A method for controlling turbulence intensity in near-ground laser wireless energy transmission based on KMGA neural network

Abstract

The present application relates to the technical field of atmospheric turbulence, and particularly relates to a turbulence intensity control method in near-ground laser wireless energy transmission based on a KMGA neural network. The method combines a clustering algorithm and a genetic algorithm to generate a KMGA-ANN model to fit the nonlinear relationship between environmental parameters such as temperature, humidity and pressure and atmospheric turbulence intensity, avoids the defect that a traditional neural network model is prone to falling into local optimization, leading to low fitting accuracy of turbulence intensity, and realizes the turbulence generation control function with low repeatability error and low stability error based on a turbulence simulation generator through the nonlinear relationship; solves the problem that a traditional turbulence simulation generation device uses an empirical formula to realize Kolmogorov turbulence but cannot accurately characterize dynamic non-Kolmogorov atmospheric turbulence; and at the same time solves the defects that an indoor turbulence simulation generation device cannot simulate normal low-altitude atmospheric environment with low repeatability error and low stability error and needs to be matched with corresponding turbulence test equipment.

Description

Technical Field

[0001] This invention relates to the field of atmospheric turbulence technology, and in particular to a method for controlling turbulence intensity in near-ground laser wireless power transmission based on a KMGA neural network. Background Technology

[0002] In recent years, with the rapid development of new energy, low-altitude economy, and unmanned aerial vehicle (UAV) technology, laser energy transfer technology, with its advantages of long-distance transmission, high energy density, and no physical wiring constraints, has become an important research direction for solving energy replenishment problems for UAVs, unmanned vehicles, and other equipment. Whether it's providing continuous power to high-altitude UAVs from the ground to extend their operating time, or transferring energy to ground-based unmanned vehicles, laser energy transfer technology shows broad application prospects. However, when lasers are transmitted in the atmosphere, they are inevitably affected by atmospheric turbulence—atmospheric turbulence has randomness, nonlinearity, diffusion, vortex, and dissipation properties, which can cause laser beam drift, expansion, intensity fluctuations, and flickering. This directly leads to a significant decrease in beam quality and a substantial reduction in energy transfer efficiency, and in severe cases, it can even cause the energy transfer system to malfunction, greatly limiting the large-scale application of laser energy transfer technology in practical scenarios. Therefore, one of the core keys to researching laser energy transfer technology is to deeply explore the interaction between lasers and atmospheric turbulence.

[0003] In the study of atmospheric turbulence, the atmospheric refractive index structure constant is usually used. or atmospheric coherence length These parameters serve as characterization parameters to quantify the intensity and characteristics of turbulence. However, due to the random and nonlinear nature of atmospheric turbulence, it is difficult to accurately capture and measure these parameters in real-time in actual atmospheric environments, leading to significant errors in the assessment of the impact of turbulence on laser energy transfer. To address this issue, researchers often conduct field experiments. However, field experiments are not only time-consuming and labor-intensive but also affected by factors such as geographical location and climate conditions, resulting in poor repeatability of experimental results and making it difficult to form systematic research data.

[0004] The emergence of indoor turbulence simulators has provided a controllable experimental environment for laser energy transfer research, effectively avoiding the limitations of field experiments. However, how to achieve stable control of turbulence generated by the turbulence simulator, and how to ensure low repeatability and low stability errors of turbulence intensity in different test batches, remain major engineering challenges. Currently, the mainstream turbulence simulation methods in the industry mainly include convection turbulence simulation and phase screen method: the convection turbulence simulation method simulates the convection motion in the atmosphere through physical means, generating a flow field environment similar to the actual atmospheric turbulence characteristics; the phase screen method is based on optical principles, simulating the disturbance of turbulence on the laser wavefront by constructing a phase screen, providing theoretical and experimental support for the anti-turbulence design of laser energy transfer systems.

[0005] Convection-based turbulence simulation primarily relies on the temperature changes on the upper and lower surfaces of a turbulence generator to induce random air movement within the turbulence pool, causing variations in the refractive index at different locations in the air and thus generating atmospheric turbulence. This method can approximate the changes in real atmospheric turbulence. Currently, mainstream simulation devices generate hot air by heating a resistance wire on the lower panel, while the upper panel uses water cooling circulation to lower the air temperature, creating a temperature gradient difference within the turbulence simulation generator cavity. Finally, the temperature difference within the turbulence pool is precisely controlled by adjusting the power of the heating resistance wire on the lower panel and the temperature of the circulating water on the upper panel to achieve stable turbulence control.

