A multi-mode tokamak device rupture warning method and system

Abstract

The present invention discloses a multi-modal tokamak device rupture warning method and system, which belongs to the field of tokamak device safety protection. The method comprises: step 1, collecting multi-modal data, and marking the rupture data in the collected multi-modal data; step 2, aligning the collected multi-modal data with the time axis, dividing the time slices, and forming a rupture warning database; step 3, using the rupture warning database to train a rupture warning model, and deploying the optimal rupture warning model obtained through training in real time; step 4, during the tokamak experiment, inputting the 0-dimensional diagnostic data and the two-dimensional image diagnostic data obtained in real time into the optimal rupture warning model, and performing rupture mitigation and device protection according to the output rupture warning results. The present invention can more accurately predict the rupture state of the tokamak device, and trigger the rupture processing actuator when it is about to rupture, so as to better prevent the device from suffering rupture damage.

Description

Technical Field

[0001] The present invention belongs to the field of tokamak device safety protection, and in particular relates to a multi-modal tokamak device rupture early warning method and system. Background Art

[0002] When a tokamak plasma ruptures, energy is deposited onto the first wall material within a very short period of time, directly causing material damage and compromising operational safety. In devices such as the ITER (International Thermonuclear Experimental Reactor) and BEST (Burning Plasma Experimental Superconducting Tokamak), damage caused by ruptures is exacerbated by factors such as high energy storage and high currents. During a rupture, eddy currents and corona currents are generated within the device's internal components, interacting with the magnetic field to generate electromagnetic forces exceeding tens of thousands of Newtons. These currents directly threaten the safety of internal components. For future fusion reactors, a single unmitigated rupture could damage the device and cause incalculable losses. However, the chain of events leading to a plasma rupture is extremely complex, nonlinear, and rapid, making it difficult to provide early warning of a rupture based on physical understanding. Therefore, there is an urgent need to develop a method and system for rupture warning that can integrate a wide range of diagnostic information, including visible and infrared images of plasma temperature, density, and plasma state, to provide early warning of a rupture. Summary of the Invention

[0003] In order to solve the problems existing in the prior art, the present invention provides a multi-modal tokamak device rupture warning method and system, which integrates the information in the 0-dimensional time series signal and the two-dimensional image signal, can accurately predict the rupture of the tokamak device, and thus effectively alleviate the rupture hazards.

[0004] In order to achieve the purpose of the present invention, the technical solution adopted by the present invention is as follows:

[0005] A multi-mode tokamak device rupture early warning method, the method comprising the following steps:

[0006] Step 1: Data acquisition and annotation: Collect multimodal data, including 0-dimensional diagnostic data collected and calculated using real-time diagnostics in the tokamak device, and plasma 2D image diagnostic data collected using a camera; annotate the fracture data in the collected multimodal data;

[0007] Step 2: Data preparation: Align the collected multimodal data with the time axis and divide them into time slices. Each time slice contains 0-dimensional diagnostic data, 2D image diagnostic data, and annotated fracture data information. The above data and information together form a fracture warning database.

[0008] Step 3: Model training: Use the fracture warning database to train the fracture warning model, and deploy the optimal fracture warning model obtained through training in real time;

[0009] Step 4, rupture warning: During the tokamak experiment, the real-time 0-dimensional diagnostic data and 2D image diagnostic data are input into the optimal rupture warning model, and rupture mitigation and device protection are performed based on the output rupture warning results.

[0010] Furthermore, in step 1, the 0-dimensional diagnostic data includes plasma current Ip, plasma current error Ip_error, plasma density Ne, loop voltage V_loop, and plasma elongation ratio Kappa; the plasma two-dimensional image diagnostic data collected by the camera includes a tokamak image, and the tokamak image includes any one of an image of the entire plasma area observed along the plasma tangent and an image of a partial plasma area, wherein the band to which the collected plasma two-dimensional image belongs is any one of the visible light band, infrared band, and ultraviolet band; the labeling of the rupture data in the collected multimodal data includes labeling by a method combining automatic judgment of the current change rate and expert experience labeling, and marking the time point of the rupture in the collected multimodal data.

[0011] Furthermore, in step 2, aligning the time axis of the collected multimodal data includes downsampling the high sampling rate data to match the low sampling rate data.

