dynamically adjusts the control parameters according to the system state, effectively compensating for the system's nonlinear characteristics and parameter uncertainties; the two-stage strategy of "offline optimization + online adjustment" balances computational efficiency and control performance. This invention not only provides key technical support for wideband ultra-high resolution spectrometers but can also be extended to related fields requiring arcsecond-level positioning accuracy, such as satellite attitude control, laser communication, high-precision radar antennas, and precision manufacturing.

[0078] The above are merely preferred embodiments of the present invention and do not constitute any limitation on the present invention. Any equivalent substitutions or modifications made by those skilled in the art to the technical solutions and content disclosed in the present invention without departing from the scope of the present invention shall be deemed to have remained within the scope of protection of the present invention.

Claims

1. A method for precise control of a grating turntable for a wide-band ultra-high resolution spectrometer based on a PSO-RBF neural network, characterized in that, Includes the following steps: S1: Construct a three-input, three-output RBF neural network model, using system error, error rate of change, and desired angle as inputs, to dynamically generate three parameters for the adaptive PID controller: K p K i K d ; S2: The particle swarm optimization algorithm is used to globally optimize the parameters of the RBF neural network: center vector, width parameter and connection weights. A dynamic inertial weight strategy is introduced to balance the global exploration and local refinement capabilities. S3: It adopts a two-stage control strategy of "offline optimization + online adjustment". In the offline stage, the initial parameters of the RBF network are iteratively optimized through the particle swarm optimization algorithm. In the online stage, real-time control is performed based on the optimized parameters and the adaptive learning capability is maintained. S4: The optimal PID parameters are calculated in real time through the RBF network to achieve arcsecond-level precision positioning control of the grating turntable.

2. The method for precise control of a wide-band ultra-high resolution spectrometer grating turntable based on a PSO-RBF neural network according to claim 1, characterized in that, In S1, the RBF neural network includes: an input layer, a hidden layer, and an output layer; The hidden layer uses 5 neurons and employs a Gaussian function as the radial basis function, the expression of which is: φ i (|XC i |)=exp(-|XC i |2 / (2σ i 2 )); In the formula, X is the input vector. C i Let σ be the center vector of the i-th neuron. i This is the width parameter.

3. The method for precise control of a wide-band ultra-high resolution spectrometer grating turntable based on a PSO-RBF neural network according to claim 1, characterized in that, In S2, the dynamic inertia weight strategy of the particle swarm optimization algorithm is: w = w max -(w max -w min ) / K max ×k where w max =0.9, w min =0.4, K max is the maximum number of iterations, and k is the current number of iterations.

4. The method for precise control of a wide-band ultra-high resolution spectrometer grating turntable based on a PSO-RBF neural network according to claim 3, characterized in that, In S3, the particle swarm optimization algorithm introduces a mutation operation: when the swarm optimal solution does not improve significantly after 10 consecutive iterations, a small random perturbation is performed on the global optimal position to avoid premature convergence of the algorithm.

5. The method for precise control of a wide-band ultra-high resolution spectrometer grating turntable based on a PSO-RBF neural network according to claim 1, characterized in that, In S3, the objective function of the particle swarm optimization algorithm is: Where e(t) is the systematic error, Let u(t) be the error rate of change, u(t) be the control output, and α1, α2, α3 and α4 be weighting coefficients, which correspond to the transient error, dynamic characteristics, control energy consumption and steady-state accuracy requirements of the system, respectively.

6. The method for precise control of a wide-band ultra-high resolution spectrometer grating turntable based on a PSO-RBF neural network according to claim 1, characterized in that, The mathematical expression for the adaptive PID controller is: u(t) = K p (X)e(t)+K i (X)∫e(τ)dτ+K d (X)de(t) / dt where, K p (X), K i (X) and K d (X) represents the adaptive PID parameters output by the RBF network.

7. The method for precise control of a wide-band ultra-high resolution spectrometer grating turntable based on a PSO-RBF neural network according to claim 6, characterized in that, The parameter adjustment range of the adaptive PID controller is: K p (X)=K p0 +ΔK p ·f p (X), K i (X)=K i0 +ΔK i ·f i (X), K d (X)=K d0 +ΔK d ·f d (X); In the formula, K p0 ,K i0 ,K d0 For the initial PID parameters, ΔK p ,ΔK i, ΔK d To adjust the range, f p (X),f i (X),f d (X) is the normalized output of the RBF network, with a value range of [-1, 1].

8. The method for precise control of a wide-band ultra-high resolution spectrometer grating turntable based on a PSO-RBF neural network according to claim 1, characterized in that, The spectrometer grating turntable adopts a "yaw-pitch" two-axis structure. The yaw axis is responsible for large-range spectral scanning ±30°, and the pitch axis is responsible for fine adjustment of small angles (±10°). Ultra-high resolution spectral measurement is achieved by changing the diffraction number of the grating.

9. The method for precise control of a wide-band ultra-high resolution spectrometer grating turntable based on a PSO-RBF neural network according to claim 8, characterized in that, The drive system of the grating turntable adopts a combination of DC servo motor and harmonic reducer. The pitch axis uses a harmonic reducer with a reduction ratio of n=20:1, and the position detection uses a high-precision magnetoelectric encoder with a resolution of 0.001°.

Images