A method and system for reconstruction of optical plasmonic boundary configurations

Abstract

The application discloses a kind of optical plasma boundary configuration reconstruction method and system, belong to plasma physics technical field, including collection optical plasma image, input the optical plasma image of collection into trained U-Net network model, carry out optical plasma boundary identification, obtain the pixel coordinates of multiple boundary points of optical plasma image;Mapping function is fitted using XGBoost model training, the pixel coordinates of lower boundary point in the plasma image are converted into the pixel coordinates of lower boundary point in polar surface using the mapping function after fitting, realize the reconstruction of the optical plasma configuration.The application selects the label value of pixel coordinates under the condition that there is no camera calibration, and realizes the boundary reconstruction of optical plasma based on XGBoost model, converts image plane into the polar surface of tokamak, compared with prior art, the reconstruction method of the application is more accurate, and there is no horizontal drift problem.

Description

Technical Field

[0001] This invention relates to the field of plasma physics, and more specifically to a method and system for reconstructing optical plasma boundary configurations. Background Technology

[0002] Real-time reconstruction of the plasma configuration of a tokamak device is of great significance for optimizing operating parameters, controlling the thermal load and particle deposition on the vacuum chamber walls and divertor target plates, and reducing damage to vacuum chamber components. Modern tokamak devices typically employ magnetic measurement methods. The plasma configuration of the EAST tokamak device is primarily estimated using the Plasma Equilibrium and Inversion Control (EFIT) algorithm. However, due to the influence of magnetic perturbations on the plasma, which randomly disrupt magnetic field lines, and the integral drift phenomenon inherent in this measurement method, this approach is unsuitable for long-term operation of future tokamak devices.

[0003] Therefore, to meet the needs of long-term operation of future tokamak devices, it is essential to develop new plasma configuration reconstruction methods. Images captured by the visible light camera on EAST clearly show the plasma boundaries. Extracting the plasma boundaries from these images and reconstructing their configuration can avoid the problems encountered in magnetic measurements.

[0004] Hao LUO et al. proposed a boundary reconstruction method that does not require camera calibration. The algorithm aims to minimize the distance between the optical reconstruction result and the EFIT data, achieving an error of 1-2 cm on EAST discharge experiments and EFIT data. However, its edge extraction algorithm requires manual setting of the ROI region, making it difficult to extract the outer plasma boundary, and the fitted data exhibits horizontal drift.

[0005] Therefore, proposing a method and system for reconstructing optical plasma boundary configurations with higher accuracy in boundary extraction and configuration reconstruction is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] In view of this, the present invention provides a method and system for reconstructing optical plasma boundary configurations. The image segmentation extraction algorithm based on U-Net network can extract plasma boundaries from the entire image without manually setting ROI regions, and can extract outer plasma boundaries. The optical reconstruction algorithm based on XGBoost has higher boundary reconstruction accuracy and does not have the problem of horizontal drift.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] On one hand, the present invention provides a method for reconstructing the optical plasma boundary configuration, comprising the following steps:

[0009] Acquire isooptic plasma images and establish a sample dataset;

[0010] Build and train the U-Net network model to obtain the trained U-Net network model;

[0011] The acquired optical plasma image is input into the trained U-Net network model to perform optical plasma boundary recognition, obtain multiple boundary points of the optical plasma, and record the pixel coordinates of the boundary points.

[0012] A mapping function between the optical plasma image coordinates and the poloidal plane coordinates is constructed; the mapping function is fitted using an XGBoost model.

[0013] The pixel coordinates of the lower boundary point of the plasma image are converted into the pixel coordinates of the lower boundary point of the poloidal plane using the fitted mapping function, thereby realizing the reconstruction of the optical plasma configuration.