[0012] Furthermore, in step 3, the structure of the rupture warning model is: using a convolutional neural network (CNN) and a long short-term memory (LSTM) network to process 0-dimensional diagnostic data to obtain a first hidden vector; using a Video Vision Transformer model to process plasma two-dimensional image diagnostic data to obtain a second hidden vector; after splicing the first hidden vector and the second hidden vector, classifying them through a multi-layer perceptron (MLP) to obtain a rupture warning result.

[0013] Furthermore, in step 4, 0-dimensional diagnostic data is obtained in real time by reading the acquisition system data through the reflective memory card, and 2D image diagnostic data is obtained in real time through the image acquisition card; the result of the rupture warning is transmitted to the plasma control system in real time through the reflective memory network.

[0014] On the other hand, the present invention also provides a multi-modal tokamak device rupture early warning system, the system comprising:

[0015] Data acquisition and annotation module: used to acquire multimodal data, including 0-dimensional diagnostic data acquired and calculated using real-time diagnostics in the tokamak device, and plasma 2D image diagnostic data acquired using a camera; and to annotate fracture data in the acquired multimodal data;

[0016] Data preparation module: used to align the time axis of the collected multimodal data and divide it into time slices. Each time slice contains 0-dimensional diagnostic data, 2D image diagnostic data, and annotated fracture data information;

[0017] Rupture warning module: It is used to input the time-aligned multi-modal data into the optimal rupture warning model during the tokamak experiment, and perform rupture mitigation and device protection based on the output rupture warning results.

[0018] The beneficial effects of the present invention are:

[0019] The present invention proposes a multimodal tokamak device rupture warning method, which combines the tokamak information contained in the 0-dimensional diagnostic signal and the 2-dimensional image data, and can more accurately predict the rupture state of the tokamak device; at the same time, it realizes a multimodal tokamak device rupture warning system, which can predict the plasma rupture state in real time through a multimodal method and trigger the rupture processing actuator when the rupture is about to occur, so as to better prevent the device from suffering rupture damage. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a flow chart of a multi-modal tokamak device rupture warning method according to the present invention;

[0021] Figure 2 This is a structural diagram of a multi-modal tokamak device rupture warning system of the present invention. DETAILED DESCRIPTION

[0022] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0023] like Figure 1 The figure below is a flow chart of a multi-modal tokamak device rupture warning method. The rupture warning model training process is divided into the following steps:

[0024] Step 1: Data Collection and Annotation: Multimodal data is collected. In this embodiment, the 0-dimensional diagnostic signals used include the plasma current signal Ip, the plasma current error signal Ip_error, the density signal Ne, the loop voltage V_loop, the plasma vertical position z, the normalized specific pressure Betan, the lithium impurity content Li, the plasma elongation ratio Kappa, the boundary safety factor Q95, and the plasma energy storage Wmhd, totaling 10 diagnostic signals. The 2D image diagnostic signal uses a tangential observation camera installed in the plane of the J window. The field of view covers the entire annular cross-section of the plasma, and the image bands collected are any of the visible light band, infrared band, and ultraviolet band. Data annotation uses a method that automatically determines the current change rate, such as the current drop rate, combined with annotation by experts in the field of rupture, to annotate the time slice of the rupture in each time data collection.

[0025] Step 2: Data Preparation: Align the multimodal data collected in Step 1 with the labels of the annotated rupture time slices along the time axis. Downsample the higher-sampling rate data to align it with the lower-sampling rate data. Each 10ms time slice contains 0-dimensional diagnostic data and 2D image diagnostic data, as well as rupture information for the current time slice, forming a rupture warning database.

[0026] Step 3, model training: Use the rupture warning database obtained in step 2 to train the rupture warning model, save the optimal rupture warning model obtained through training, and deploy the optimal rupture warning model in real time. The rupture warning model structure is as follows: use a convolutional neural network (CNN) and a long short-term memory (LSTM) network to process 0-dimensional diagnostic data to obtain a first hidden vector; use a video vision transformer model to process plasma two-dimensional image diagnostic data to obtain a second hidden vector; after splicing the first hidden vector and the second hidden vector, classify them through a multi-layer perceptron (MLP) to obtain a rupture warning result;

[0027] Step 4, rupture warning: During the tokamak experiment, the real-time 0-dimensional diagnostic data and 2D image diagnostic data are input into the optimal rupture warning model, and rupture mitigation and device protection are performed based on the output rupture warning results.