[0006] The phase screen method is a core approach for modeling and simulating atmospheric turbulence from an optical theory perspective. It divides the long-distance propagation path of light waves in the atmosphere into multiple short-distance thin layers. The turbulence effect of each thin layer is equivalent to a phase screen with specific statistical characteristics (consistent with the Kolmogorov power spectrum, etc.). Simulation is achieved through an alternating process of "free-space Fresnel diffraction propagation + phase screen phase modulation." This method first sets fundamental parameters such as propagation distance, wavelength, sampling parameters, and turbulence intensity. Then, it uses Fourier synthesis to generate a phase screen conforming to the turbulence power spectrum (achieved through frequency domain random perturbation correction and inverse Fourier transform). Subsequently, each thin layer undergoes a Fourier transform of the light field, Fresnel transfer function application, inverse Fourier transform (simulating free propagation), and phase screen modulation sequentially. Finally, the propagated light field is obtained. By analyzing key parameters such as the intensity distribution, intensity fluctuation variance, and beam spread radius, it can provide strong data support for the optical design of laser energy transfer systems (such as beam shaping and adaptive optics correction).

[0007] Currently, the main methods available on the market are convection-based turbulence simulation and phase-screen-based turbulence simulation. However, these methods have significant drawbacks in controlling low repeatability and low stability errors of turbulence in practical applications, as detailed below: For convective turbulence simulation methods, although the method can generate relatively stable turbulence by controlling the temperature difference between the upper and lower panels, traditional methods sample turbulence data to obtain an empirical turbulence state. However, when the environment of the turbulence generator (such as temperature, humidity and pressure) changes, these empirical turbulence intensity data have large errors, which seriously affect the repeatability and stability errors of the turbulence simulation.

[0008] For convective turbulence simulation methods, traditional methods typically require a turbulence intensity measurement device (scintillator, atmospheric coherence length meter, or Hartmann wavefront sensor) for real-time measurement to ensure the generation of a specified atmospheric turbulence intensity. However, this approach requires additional hardware. Furthermore, in applications involving laser wireless energy transmission, the high energy transmitted by the laser results in high power tolerance requirements for the testing equipment, and it is inconvenient to set up the corresponding testing equipment in the transmission link.

[0009] The phase screen simulation method has serious limitations in dynamic turbulence. It is mostly a static simulation and cannot efficiently reflect the time evolution characteristics of turbulence. Even with the addition of a time-series phase screen, it is difficult to simulate the high-frequency changing turbulence state. Especially in the field of laser atmospheric transmission, the phase screen simulation method cannot characterize the real-time changing atmospheric turbulence.

[0010] For the phase screen simulation method, the fitting formula of this method adopts a simplified power spectrum model, and the generated phase distribution conforms to the Kolmogorov spectrum. However, the actual atmospheric environment is difficult to be in a stable Kolmogorov spectrum atmospheric turbulence. Therefore, this method cannot play a role in the experiment of laser atmospheric transmission. Summary of the Invention

[0011] This invention discloses a method for controlling turbulence intensity in near-ground laser wireless power transfer based on a KMGA neural network. The specific method is as follows: Set the atmospheric coherence length and temperature control range; Obtain all impact factors; The influencing factors are input into the trained turbulence neural network prediction model to calculate the predicted atmospheric coherence length. When the error between the set atmospheric coherence length and the predicted atmospheric coherence length exceeds the threshold, and the predicted atmospheric coherence length is less than the set atmospheric coherence length, a cooling operation is performed to make the maximum value of the temperature control range equal to the current temperature. When the error between the set atmospheric coherence length and the predicted atmospheric coherence length exceeds the threshold, and the predicted atmospheric coherence length is greater than the set atmospheric coherence length, a heating operation is performed to make the minimum value of the temperature control range equal to the current temperature. Repeatedly adjust the maximum and / or minimum values ​​of the temperature control range until the error between the set atmospheric coherence length and the predicted atmospheric coherence length is less than the threshold, and output the final control temperature of atmospheric turbulence intensity.

[0012] Furthermore, after determining the final control temperature, the temperature of the turbulence simulation generator is brought close to the final control temperature through a PID closed-loop control algorithm.

[0013] Furthermore, all impact factors include any one or more of the following: Temperature, pressure.