[0014] Preferably, the specific steps to obtain a trained U-Net network model include:

[0015] The sample dataset is divided into a training set and a test set;

[0016] The optical plasma images in the training set are labeled;

[0017] The optical plasma images in the training set and the test set are preprocessed;

[0018] Construct a U-Net network model and train it using the preprocessed training set;

[0019] The trained U-Net network model was validated using a preprocessed test set, and the optimal U-Net network model was selected.

[0020] Preferably, the structure of the U-Net network model includes an encoding module and a decoding module.

[0021] Preferably, the loss function of the U-Net network model is cross-entropy loss, as shown in the following formula:

[0022]

[0023] Where M represents the number of categories; p c yc is the predicted value, representing the probability that the sample belongs to class c; yc represents the true distribution of the sample, the true probability that it belongs to class c, which is usually 0 or 1.

[0024] Preferably, the specific steps for constructing the mapping function between the optical plasma image coordinates and the poloidal plane coordinates are the pixel coordinate label value selection process, wherein the label value selection process is as follows:

[0025] The pixel coordinates (u, v) of multiple boundary points of optical plasma were obtained using the Unet network model.

[0026] The pixel coordinates (u, v) are scaled and translated to obtain the pixel coordinates (u', v') under the poloidal plane. The pixel coordinates (u', v') under the poloidal plane are placed in the same coordinate system as the EFIT data, and the geometric center of the optical plasma boundary is made to coincide with the geometric center of the EFIT data graphic.

[0027] Connect the geometric center to the pixel coordinates (u', v'), and extend the line connecting the geometric center and the pixel coordinates (u', v') to intersect the EFIT fitting curve to obtain the intersection point coordinates (r, z), which are used as the label value.

[0028] Preferably, the XGBoost model is constructed as follows:

[0029] Define the XGBoost model as follows:

[0030]

[0031] Among them, f k (x) represents the k-th decision tree, x i This represents the input feature vector. This represents the predicted value; the model contains K trees. Represents the regression tree space;

[0032] The objective function of the XGBoost model is defined as follows:

[0033]

[0034] In the XGBoost model, the objective function Obj consists of a loss function L and a regularization term Ω. The loss function L is used to evaluate the error between the predicted and actual values, and the regularization term Ω is used to control the complexity of the model and avoid overfitting. The definition of the regularization term Ω is:

[0035]

[0036] Where γ and λ are custom coefficients, T is the number of leaves, and ω is the weight score of the leaf node;

[0037] The model for round s is:

[0038]

[0039] Where s is the iteration round number, and the model in the s-th round is equal to the model in the (s-1)-th round plus the new model f(x). i ), where f(x) i This minimizes the objective function Obj.

[0040] After round s, the objective function Obj is rewritten as:

[0041]

[0042] in, For the (s-1)th round model, the sample x i The predicted value, f s (x i ) represents the predicted value of the new model trained in the s-th round;

[0043] Taylor expansion of the objective function Obj:

[0044]

[0045] Among them, g i h is the first derivative of the loss function. i This is the second derivative of the loss function;

[0046] Due to the constant term It does not affect the optimization of the objective function, so it can be removed, and Ω(f) can be removed. s Substituting into the formula, the objective function Obj can be written as:

[0047]

[0048] Among them I j Let j be all samples falling on leaf node j; And for ω j Find the partial derivatives such that the partial derivatives equal to 0, and we get:

[0049]

[0050]

[0051] Based on the optimal solution objective function Obj (s) Split sample data.

[0052] On the other hand, the present invention provides a system for reconstructing optical plasma boundary configurations, used to implement any of the above-mentioned methods for reconstructing optical plasma boundary configurations, the system comprising:

[0053] The data acquisition module is used to acquire data and construct a sample dataset, which includes a training set and a test set.

[0054] The data preprocessing module is used to preprocess the constructed sample dataset;

[0055] The U-Net network model building module is used to build a U-Net network model, and to train the built U-Net network model using the training set to obtain a trained U-Net network model.

