Tokamak plasma rupture prediction method based on data enhancement
Through the tokamak plasma rupture prediction method based on data augmentation, the neural network is optimized using Gaussian noise model and ACO-BP algorithm, combined with an integrated learning framework, the problem of insufficient rupture prediction accuracy and early warning time in the existing technology is solved, and higher rupture prediction accuracy and safe operation of the device is achieved.
Patent Information
- Application Number
- CN202411422525.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-12
- Publication Date
- 2025-02-28
AI Technical Summary
The prior art is difficult to accurately predict the tokamak plasma rupture event, resulting in insufficient prediction accuracy and early warning time, affecting the safe operation of the device.
The tokamak plasma rupture prediction method based on data augmentation is adopted to generate a Gaussian noise model to perform data augmentation, combine the ACO algorithm to optimize the weight of the BP neural network, and an integrated learning framework is adopted to solve the deviation problem caused by data augmentation.
Improve the accuracy of rupture prediction, generate more rupture samples, and optimize the training process of the prediction model, thereby providing operators with sufficient time to take control and intervention measures to avoid or mitigate the impact of rupture and ensure the safe operation of the device.
Smart Images
Figure 00000000_0000_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of nuclear fusion, and in particular relates to a tokamak plasma rupture prediction method based on data enhancement. Background Art
[0002] Magnetic confinement controlled nuclear fusion is currently one of the most feasible ways to solve future energy problems.
[0003] Tokamak is the most popular magnetic confinement device currently. It uses magnetic field to confine particles and achieve fusion conditions for output. However, plasma rupture is a common event during the operation of tokamak experimental devices, and it is even an inevitable emergency. When plasma ruptures, a large amount of energy (thermal energy and magnetic energy) is deposited on the first wall material or other components of the device in the form of heat flow, corona current and escaped electrons in a very short time, which will cause damage to the device over a long period of time. As the size of the fusion device increases and the operating parameters increase, the harm of plasma rupture to the device will be greater. Therefore, it is very important for the safety of the device to provide early warning of rupture and implement effective rupture avoidance or mitigation.
[0004] At present, the research on rupture prediction mainly includes two methods, including physical model prediction and data-driven prediction. However, due to the complexity and nonlinear characteristics of plasma rupture development, the understanding of the physical mechanism of rupture is not clear enough; in addition, the rupture occurs in an extremely short time and the relationship between various factors is very complex. It is difficult for traditional machine learning models to accurately capture these nonlinear relationships, resulting in unsatisfactory prediction accuracy and warning time; although big data-driven rupture prediction can effectively avoid the above problems, due to the relatively limited rupture data of future large-scale tokamaks, it is difficult for data-driven models to learn the characteristics of rupture precursors. Summary of the invention
[0005] The purpose of the embodiments of the present invention is to provide a tokamak plasma disruption prediction method based on data enhancement, aiming to solve the problems existing in the above-mentioned background technology.
[0006] The embodiment of the present invention is implemented as follows: a method for predicting tokamak plasma rupture based on data enhancement, comprising the following steps:
[0007] S1. Use the original data combined with the normal distribution formula to generate a Gaussian noise model for data enhancement;
[0008] S2, adding normally distributed random values to the original data, and combining the data with Gaussian noise added with the original data to form an enhanced data set;
[0009] S3, optimizing the initial weights of the BP algorithm through the ACO algorithm;
[0010] S4. Integrated learning framework to solve the bias problem caused by data enhancement.
[0011] Preferably, it also includes a BP neural network, wherein the BP neural network has multiple layers of feedforward neurons.
[0012] Preferably, the BP neural network comprises: an input layer, an output layer and a plurality of hidden layers.
[0013] Preferably, the probability density function of the Gaussian noise model is:
[0014]
[0015] Here, the mean is zero and σ is the standard deviation.
[0016] Preferably, the ACO algorithm in S3 combines the characteristics of distributed computing, positive feedback mechanism and greedy search, and uses pheromones to select paths by simulating the behavior of ants looking for food.
[0017] Preferably, in the ACO-BP algorithm, the network weights are recorded as q1, q2, ..., qp, respectively, and the value range of any parameter qi (1, 2, ..., p) is divided into N sub-intervals to form a set Iqi, and the boundary value of each sub-interval is used as a temporary candidate value. The pheromone in each sub-interval is τj(Iqi)(t), j = 1, 2, ..., N;
[0018] If the total number of ants in the system is K, the probability that any ant k randomly selects element j from the set Iqi is:
[0019]
[0020] The expression of pheromone update is:
[0021]
[0022] Where ρ(0<ρ<1) is the pheromone residual coefficient; Q is a constant used to adjust the update speed of pheromone; ek is the maximum value of the output error of each training sample when the element selected by ant k in the set is used as the model weight.
[0023] Preferably, in S4, a more powerful and comprehensive model is obtained by combining multiple weak learning models.
[0024] A tokamak plasma rupture prediction method based on data enhancement provided in an embodiment of the present invention uses data enhancement technology. The model based on data enhancement has excellent generalization function, can generate more rupture samples under the condition of limited data, and optimize the training process of the prediction model, thereby improving the accuracy of rupture prediction. This provides operators with sufficient time to take corresponding control and intervention measures to avoid or reduce the impact of the rupture, thereby ensuring the safe operation of the tokamak device. BRIEF DESCRIPTION OF THE DRAWINGS
[0025] Figure 1 A schematic flow chart of a method for predicting tokamak plasma rupture based on data enhancement provided by an embodiment of the present invention;
[0026] Figure 2 A schematic diagram of an integrated learning method in a tokamak plasma disruption prediction method based on data enhancement provided by an embodiment of the present invention;
[0027] Figure 3 A schematic diagram of a ROC curve in a tokamak plasma rupture prediction method based on data enhancement provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0028] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.
