Fusion reactor solid-state switch network unit redundancy turn-off control method

Abstract

The invention discloses a fusion reactor solid-state switch network unit redundancy turn-off control method, and relates to the technical field of controllable fusion reactors. A redundant reverse turn-off unit formed by multiple groups of pulse capacitors and corresponding controllable switch devices is connected in parallel to two sides of a power device of a solid-state switch network unit, meanwhile, loop current is sampled in real time, and a loop current value after control delay is predicted based on a time sequence prediction model. And according to the predicted current, selecting a corresponding pulse capacitor unit to perform reverse discharge, thereby realizing accurate control of the reverse turn-off current amplitude. The arithmetic unit adopts a double-processor architecture, the real-time processor is used for executing prediction model reasoning and turn-off control, and the high-performance processor is used for performing online learning and parameter updating on a prediction model and periodically issuing model parameters to the real-time processor. According to the invention, the turn-off success rate can be improved when the power device fails to be turned off or is turned off abnormally, and the robustness and reliability of the system are enhanced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of controllable fusion reactor, and particularly relates to a redundancy turn-off control method for a fusion reactor solid-state switch network unit. BACKGROUND

[0002] The switch network unit is a igniter of a fusion reactor. A fusion reactor with a tokamak as a core generates a pulsed high voltage by active breaking, so as to excite a plasma current. The unit needs to complete more than 100,000 times of direct current breaking in the life cycle of the device. At present, a commutation switch network composed of full-controlled solid-state switch devices is generally used.

[0003] With the increase of the scale of the fusion device, the rated current of the magnet power supply continues to rise. In order to meet the capacity demand, a parallel structure of multiple solid-state switch devices is often used. However, the inconsistency of the parameters of the parallel devices and the difference of the loop stray parameters may cause a dynamic current sharing problem, and may cause the breaking failure of individual devices after long-term operation. Once the breaking failure occurs, the main loop current will be concentrated to the failed device for commutation, causing damage to the device and failure of the plasma start, which seriously affects the safety and operation reliability of the device.

[0004] The parallel RC absorption circuit or the fixed reverse breaking circuit used in the traditional scheme has the problems of unadjustable reverse breaking current, easy introduction of overcurrent impact, and difficulty in adapting to the rapidly changing current, because the current rate of change during the breaking process of the fusion power supply is very high (more than 10 kA / ms). In addition, the existing scheme is based on instantaneous sampling current control, and the influence of sampling and control delay on the breaking effect is not considered.

[0005] Therefore, it is necessary to propose a new breaking method and device which can realize accurate control of the reverse breaking current and has self-adaptive ability even in the presence of control delay. SUMMARY

[0006] To solve the above technical problems, the present application provides a redundancy turn-off control method for a fusion reactor solid-state switch network unit. The specific technical scheme is as follows:

[0007] A redundancy turn-off control method for a fusion reactor solid-state switch network unit, applied to a direct current circuit breaker system containing semiconductor power devices, the method comprising the following steps:

[0008] Step 1, real-time acquisition of a main loop current signal flowing through the semiconductor power device;

[0009] Step 2, according to the acquired main loop current historical data, a prediction current value after the control delay is calculated by using a prediction model;

[0010] Step 3, when detecting that the semiconductor power device fails to turn off or has an abnormal trend of turning off, determining a required reverse turn-off current amplitude according to the predicted current value;

[0011] Step 4, selecting at least one group of pulse capacitor units from a plurality of groups of pulse capacitor units in parallel or series-parallel configuration according to the reverse turn-off current amplitude;

[0012] Step 5, triggering a controllable switching device corresponding to the pulse capacitor unit to make the pulse capacitor unit discharge reversely to the main loop to form a controlled reverse turn-off current, thereby achieving forced turn-off of the semiconductor power device.