[0056] The boundary recognition module is used to perform optical plasma boundary recognition based on the trained U-Net network model, obtain multiple boundary points of the optical plasma, and record the pixel coordinates of the boundary points;

[0057] The mapping model construction module constructs a mapping function between the optical plasma image coordinates and the poloidal plane coordinates, and uses the XGBoost model to train and fit the mapping function.

[0058] The configuration reconstruction module is used to convert the pixel coordinates of the lower boundary point of the plasma image into the pixel coordinates of the lower boundary point of the poloidal plane using a fitted mapping function, thereby realizing the reconstruction of the optical plasma configuration.

[0059] As can be seen from the above technical solutions, compared with the prior art, this invention discloses a method and system for reconstructing the boundary configuration of optical plasma. Based on the U-Net network model, it performs semantic segmentation of the image, eliminating the need for preprocessing of the original image or manually setting the ROI region. Compared with traditional edge detection algorithms, using the U-Net network model for optical plasma boundary recognition can extract boundaries better and avoids a large number of false boundaries. Furthermore, without camera calibration, this invention selects pixel coordinate label values ​​and implements optical plasma boundary reconstruction based on the XGBoost model, converting the image plane into the tokamak's poloidal surface. Compared with the prior art, the reconstruction method of this invention has higher accuracy and does not suffer from horizontal drift. Attached Figure Description

[0060] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0061] Figure 1 This is a schematic flowchart of a method for reconstructing the optical plasma boundary configuration according to the present invention.

[0062] Figure 2 This is a schematic diagram of the U-Net network model used for optical plasma boundary recognition in this invention.

[0063] Figure 3(a) shows the original image, and Figure 3(b) shows the detection results of the plasma boundary adaptive Canny edge detection algorithm.

[0064] Figure 4(a) shows the original image, Figure 4(b) shows the boundary region of the plasma at 6.9 seconds of shot #110790 identified by the U-Net network model, and Figure 4(c) shows the obtained boundary center point.

[0065] Figure 5(a) shows the EFIT data fitting curve for the 2.61st second of shot #110790, and Figure 5(b) shows the set of pixel coordinates (u', v') for the 2.61st second of shot #110790.

[0066] Figure 6(a) shows the EFIT and optical reconstruction coordinates of gun #110793 at 1.02 seconds; Figure 6(b) shows the EFIT and optical reconstruction coordinates of gun #110793 at 6.12 seconds; Figure 6(c) shows the EFIT and optical reconstruction coordinates of gun #110790 at 3.16 seconds; and Figure 6(d) shows the EFIT and optical reconstruction coordinates of gun #110790 at 10.09 seconds.

[0067] Figure 7 This is a schematic diagram of the structure of an optical plasma boundary configuration reconstruction system according to the present invention. Detailed Implementation

[0068] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0069] This invention discloses a method for reconstructing the boundary configuration of an optical plasma, such as... Figure 1 As shown, it includes the following steps:

[0070] Optical plasma images were acquired using a visible light camera, and a sample dataset was established.

[0071] A U-Net network model was constructed and trained using a training set to obtain the trained U-Net network model. The sample dataset was constructed by dividing it into training and test sets; plasma boundary regions were manually labeled in the images to serve as the real segmentation images for network training. Before model training, the image size was adjusted to 512×512, and data augmentation was performed to prevent overfitting. Figures 3 and 4 show the detection results of the Canny algorithm and the recognition results of the U-Net network.

[0072] A U-Net network model is constructed and trained using a preprocessed training set. The trained U-Net network model is then validated using a preprocessed test set, and the optimal U-Net network model is selected.