[0029] The specific implementation of the present invention is described in detail below in conjunction with specific embodiments.
[0030] like Figure 1-3 As shown, a structural diagram of a method for predicting tokamak plasma disruption based on data enhancement provided by an embodiment of the present invention includes the following steps:
[0031] S1. Use the original data combined with the normal distribution formula to generate a Gaussian model for data enhancement;
[0032] Also included is a BP neural network with multiple layers of feedforward neural networks, the BP neural network comprising: an input layer, an output layer and multiple hidden layers;
[0033] The probability density function of the Gaussian noise model is:
[0034]
[0035] Here, the mean is zero and σ is the standard deviation.
[0036] S2, adding normally distributed random values to the original data, and combining the data with Gaussian noise added with the original data to form an enhanced data set;
[0037] S3, optimizing the initial weights of the BP algorithm through the ACO algorithm;
[0038] The ACO algorithm combines the characteristics of distributed computing, positive feedback mechanism and greedy search, and uses pheromones to select paths by simulating the behavior of ants looking for food;
[0039] In the ACO-BP algorithm, the following assumptions are made: the network weights are recorded as q1, q2, ..., qp, and the value range of any parameter qi (1, 2, ..., p) is divided into N subintervals to form a set Iqi. The boundary value of each subinterval is used as a temporary candidate value. The pheromone in each subinterval is τj(Iqi)(t), j = 1, 2, ..., N;
[0040] If the total number of ants in the system is K, the probability that any ant k randomly selects element j from the set Iqi is:
[0041]
[0042] The expression of pheromone update is:
[0043]
[0044] Where ρ(0<ρ<1) is the pheromone residual coefficient; Q is a constant used to adjust the update speed of pheromone; ek is the maximum value of the output error of each training sample when the element selected by ant k in the set is used as the model weight.
[0045] S4, integrated learning framework to solve the bias problem caused by data enhancement;
[0046] By combining multiple weak learning models, a more powerful and comprehensive model is obtained. Although data augmentation improves the accuracy of the model, the increase in rupture data not only expands the training set, but may also introduce some unexpected biases. As the model receives more rupture-related information, its sensitivity to the data increases, making it susceptible to subtle fluctuations, which can be addressed by adopting an ensemble learning approach.
[0047] In the effective supervised learning algorithm of machine learning, our goal is to train a stable model that performs well in all aspects, but the actual situation is often not as expected. Sometimes we can only get multiple preferred models (weak learning models that perform better in some aspects). Through ensemble learning, when a weak learner makes an incorrect prediction, other weak learners can also correct these errors, as shown in the attached figure of the manual. Figure 2 As shown in the figure, since a single learner is regarded as a weak learner, ensemble learning forms a strong learner, which can effectively alleviate the deviation introduced by the data enhancement method.
[0048] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the present invention can be implemented in other specific forms without departing from the spirit or essential features of the present invention. Therefore, the embodiments should be considered exemplary and non-restrictive in all respects, and the scope of the present invention is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims be included in the present invention.
[0049] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.
Claims
1. A method for predicting tokamak plasma disruption based on data enhancement, characterized in that: The following steps are involved: S1. Use the original data combined with the normal distribution formula to generate a Gaussian noise model for data enhancement; S2, adding normally distributed random values to the original data, and combining the data with Gaussian noise added with the original data to form an enhanced data set; S3, optimizing the initial weights of the BP algorithm through the ACO algorithm; S4. Integrated learning framework to solve the bias problem caused by data enhancement.
2. A method for predicting tokamak plasma disruption based on data enhancement according to claim 1, characterized in that: It also includes a BP neural network, which has multiple layers of feedforward neurons.
3. A method for predicting tokamak plasma disruption based on data enhancement according to claim 2, characterized in that: The BP neural network includes: an input layer, an output layer and multiple hidden layers.
4. The method for predicting tokamak plasma disruption based on data enhancement according to claim 1, characterized in that: The probability density function of the Gaussian noise model is: Here, the mean is zero and σ is the standard deviation.
5. The method for predicting tokamak plasma disruption based on data enhancement according to claim 1, characterized in that: The ACO algorithm in S3 combines the characteristics of distributed computing, positive feedback mechanism and greedy search, and uses pheromones to select paths by simulating the behavior of ants looking for food.
6. The method for predicting tokamak plasma disruption based on data enhancement according to claim 1, characterized in that: In the ACO-BP algorithm, the network weights are recorded as q1, q2, ..., qp, and the value range of any parameter qi (1, 2, ..., p) is divided into N sub-intervals to form a set Iqi. The boundary value of each sub-interval is used as a temporary candidate value. The pheromone in each sub-interval is τj(Iqi)(t), j = 1, 2, ..., N; If the total number of ants in the system is K, the probability that any ant k randomly selects element j from the set Iqi is: The expression of pheromone update is: Where ρ(0<ρ<1) is the pheromone residual coefficient; Q is a constant used to adjust the update speed of pheromone; ek is the maximum value of the output error of each training sample when the element selected by ant k in the set is used as the model weight.
7. The method for predicting tokamak plasma disruption based on data enhancement according to claim 1, characterized in that: In S4, a more powerful and comprehensive model is obtained by combining multiple weak learning models.