[0013] The present application has the following beneficial effects:

[0014] The present application has the following beneficial effects: BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 is a system overall diagram;

[0016] Figure 2 is a schematic diagram of a pre-charge pulse capacitor array;

[0017] Figure 3 is a system operation logic. DETAILED DESCRIPTION

[0018] Embodiments of the present application are described in detail below, examples of which are shown in the accompanying drawings. The embodiments described below by reference to the accompanying drawings are exemplary and are intended to explain the present application, and cannot be understood as a limitation of the present application.

[0019] Referring to Figure 1 is an embodiment of a fusion reactor solid-state switch network unit redundant turn-off control method of the present application, taking a 47kA all-solid-state circuit breaker of a BEST fusion device switch network unit as an example, the main loop of which is 16 IGCTs in parallel, and a fusion reactor solid-state switch network unit redundant turn-off control method proposed by the present application is adopted.

[0020] Referring to Figure 1 The present application is composed of the following parts, which are a current sensor, a pre-charge pulse capacitor array, a high-performance processor, and a real-time calculation processor. The current sensor is used to collect the current flowing through the power device in the main loop in real time and transmit it to the two processors; the pre-charge pulse capacitor array is composed of small pulse capacitors and pulse thyristors, referring to Figure 2By controlling the turn-on and turn-off of the thyristor, reverse currents of different amplitudes can be generated; the real-time computing processor is used to run a current prediction model to predict the current value at a future time instant in real time after receiving the current signal; the high-performance processor receives the current signal to train the current prediction model and sends the model parameters to the real-time computing processor to update the parameters of the prediction model. The method comprises the following steps, as shown in Figure 3

[0021] Step 1: The current sensor acquires the current signal flowing through the main circuit power device in real time and transmits it to the real-time computing processor and the high-performance processor. The current flowing through the main circuit power device at time instant k is acquired in real time , and the sampling period is .

[0022] (1)

[0023] Step 2: According to the acquired main circuit current historical data, the real-time computing processor calculates the predicted current value after the control delay by using the prediction model; the high-performance processor trains the prediction model according to the current data and periodically transmits the obtained model parameters to the real-time computing processor.

[0024] The sampling period is , and the prediction model is:

[0025] (2)

[0026] In the formula, is the adaptive gain, which can be updated slowly by the high-performance processor, is the first-order differential of the main circuit current at time instant k, is the first-order differential of the main circuit current at time instant k+1, is the second-order differential of the main circuit current at time instant k, is the second-order differential of the main circuit current at time instant k+1, is the error between the predicted value and the actual value of the main circuit current at time instant k, which is represented by the following formula:

[0027] (3)

[0028] In the formula, is the predicted value of the main circuit current at time instant k.

[0029] The update rule of the parameters in the model by the high-performance processor is as follows:

[0030] (4)

[0031] In the above formula, is the adaptive gain at time instant k,​ is the adaptive gain for k+1 time, η is the online learning type adaptive gain, J is the error in a predetermined time, which can be expressed by the following formula:

[0032] (5)

[0033] In the formula, N is the period of model updating, and the model is updated once after N current values are collected.

[0034] The parameter revision formula can be expressed as follows by expanding the above formula:

[0035] (6)

[0036] Step 3, when the power device opening fails, the real-time calculation processor rapidly calculates the current estimated value at the next time according to the current current value, and selects the switch combination in the pulse capacitor array based on this, and finally outputs the reverse current matched with the fault current. In order to quickly obtain the combination result at the time of fault, the dichotomy table method is adopted in the application, that is, the reverse current values generated by all combinations are calculated in advance and stored in the memory of the real-time calculation processor according to the current size, when the fault occurs, the combination result can be quickly obtained according to the size of the fault current prediction value.

[0037] Step 4, selecting at least one pulse capacitor unit from a plurality of pulse capacitor units in parallel or series-parallel configuration according to the reverse turn-off current amplitude.

[0038] Step 5, triggering the controllable switching device corresponding to the pulse capacitor unit, making the pulse capacitor unit discharge reversely to the main loop, forming a controlled reverse turn-off current, thereby realizing the forced turn-off of the power device.

