Method for real-time fault detection of tokamak magnetic probe

Abstract

The present invention relates to the technical field of fault detection of tokamak magnetic probes, and in particular to a method for real-time fault detection of a tokamak magnetic probe. The technical solution comprises the following steps: collecting neural network model training data to construct a training database, and using physical parameters related to a magnetic probe signal as a model input; using a neural network model having a plurality of encoding layers and decoding layers to capture temporal evolution characteristics of the magnetic probe signal, and predicting a magnetic probe signal in a subsequent time period on the basis of an input time period; and transmitting the model input to a trained neural network model to predict the magnetic probe signal in real time. The present invention proposes the use of an artificial intelligence algorithm to realize fault detection for magnetic probe diagnosis in a tokamak device, so that the calculation accuracy of the model can be improved, and the calculation time of the model can be greatly optimized, thereby making real-time fault detection of magnetic probes become feasible and have strong general applicability.

Description

A method for real-time fault detection of magnetic probes in tokamak Technical Field

[0001] This invention relates to the field of tokamak magnetic probe fault detection technology, and in particular to a method for real-time fault detection of tokamak magnetic probes. Background Technology

[0002] With the continuous development of magnetic confinement fusion research, the safe and stable operation of the device has become a major concern. In a magnetically confined tokamak device, a strong magnetic field exists, generated by currents flowing in the outer conductors (such as longitudinal and poloidal conductors) or the plasma itself. These magnetic fields interact with the plasma, confining the high-temperature plasma within a vacuum chamber. Therefore, in tokamak plasma experiments, magnetic field measurement is one of the most fundamental conditions for stable device operation, and electromagnetic measurement diagnostic systems are a simple and effective tool for measuring magnetic fields and electromagnetic processes. Inversion based on electromagnetic measurement diagnostics can reveal the state of the high-temperature plasma inside the device, thereby enabling stable plasma control. Among these, the magnetic probe is one of the fundamental diagnostic methods in electromagnetic measurement, characterized by its simple measurement method and high accuracy. However, the harsh operating environment of fusion devices can lead to measurement failures in the magnetic probe during long-term operation, affecting the stable operation of the device. To ensure stable operation, it is necessary to monitor the operating status of the magnetic probe in real time.

[0003] Existing magnetic probe fault diagnosis and testing methods have the following drawbacks:

[0004] (1) Evolutionary model construction is difficult

[0005] The tokamak discharge process is extremely complex, and traditional physical evolution methods struggle to accurately simulate it. Existing techniques generally simplify the physical process based on certain assumptions to achieve discharge simulation. However, these simplified physical processes can only accommodate a portion of the discharge process, and when discharge parameters are high, the simulation fails to obtain accurate evolution results. Therefore, specific adjustments to the model are typically needed to adapt to these scenarios.

[0006] (2) The calculation speed is slow and the real-time performance is difficult to guarantee.

[0007] To obtain a more accurate evolutionary discharge model, a more complex physical model needs to be constructed. However, the calculation of these models is difficult to perform quickly based on the current computing power of computers, with calculation times reaching hours or even astronomical levels. Real-time performance cannot be guaranteed, which seriously affects real-time deployment.

[0008] (3) Failure to achieve online detection

[0009] Fault diagnosis and detection of tokamak magnetic probes cannot achieve real-time online detection due to limitations in current technology; calibration is typically performed before the discharge experiment. If a fault occurs during discharge, the discharge will be interrupted and fail. After the discharge failure, researchers will assess the situation and perform calibration to troubleshoot the fault, severely impacting the experimental process.

[0010] In summary, this application proposes a method for real-time fault detection of magnetic probes in tokamak. Summary of the Invention

[0011] The purpose of this invention is to address the problems of high difficulty and slow speed in fault diagnosis and detection of existing magnetic probes in the background art, and to propose a real-time fault detection method for magnetic probes in tokamak.

[0012] The technical solution of the present invention: a method for real-time fault detection of magnetic probes in tokamak, comprising the following steps:

[0013] Step 1: Collect training data for the neural network model to build a training database, using the physical parameters related to the magnetic probe signal as model input;

[0014] Step 2: Use a neural network model with multiple coding and decoding layers to capture the temporal evolution characteristics of the magnetic probe signal and predict the magnetic probe signal for subsequent time periods based on the input time period.

[0015] Step 3: Transmit the model input to the trained neural network model to predict the magnetic probe signal in real time;

[0016] Step 4: By comparing the actual measured magnetic probe signal with the predicted magnetic probe signal, determine whether there is a problem with the actual measured magnetic probe signal based on the set threshold.

[0017] Step 5, Abnormal Signal Feedback Processing: When a problem is detected in the actual measured signal, the predicted signal is used to replace the problematic signal and then output to ensure the stability of the experiment.