[0014] Furthermore, the final controlled temperature is equal to the average of the maximum and minimum values ​​of the temperature control range.

[0015] Furthermore, the turbulence neural network prediction model is a clustering-genetic algorithm neural network model.

[0016] Furthermore, the clustering-genetic algorithm neural network model is constructed as follows: Divide the training set and test set ; Initial population Cross rate and variability Number of outstanding individuals in the parent generation ; Clustering range Number of clusters ; Input training set The neural network model architecture is determined using grid search; The binary code length is determined based on the number of hyperparameters; Initialize the population to obtain the initial parent generation Calculate fitness; Retain the previous generation based on the parent's fitness level. Outstanding individuals from the father generation ; Genetic manipulation to obtain offspring and with Merging ; right Perform individual clustering, with the number of clusters being... ; cluster Sort all individuals within the group and add a penalty; Merge all individuals with added penalties and select the previous ones. Each individual as the next parent generation ; right Genetic manipulation to obtain offspring and use it as the next parent generation ; Calculate the fitness value of each individual and determine the optimal individual. ; The optimal individual Used as initial hyperparameters for ANN models And the optimal model is obtained using optimization algorithms; Input test set Substitute into the prediction model to obtain the predicted value ; Output optimal initial hyperparameters Predicted value .

[0017] Due to the adoption of the above technical solutions, the present invention has the following beneficial effects: 1. In convective turbulence simulation methods, empirical data is usually collected and low-order nonfit formulas are used to generate turbulence simulations. However, when the environment of the turbulence generator (such as temperature, humidity, and pressure) changes, these empirical turbulence intensity data have large errors, which seriously affect the stability and repeatability of the turbulence simulation. This invention adopts an improved neural network model with high-order nonlinear fitting (with thousands of model parameters) and considers multiple turbulence intensity influencing variables to avoid the fitting model getting trapped in local optima, and can better fit the dynamic complex turbulence intensity.

[0018] 2. Traditional turbulence simulation methods typically require the use of turbulence intensity measurement instruments for real-time measurement and control of turbulence intensity. This method is not only costly but also requires calibration of the turbulence measurement instruments. This invention eliminates the need for additional measurement instruments and uses a neural network model to accurately predict and control temperature. This not only saves hardware costs but also enables turbulence intensity control with lower stability and repeatability errors.

[0019] 3. To address the limitations of phase screen simulation methods in dynamic turbulence and their inability to fit real atmospheric turbulence, this invention is based on a convection-type turbulence simulation generator. This not only enables real-time control of the turbulence simulator and generates turbulence intensity over a wide dynamic range, but also reflects the temporal evolution characteristics of turbulence through temperature difference control, avoiding the static aberrations of traditional phase screen intelligent simulation. Furthermore, it incorporates turbulence generation principles and comprehensively considers various turbulence influencing factors to generate non-Kolmogorov spectral turbulence that conforms to real atmospheric conditions.

[0020] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0021] The accompanying drawings of this invention are described below.

[0022] Figure 1 This is a schematic diagram of a neural network process.

[0023] Figure 2 This is a schematic diagram of the overall process.

[0024] Figure 3A schematic diagram of the process for constructing a turbulence neural network prediction model.

[0025] Figure 4 This is a schematic diagram illustrating the principle of the KMGA algorithm.

[0026] Figure 5 This is a comparison chart showing the results before and after adding clustering to the genetic algorithm.

[0027] Figure 6 This is a diagram illustrating the comparison of the model results. Detailed Implementation

[0028] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0029] This application combines clustering and genetic algorithms to generate a KMGA-ANN model (kmeans-GA ANN) that fits the nonlinear relationship between environmental parameters such as temperature, humidity, and pressure and atmospheric turbulence intensity. This avoids the shortcomings of traditional neural network models, which are prone to getting trapped in local optima and resulting in low accuracy in fitting turbulence intensity. Through this nonlinear relationship, it achieves turbulence generation control with low repeatability and low stability errors based on a turbulence simulator. It addresses the issue that traditional turbulence simulators use empirical theoretical formulas to simulate Kolmogorov turbulence but cannot accurately characterize dynamic non-Kolmogorov atmospheric turbulence. It also solves the shortcomings of indoor turbulence simulators, which cannot simulate normal low-altitude atmospheric environments with low repeatability and low stability errors, and require corresponding turbulence testing equipment. This provides strong technical support for research on laser transmission through non-Kolmogorov atmospheric turbulence during laser atmospheric transmission, and provides indoor simulation conditions for atmospheric disturbance effects in low-altitude wireless laser energy transmission technology, laying a foundation for early research and development of laser wireless energy transmission technology in the low-altitude economic field.