[0073] like Figure 2 As shown, the U-Net network model consists of an encoding module and a decoding module. The U-Net network is a fully convolutional network improved from FCN. Its structure comprises a contraction path and an expansion path. The left side represents the contraction path, where each layer consists of two 3×3 convolutional layers and a 2×2 max-pooling layer, using ReLU activation. Each layer doubles the number of feature channels. The right side represents the expansion path, where each layer first uses deconvolution, halving the number of feature channels, then concatenates with the feature maps from the corresponding contraction path, followed by two 3×3 convolutional layers. The last layer of the expansion path uses a 1×1 convolutional kernel, resulting in only one feature channel. The Sigmoid function restricts the output range to [0,1]. The output segmentation image represents the probability that each pixel belongs to the plasma boundary. Pixels with a probability greater than 0.5 are ultimately selected as belonging to the ELMs filamentary structure.

[0074] The U-Net network model uses cross-entropy loss as its loss function, as shown in the following formula:

[0075]

[0076] Where M represents the number of categories, p c For the predicted value, y c This represents the true distribution of the sample, indicating the true probability that it belongs to category c, typically 0 or 1. In this embodiment, M is set to 2 (whether it belongs to the plasma boundary). The value is 0 or 1; it is 1 when the category is the same as the true category, and 0 otherwise. This represents the probability that the sample belongs to category c.

[0077] In this embodiment, data from gun numbers #110790, #110792, and #110793 are selected for processing and analysis. The acquired optical plasma images are input into a trained U-Net network model to perform optical plasma boundary recognition, obtain multiple boundary points of the optical plasma, and record the pixel coordinates (u,v) of the boundary points.

[0078] A mapping function between optical plasma image coordinates and poloidal plane coordinates is constructed. The pixel coordinates (u, v) are scaled and translated to obtain the pixel coordinates (u', v') under the poloidal plane. The pixel coordinates (u', v') under the poloidal plane are placed in the same coordinate system as the EFIT data, and the geometric center of the optical plasma boundary coincides with the geometric center of the EFIT data graphic, as shown in Figures 5(a) and 5(b). The geometric center is connected to the pixel coordinates (u', v'), and the line connecting the geometric center and the pixel coordinates (u', v') is extended to intersect the EFIT fitting curve to obtain the intersection point coordinates (r, z). The intersection point coordinates (r, z) are used as the label value.

[0079] The selection of label values ​​can be viewed as a functional mapping relationship with pixel coordinates. The XGBoost model can fit this function through training. XGBoost consists of multiple CART (Classification and Regression Trees), with multiple decision trees making decisions together and summing all the results to obtain the final result. It adopts the ensemble idea of ​​Boosting, integrating multiple weak learners into a strong learner to improve the overall model performance. Therefore, the XGBoost model is used to fit the mapping function during training. The EFIT data and image data of all gun numbers at the same time are processed in the same way. The (u',v') and (r,z) sets obtained from gun numbers #110792 and #110793 are used as the training set, and the (u',v') and (r,z) sets obtained from gun number #110790 are used as the test set.

[0080] We trained two XGBoost models, model_R and model_Z, to fit the r and z values, respectively. Model_R takes the training set (u', v') as input and the training set's r-coordinates as the true values. Model_Z takes the training set (u', v') as input and the training set's z-coordinates as the true values. The mean squared error of model_R on the training set is 3.5149 × 10⁻⁶. -5 m 2 The mean squared error of model_Z is 1.4034 × 10⁻⁶. -5 m 2 The mean squared error of model_R obtained on the test set is 3.7119 × 10⁻⁶. -5 m 2 The mean squared error of model_Z is 1.7045 × 10⁻⁶. -5 m 2 Therefore, the average distance error between the (r, z) coordinates predicted by the model on the test set and the label value is 7.36 mm. Figures 6(a) - 6(d) The image shows a comparison between the model's predicted values ​​and the EFIT data:

[0081] By using the fitted mapping function, the pixel coordinates of the lower boundary points of the plasma image are converted into the pixel coordinates of the lower boundary points of the poloidal plane, thereby realizing the reconstruction of the optical plasma configuration.