[0018] Optionally, in step one, the physical parameters related to the magnetic probe signal include the voltage signals of the PF coil and IC coil and the total plasma current, while the state parameters characterizing the current plasma, such as energy storage Wmhd, loop voltage Vloop, poloidal specific pressure, and internal inductance Li, are used as inputs to the model.

[0019] Optionally, in the step of constructing the training database, the data source is the discharge data of the EAST tokamak device, including multiple discharge configurations such as double zero and single zero, and the total plasma current at the discharge level top section covers the range of 200-500kA.

[0020] Optionally, step two specifically includes: using a Transformer neural network model as the base model to capture the evolutionary characteristics over time, wherein the Transformer neural network model has 4 encoding layers and 6 decoding layers, the input is the time period from 0 to t, and the prediction of the magnetic probe signal evolving from the time period t to t+999.

[0021] Optionally, the input to the Transformer neural network model includes plasma state parameters, which enter the Conv1D layer after Embedding and PE operations, and then undergo encoding and decoding processing through multiple layers of LN, Conv FFN, MHA and Masked MHA operations, and finally obtain the prediction signal through Projection and Linear operations.

[0022] Optionally, in step four, the actual measured signal is compared with the predicted signal. If the difference between the two is greater than a set threshold, the actual measured signal is considered to have a problem; otherwise, it is normal.

[0023] Optionally, in step four, the threshold value in the step of comparing the actual measured signal with the predicted signal is set according to the actual operating environment and the performance requirements of the magnetic probe. The threshold value is set to 0.05, i.e. .

[0024] Optionally, in the abnormal signal feedback processing, after replacing the problematic signal, the output signal is marked.

[0025] Compared with the prior art, this application includes at least one of the following beneficial technical effects:

[0026] (1) By constructing an accurate plasma evolution model to predict the magnetic probe signal, the actual measured value is compared with the predicted value to determine the operating status of the magnetic probe, thereby realizing the fault detection of the magnetic probe.

[0027] (2) In terms of constructing evolutionary models, the Transformer model is used to adapt information containing time evolution and improve the accuracy of model evolution.

[0028] (3) The GPU-based neural network evolution model overcomes the problem of long computation time of traditional evolution models and realizes rapid evolution of plasma fault diagnosis.

[0029] (4) The invention utilizes data-driven artificial intelligence algorithms to replace traditional physical modeling, which improves the accuracy of model prediction and reduces the difficulty of model construction.

[0030] This invention proposes to use artificial intelligence algorithms to detect faults in magnetic probes in tokamak devices. This not only improves the calculation accuracy of the model, but also significantly optimizes the calculation time, making real-time detection of magnetic probe faults feasible and highly universal. Attached Figure Description

[0031] Figure 1 shows a principle block diagram of a real-time fault detection method for a magnetic probe in a tokamak;

[0032] Figure 2 is a schematic diagram of the Transformer neural network model;

[0033] Figure 3 shows the results of predicting 1000 evolution steps using a neural network model trained on a database. Detailed Implementation

[0034] The technical solution of the present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0035] Example

[0036] As shown in Figure 1, the present invention proposes a real-time fault detection method for magnetic probes in tokamaks, the specific steps of which are as follows:

[0037] (1) Collect training data for neural network models and construct a training database.

[0038] Based on the physical context, the contribution sources of the magnetic probe signal are fully considered, including the voltage signals of the PF coil and IC coil, and the total plasma current. Furthermore, the state parameters characterizing the current plasma, such as stored energy Wmhd, loop voltage Vloop, poloidal specific pressure, and internal inductance Li, are also considered as inputs to the model.

[0039] (2) Construct a neural network model and train the neural network model.

[0040] Since this invention aims to obtain a plasma evolution model, which needs to consider temporal evolution, a Transformer neural network model is used as the basic model to capture the temporal evolution characteristics. The Transformer neural network model can effectively capture long-range dependencies in a sequence and can significantly shorten the training and inference cycles through parallel computation. As shown in Figure 2, the Transformer neural network model has 4 encoding layers and 6 decoding layers. The model's input is the time interval 0-t (blue rectangle in Figure 2), and it predicts the magnetic probe signal evolving from t to t+999 (green rectangle in the figure).

[0041] (3) Predict magnetic probe signals in real time using a trained evolutionary model.

[0042] The input required by the neural network model is transmitted to the neural network model to predict the signal of the magnetic probe in real time.

[0043] (4) Compare the actual measured signal with the predicted signal

[0044] The actual measured signal is compared with the predicted signal to determine the current operating status of the magnetic probe diagnostics. If the difference between the two is greater than a set threshold, the actual measured signal is considered to have a problem; otherwise, it is normal.

[0045] (5) Abnormal signal feedback processing

[0046] When a problem is detected in the actual measured signal, the predicted signal is used to replace the problematic signal and then output, thereby ensuring the stability of the experiment.