[0028] When the tokamak device is in operation, a multi-modal tokamak device rupture warning system is used to perform real-time rupture prediction. Figure 2Figure 1 shows the structure of a multi-modal tokamak device rupture warning system according to the present invention. The system comprises a data acquisition module, a data preparation module, and a rupture warning module. During the experiment, the rupture warning system reads the 0-dimensional diagnostic signal from the tokamak diagnostic system, a front-end signal source, into the data acquisition module in real time via a reflective memory network. Simultaneously, the image acquisition card acquires high-speed 2D image diagnostic signals in real time. The data preparation module then denoises and filters the acquired 0-dimensional diagnostic signal, resizes the high-speed 2D image signal, and aligns the 0-dimensional diagnostic and 2D image signals before sending them to the rupture warning module. In the rupture warning module, the 0-dimensional diagnostic signal is processed using a CNN+LSTM network, and the video image signal is processed using a Video Vision Transformer (ViViT). After obtaining the rupture warning result, the rupture warning result is transmitted in real time via the reflective memory network to the plasma control system, i.e., the actuator, for rupture avoidance or mitigation. When an impending rupture is predicted, the plasma current is steadily reduced via a control coil to minimize damage to the device. When the plasma current cannot be controlled, MGI (massive gas injection) is triggered to rapidly mitigate the plasma rupture.

[0029] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above are only specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A multi-mode tokamak device rupture early warning method, characterized in that: The method comprises the following steps: Step 1. Data acquisition: Collect multimodal data, including 0-dimensional diagnostic data collected and calculated using real-time diagnostics in the tokamak device, and plasma 2D image diagnostic data collected using a camera; annotate the fracture data in the collected multimodal data; Step 2: Data preparation: Align the collected multimodal data with the time axis and divide them into time slices. Each time slice contains 0-dimensional diagnostic data, 2D image diagnostic data, and annotated fracture data information. The above data and information together form a fracture warning database. Step 3: Model training: Use the fracture warning database to train the fracture warning model, and deploy the optimal fracture warning model obtained through training in real time; Step 4: Rupture warning: During the tokamak experiment, the real-time 0D diagnostic data and 2D image diagnostic data are input into the optimal rupture warning model, and rupture mitigation and device protection are performed based on the output rupture warning results. Wherein, in said step 1, the 0-dimensional diagnostic data includes the plasma current Ip, the plasma current error Ip_error, the plasma density Ne, the loop voltage V_loop, and the plasma elongation ratio Kappa; the plasma two-dimensional image diagnostic data collected by the camera includes the tokamak position, and the tokamak position includes any one of an image of the entire plasma area observed along the plasma tangent direction and an image of a partial plasma area observed, wherein the waveband to which the collected plasma two-dimensional image belongs is any one of the visible light band, the infrared band, and the ultraviolet band; said marking of the rupture data in the collected multimodal data includes marking by a method combining automatic judgment of the current change rate and expert experience marking, and marking the time point of the rupture in the collected multimodal data; In step 2, aligning the time axis of the collected multimodal data includes downsampling the high sampling rate data to match the low sampling rate data.

2. A multi-mode tokamak device rupture early warning method according to claim 1, characterized in that: In step 3, the structure of the rupture warning model is as follows: using a convolutional neural network (CNN) and a long short-term memory (LSTM) network to process 0-dimensional diagnostic data to obtain a first hidden vector; using a video vision transformer model to process plasma two-dimensional image diagnostic data to obtain a second hidden vector; after splicing the first hidden vector and the second hidden vector, classifying them through a multi-layer perceptron (MLP) to obtain a rupture warning result.

3. A multi-mode tokamak device rupture early warning method according to claim 2, characterized in that: In step 4, the 0-dimensional diagnostic data is obtained in real time by reading the acquisition system data through the reflective memory card, and the 2D image diagnostic data is obtained in real time through the image acquisition card; the result of the rupture warning is transmitted to the plasma control system in real time through the reflective memory network.

4. A multi-mode tokamak device rupture warning system, applied to the method according to any one of claims 1 to 3, characterized in that: The system comprises: Data acquisition module: used to collect multimodal data, including 0-dimensional diagnostic data collected and calculated using real-time diagnostics in the tokamak device, and plasma 2D image diagnostic data collected using a camera; and to annotate the fracture data in the collected multimodal data; Data preparation module: used to align the time axis of the collected multimodal data and divide it into time slices. Each time slice contains 0-dimensional diagnostic data, 2D image diagnostic data, and annotated fracture data information; Rupture warning module: It is used to input the time-aligned multi-modal data into the optimal rupture warning model during the tokamak experiment, and perform rupture mitigation and device protection based on the output rupture warning results.

Images