[0030] The following explains the basic knowledge involved in this application: 1. Methods for generating atmospheric turbulence Atmospheric refractive index structure constant It describes the intensity of atmospheric turbulence fluctuations, characterizing the severity of random inhomogeneities in the atmospheric refractive index. Light waves and radio waves propagating in the atmosphere are affected by atmospheric turbulence, resulting in various undesirable effects such as spot drift, flicker, and phase fluctuations. In the free atmosphere, the refractive index is generally considered to be a function of temperature, humidity, and pressure. In the formula: n is the refractive index of air, T is the temperature, Q is the absolute humidity, and P is the air pressure. At small scales, temperature fluctuations or refractive index fluctuations can also be considered to satisfy the assumption of local homogeneity and isotropy, and the 2 / 3 law of the structure function. This can be expressed as: In the formula and are the structure functions of refractive index and temperature respectively; and are the structure constants of refractive index and temperature respectively.

[0031] 2. Neural network prediction model Based on the characteristics that artificial neural networks can solve regression tasks, the neural network prediction model is used to predict the relationship between the input feature vector and the output vector , so as to find a very wide family of functions that map from the input features to the output vector , where is expressed as the hyperparameter in the family of functions . This family of functions provides a new expression for , and the best-performing can be obtained through continuous training, so as to be used to predict the unknown output vector of . The neural network process is as shown in Figure 1 , where is the training learning rate.

[0032] A method for controlling the turbulence intensity in near-earth laser wireless power transmission based on the KMGA neural network. The method for predicting and controlling atmospheric turbulence based on the neural network model does not require additional instruments for detecting the atmospheric turbulence intensity in real time. The accurate prediction of the neural network model can be used to achieve turbulence simulation with low repeatability error and low stability error. The specific process is as shown in Figure 2 : S1. Set the atmospheric coherence length r0 and the temperature control range [Tmin, Tmax].

[0033] S2. Obtain all influencing factors.

[0034] In step S2, all influencing factors include: temperature T, humidity H, and pressure P.

[0035] S3. Input the influencing factors into the trained turbulence neural network prediction model to calculate the predicted atmospheric coherence length r0'; S4. When the error between the set atmospheric coherence length r0 and the predicted atmospheric coherence length r0' exceeds the threshold, and the predicted atmospheric coherence length is less than the set atmospheric coherence length, i.e., r0'<r0, perform a cooling operation, and set the maximum value of the temperature control range equal to the current temperature, Tmax = T.

[0036] S5. When the error between the set atmospheric coherence length r0 and the predicted atmospheric coherence length r0' exceeds the threshold, and the predicted atmospheric coherence length is greater than the set atmospheric coherence length, r0'>r0, a heating operation is performed to make the minimum value of the temperature control range equal to the current temperature Tmin=T.

[0037] S6. Repeatedly adjust the maximum and / or minimum values ​​of the temperature control range until the error between the set atmospheric coherence length and the predicted atmospheric coherence length is less than the threshold, and output the final control temperature of atmospheric turbulence intensity, T(i)=(Tmax+Tmin) / 2.

[0038] S7. After determining the final control temperature, the temperature of the turbulence simulation generator is brought close to the final control temperature through a PID closed-loop control algorithm.

[0039] In this embodiment, a turbulence neural network prediction model is constructed, such as... Figure 3 As shown in the table below, the specific operations are as follows: The KMGA algorithm, by performing individual clustering on the population, suppresses the generation of similar individuals in the genetic algorithm, ensures population diversity, and avoids premature convergence of the genetic algorithm, thereby preventing the prediction model from getting trapped in local optima. A schematic diagram of this algorithm is shown below. Figure 4 As shown.

[0040] The difference between the KMGA and GA algorithms lies in the fact that the KMGA algorithm can quickly identify similar individuals through clustering and apply corresponding penalties to ensure that similar individuals with low fitness are discarded during the selection process. However, clustering is not suitable for all stages of the genetic algorithm. In the early stages of the genetic algorithm, since randomly generated individuals need to be continuously reproduced to produce more excellent individuals, clustering should not be added in the early stages. At the same time, the cluster range should not be too long, as a too-long cluster range will weaken the effect of the genetic algorithm itself and easily cause the model to converge to a local minimum prematurely. Figure 5 The graph shows a comparison of the results of the genetic algorithm before and after adding clustering. It can be seen from the graph that without clustering, the highest fitness of the population after 100 iterations is 80, but after adding clustering, the highest fitness can reach 170.