[0082] The construction process of the XGBoost model is as follows:

[0083] Define the XGBoost model as follows:

[0084]

[0085] Among them, f k (x) represents the k-th decision tree, x i This represents the input feature vector. This represents the predicted value; the model contains K trees. Represented as the regression tree space;

[0086] The objective function of the XGBoost model is defined as follows:

[0087]

[0088] In the XGBoost model, the objective function Obj consists of a loss function L and a regularization term Ω. The loss function L is used to evaluate the error between the predicted and actual values, and the regularization term Ω is used to control the complexity of the model and avoid overfitting. The definition of the regularization term Ω is:

[0089]

[0090] Where γ and λ are custom coefficients, T is the number of leaves, and ω is the weight score of the leaf node;

[0091] The model for round s is:

[0092]

[0093] Where s is the iteration round number, and the model in the s-th round is equal to the model in the (s-1)-th round plus the new model f(x). i ), where f(x) i This minimizes the objective function Obj.

[0094] After round s, the objective function Obj is rewritten as:

[0095]

[0096] in, For the (s-1)th round model, the sample x i The predicted value, f s (x i ) represents the predicted value of the new model trained in the s-th round;

[0097] Taylor expansion of the objective function Obj:

[0098]

[0099] Among them, g i h is the first derivative of the loss function. i This is the second derivative of the loss function;

[0100] Due to the constant term It does not affect the optimization of the objective function, so it can be removed, and Ω(f) can be removed. s Substituting into the formula, the objective function Obj can be written as:

[0101]

[0102] Among them I j Let j be all samples falling on leaf node j; And for ω j Find the partial derivatives such that the partial derivatives equal to 0, and we get:

[0103]

[0104]

[0105] Based on the optimal solution objective function Obj (s) Split sample data.

[0106] On the other hand, the present invention provides a system for reconstructing optical plasma boundary configurations, used to implement any of the aforementioned methods for reconstructing optical plasma boundary configurations, such as... Figure 7 As shown, it includes:

[0107] The data acquisition module is used to acquire data and build a sample dataset, which includes a training set and a test set.

[0108] The data preprocessing module is used to preprocess the constructed sample dataset;

[0109] The U-Net network model building module is used to build a U-Net network model, and to train the built U-Net network model using the training set to obtain a trained U-Net network model.

[0110] The boundary recognition module is used to perform optical plasma boundary recognition based on the trained U-Net network model, obtain multiple boundary points of the optical plasma, and record the pixel coordinates of the boundary points;

[0111] The mapping model construction module constructs a mapping function between optical plasma image coordinates and poloidal plane coordinates, and uses the XGBoost model to train and fit the mapping function.

[0112] The configuration reconstruction module is used to convert the pixel coordinates of the lower boundary points of the plasma image into the pixel coordinates of the lower boundary points of the poloidal plane using a fitted mapping function, thereby realizing the reconstruction of the optical plasma configuration.

[0113] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0114] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for reconstructing the boundary configuration of an optical plasma, characterized in that, Includes the following steps: Acquire optical plasma images and build a sample dataset; Build and train the U-Net network model to obtain the trained U-Net network model; The acquired optical plasma image is input into the trained U-Net network model to perform optical plasma boundary recognition, obtain multiple boundary points of the optical plasma, and record the pixel coordinates of the boundary points; A mapping function between the optical plasma image coordinates and the poloidal plane coordinates is constructed; the mapping function is fitted using an XGBoost model; and the fitted mapping function is used to convert the pixel coordinates of the lower boundary points of the optical plasma image into the pixel coordinates of the lower boundary points of the poloidal plane, thereby reconstructing the optical plasma configuration. The specific steps for constructing the mapping function between the optical plasma image coordinates and the poloidal plane coordinates are the pixel coordinate label value selection process, which is as follows: The pixel coordinates (u, v) of multiple boundary points of the optical plasma were obtained using the Unet network model. The pixel coordinates (u, v) are scaled and translated to obtain the pixel coordinates (u', v') under the poloidal plane. The pixel coordinates (u', v') under the poloidal plane are placed in the same coordinate system as the EFIT data, and the geometric center of the optical plasma boundary coincides with the geometric center of the EFIT data graphic. Connect the geometric center to the pixel coordinates (u', v'), and extend the line connecting the geometric center and the pixel coordinates (u', v') to intersect the EFIT fitting curve to obtain the intersection coordinates (r, z). The intersection coordinates (r, z) are used as the label values ​​for the pixel coordinates.