[0047] Experimental verification

[0048] Based on the currently operating tokamak device, discharge data from December 2022 to June 2024 was selected as the database. Among these, discharge times exceeding 2 seconds resulted in approximately 15,000 historical discharge data points (sampling rate 1kHz). This data includes various discharge configurations such as double-zero and single-zero discharges, and the total plasma current at the discharge level peak ranges from 200 kA to 500 kA. A neural network model was trained based on this database, and the predicted evolution over 1000 steps is shown in Figure 3. The red line represents the actual measured magnetic probe signal, and the blue line represents the predicted magnetic probe signal. The figure shows that the predicted evolution results are largely consistent with the actual measurements. The core of this invention is the need for an accurate evolution model to achieve fault detection of the magnetic probe. This example demonstrates that the method in this invention can accurately achieve evolution prediction, thereby enabling equipment fault monitoring. The model evolution prediction time in this example is approximately 1 ms, verifying that this invention has real-time monitoring capabilities.

[0049] This invention proposes using artificial intelligence algorithms to detect faults in magnetic probe diagnostics within tokamak devices. This not only improves the computational accuracy of the model but also significantly optimizes its computation time, making real-time fault detection of magnetic probes practically feasible. Furthermore, the fault detection method proposed in this invention (including the establishment of a data-driven evolutionary model) is not limited to magnetic probe diagnostic fault detection in tokamak devices but is also applicable to other diagnostic systems and even other fields requiring real-time monitoring of the operating status of mechanical equipment, demonstrating strong versatility.

[0050] The above specific embodiments are merely several optional embodiments of the present invention. Based on the technical solutions of the present invention and the relevant teachings of the above embodiments, those skilled in the art can make various alternative improvements and combinations to the above specific embodiments.

Claims

1. A method for real-time fault detection of magnetic probes in tokamaks, characterized in that, Includes the following steps: Step 1: Collect training data for the neural network model to build a training database, using the physical parameters related to the magnetic probe signal as model input; Step 2: Use a neural network model with multiple coding and decoding layers to capture the temporal evolution characteristics of the magnetic probe signal and predict the magnetic probe signal for subsequent time periods based on the input time period. Step 3: Transmit the model input to the trained neural network model to predict the magnetic probe signal in real time; Step 4: By comparing the actual measured magnetic probe signal with the predicted magnetic probe signal, determine whether there is a problem with the actual measured magnetic probe signal based on the set threshold. Step 5, Abnormal Signal Feedback Processing: When a problem is detected in the actual measured signal, the predicted signal is used to replace the problematic signal and then output.

2. The method for real-time fault detection of a magnetic probe in a tokamak according to claim 1, characterized in that, In step one, the physical parameters related to the magnetic probe signal include the voltage signals of the PF coil and IC coil and the total plasma current. At the same time, the state parameters of the current plasma, such as energy storage Wmhd, loop voltage Vloop, poloidal specific pressure and internal inductance Li, are used as inputs to the model.

3. The method for real-time fault detection of a magnetic probe in a tokamak according to claim 2, characterized in that, In the step of constructing the training database, the data source is the discharge data of the EAST tokamak device, including multiple discharge configurations such as double zero and single zero, and the total plasma current at the discharge level top section covers the range of 200-500kA.

4. The method for real-time fault detection of a magnetic probe in a tokamak according to claim 1, characterized in that, Step two specifically includes: using the Transformer neural network model as the basic model to capture the evolutionary characteristics over time, wherein the Transformer neural network model has 4 encoding layers and 6 decoding layers, the input is the time period from 0 to t, and the prediction is made for the magnetic probe signal evolving from the time period t to t+999.

5. A method for real-time fault detection of a magnetic probe in a tokamak according to claim 4, characterized in that, The input to the Transformer neural network model includes plasma state parameters, which enter the Conv1D layer after Embedding and PE operations. Then, they undergo encoding and decoding processing through multiple layers of LN, Conv FFN, MHA, and Masked MHA operations, and finally obtain the prediction signal through Projection and Linear operations.

6. The method for real-time fault detection of a magnetic probe in a tokamak according to claim 1, characterized in that, In step four, the actual measured signal is compared with the predicted signal. If the difference between the two is greater than a set threshold, the actual measured signal is considered to have a problem; otherwise, it is normal.

7. A method for real-time fault detection of a magnetic probe in a tokamak according to claim 6, characterized in that, In step four, the threshold for comparing the actual measured signal with the predicted signal is set according to the actual operating environment and the performance requirements of the magnetic probe. The threshold is set to 0.05, i.e. .

8. The method for real-time fault detection of a magnetic probe in a tokamak according to claim 1, characterized in that, In the abnormal signal feedback processing, after replacing the problematic signal, the output signal is marked.

Images