[0041] This model was compared with different prediction models, using root mean square error (RMSE) or mean absolute error (MAE) as evaluation criteria. The model proposed in this invention showed the best prediction performance. Its prediction results are as follows: Figure 6 As shown in the figure and the table below.

[0042] The proposed KMGA-ANN prediction model was applied to the control and prediction of an atmospheric turbulence simulation generator. The turbulence intensity of strong turbulence (r0 < 10 cm) and weak turbulence (r0 > 10 cm) was controlled respectively. The repeatability and stability accuracy are shown in the table below.

[0043] The stability error refers to the fluctuations in the change of r0 across consecutive frames after the turbulence simulation control has stabilized. A smaller stability error indicates a more stable r0. Its formula is expressed as: Repeatability error refers to the deviation between the actual generated r0' and the set r0 in consecutive frames after the turbulence simulation control has stabilized. The smaller the repeatability error, the closer r0' is to the set r0. Its formula is expressed as:

[0044] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for controlling turbulence intensity in near-ground laser wireless power transfer based on a KMGA neural network, characterized in that, The specific method is as follows: Set the atmospheric coherence length and temperature control range; Obtain all impact factors; The influencing factors are input into the trained turbulence neural network prediction model to calculate the predicted atmospheric coherence length. When the error between the set atmospheric coherence length and the predicted atmospheric coherence length exceeds the threshold, and the predicted atmospheric coherence length is less than the set atmospheric coherence length, a cooling operation is performed to make the maximum value of the temperature control range equal to the current temperature. When the error between the set atmospheric coherence length and the predicted atmospheric coherence length exceeds the threshold, and the predicted atmospheric coherence length is greater than the set atmospheric coherence length, a heating operation is performed to make the minimum value of the temperature control range equal to the current temperature. Repeatedly adjust the maximum and / or minimum values ​​of the temperature control range until the error between the set atmospheric coherence length and the predicted atmospheric coherence length is less than the threshold, and output the final control temperature of atmospheric turbulence intensity.

2. The method for controlling turbulence intensity in near-ground laser wireless power transfer based on KMGA neural network as described in claim 1, characterized in that, After determining the final control temperature, the temperature of the turbulence simulation generator is brought close to the final control temperature through a PID closed-loop control algorithm.

3. The method for controlling turbulence intensity in near-ground laser wireless power transfer based on KMGA neural network as described in claim 1, characterized in that, All impact factors include any one or more of the following: Temperature, pressure.

4. The method for controlling turbulence intensity in near-ground laser wireless power transfer based on KMGA neural network as described in claim 1, characterized in that, The final controlled temperature is equal to the average of the maximum and minimum values ​​of the temperature control range.

5. The method for controlling turbulence intensity in near-ground laser wireless power transfer based on KMGA neural network as described in claim 1, characterized in that, The turbulence neural network prediction model is a clustering-genetic algorithm neural network model.

6. The method for controlling turbulence intensity in near-ground laser wireless power transfer based on a KMGA neural network as described in claim 5, characterized in that, The clustering-genetic algorithm neural network model is constructed as follows: Divide the training set and test set ; Initial population Crossover rate and variability Number of outstanding individuals in the parent generation ; Clustering range Number of clusters ; Input training set The neural network model architecture is determined using grid search; The binary code length is determined based on the number of hyperparameters; Initialize the population to obtain the initial parent generation Calculate fitness; Retain the previous generation based on the parent's fitness level. Outstanding individuals from the father generation ; Genetic operations are performed to obtain offspring. and with Merging ; right Perform individual clustering, with the number of clusters being... ; cluster Sort all individuals within the group and add a penalty; Merge all individuals with added penalties and select the previous ones. Each individual as the next parent generation ; right Genetic operations are performed to obtain offspring. and use it as the next parent generation ; Calculate the fitness value of each individual and determine the optimal individual. ; The optimal individual Used as initial hyperparameters for ANN models And the optimal model is obtained using optimization algorithms; Input test set Substitute into the prediction model to obtain the predicted value ; Output optimal initial hyperparameters Predicted value .

Images