2. The method for reconstructing the optical plasma boundary configuration according to claim 1, characterized in that, The specific steps to obtain a trained U-Net network model include: The sample dataset is divided into a training set and a test set; The optical plasma images in the training set are labeled; The optical plasma images in the training set and the test set are preprocessed; Construct a U-Net network model and train it using the preprocessed training set; The trained U-Net network model was validated using a preprocessed test set, and the optimal U-Net network model was selected.

3. The method for reconstructing the optical plasma boundary configuration according to claim 1, characterized in that, The structure of the U-Net network model includes an encoding module and a decoding module.

4. The method for reconstructing the optical plasma boundary configuration according to claim 1, characterized in that, The loss function of the U-Net network model is cross-entropy loss, as shown in the following formula: Where M represents the number of categories; y is the predicted value, representing the probability that the sample belongs to class c; c This represents the true distribution of the sample and the true probability that it belongs to class c, usually 0 or 1.

5. The method for reconstructing the optical plasma boundary configuration according to claim 1, characterized in that, The XGBoost model is constructed as follows: Define the XGBoost model as follows: in, Indicates the first k A decision tree, This represents the input feature vector. This represents the predicted value; the model contains K trees. Represented as the regression tree space; The objective function of the XGBoost model is defined as follows: Among them, y i Representing the true value, the objective function Obj of the XGBoost model is determined by the loss function. L The loss function L is composed of a loss function L and a regularization term Ω. The loss function L is used to evaluate the error between the predicted and actual values, while the regularization term Ω controls the complexity of the model and avoids overfitting. The definition of the regularization term Ω is: in, and Here, T is a user-defined coefficient, and T is the number of leaves. The weight score of the leaf node; The model for round s is: Where s is the iteration round number, and the model in the s-th round is equal to the model in the (s-1)-th round plus the new model. ,in, Minimize the objective function Obj; After round s, the objective function Obj is rewritten as: in, For the sample of the model in the (s-1)th round The predicted value, The predicted value is the value of the new model trained in the s-th round; Taylor expansion of the objective function Obj: in, The first derivative of the loss function. This is the second derivative of the loss function; Due to the constant term It does not affect the optimization of the objective function, so it can be removed, and... Substituting into the formula, the objective function Obj can be written as: in Let j be all samples falling on leaf node j; , and to Find the partial derivatives such that the partial derivatives equal to 0, and we get: Based on the optimal solution objective function Split sample data.

6. A system for reconstructing optical plasma boundary configurations, characterized in that, The system is used to implement the method for reconstructing the optical plasma boundary configuration according to any one of claims 1-5, the system comprising: The data acquisition module is used to acquire data and construct a sample dataset, which includes a training set and a test set. The data preprocessing module is used to preprocess the constructed sample dataset; The U-Net network model building module is used to build a U-Net network model, and to train the built U-Net network model using the training set to obtain a trained U-Net network model. The boundary recognition module is used to perform optical plasma boundary recognition based on the trained U-Net network model, obtain multiple boundary points of the optical plasma, and record the pixel coordinates of the boundary points; The mapping model construction module is used to construct the mapping function between the optical plasma image coordinates and the poloidal plane coordinates, and to fit the mapping function using the XGBoost model training. The configuration reconstruction module is used to convert the pixel coordinates of the lower boundary point of the optical plasma image into the pixel coordinates of the lower boundary point of the poloidal plane using a fitted mapping function, thereby realizing the reconstruction of the optical plasma configuration.

Images