[0039] Further, the pulse capacitor unit includes at least one pulse capacitor, and each pulse capacitor unit is in series with a pulse thyristor.

[0040] Further, the pulse capacitor unit constitutes a pulse capacitor array with capacity that can be configured in stages, and different pulse capacitor units have different equivalent capacities, and the reverse turn-off current is adjusted in stages or continuously by combination gating.

[0041] Further, the predicted current value is the current value at the future time after the sampling delay, calculation delay and switch triggering delay are compensated.

[0042] Further, the prediction model is a dynamic prediction model established based on the time sequence of the main loop current, and the input at least includes the current value at the current and historical sampling time.

[0043] Further, the running and updating of the prediction model is realized by a dual-processor architecture, wherein a real-time processor is used to perform prediction model inference and reverse shutdown control; a high-performance processor is used to perform online training and parameter updating of the prediction model, and periodically transmit the updated model parameters to the real-time processor.

[0044] Further, the parameters of the prediction model are adjusted online by an adaptive updating law based on prediction error.

[0045] Further, the adaptive updating law contains a parameter updating gain for adjusting the influence of the prediction error on the correction amplitude of the model parameters.

Claims

1. A method for controlling redundancy shutdown of a fusion reactor solid-state switch network unit, applied to a direct current circuit breaker system comprising semiconductor power devices, characterized in that, The method comprises the following steps: Step 1, collecting the main loop current signal flowing through the semiconductor power device in real time; Step 2, calculating the predicted current value after the control delay according to the collected main loop current historical data and using a prediction model; Step 3, when the semiconductor power device shutdown failure or abnormal trend is detected, determining the required reverse shutdown current amplitude according to the predicted current value; Step 4, selecting at least one group of pulse capacitor units from the multiple groups of pulse capacitor units in parallel or series-parallel configuration according to the reverse shutdown current amplitude; Step 5, triggering the controllable switching device corresponding to the pulse capacitor unit to make the pulse capacitor unit discharge reversely to the main loop to form a controlled reverse shutdown current, thereby achieving forced shutdown of the semiconductor power device.

2. The method according to claim 1, wherein, The pulse capacitor unit comprises at least one pulse capacitor, and each pulse capacitor unit is connected in series with a pulse thyristor.

3. The method according to claim 1 or 2, wherein, The pulse capacitor unit constitutes a pulse capacitor array with capacity that can be configured in stages, and different pulse capacitor units have different equivalent capacities, and the reverse shutdown current is adjusted in stages or continuously by combination gating.

4. The method of claim 1, wherein, The predicted current value is the current value at the future time after compensation of the sampling delay, calculation delay and switch triggering delay.

5. The method of claim 1, wherein, The prediction model is a dynamic prediction model established based on the time series of the main loop current, and the input at least includes the current values at the current and historical sampling times.

6. The method of claim 1, wherein, The operation and update of the prediction model are realized by a dual-processor architecture, wherein a real-time processor is used to perform prediction model inference and reverse shutdown control; a high-performance processor is used to perform online training and parameter update of the prediction model, and periodically transmits the updated model parameters to the real-time processor.

7. The method according to claim 6, wherein, The parameters of the prediction model are adjusted online by an adaptive update law based on the prediction error.

8. The method according to claim 7, wherein, The adaptive update law contains a parameter update gain for adjusting the influence of the prediction error on the correction amplitude of the model parameters.

9. The method of claim 5, wherein the method further comprises: The sampling period is The prediction model is: ； In the formula, is the adaptive gain, is the current flowing through the main circuit power device at time k, is the first-order differential of the main circuit current at time k, is the first-order differential of the main circuit current at time k+1, is the second-order differential of the main circuit current at time k, is the second-order differential of the main circuit current at time k+1, is the error between the predicted value and the actual value of the main circuit current at time k, and is expressed by the following formula: ； In the formula, is the predicted value of the main circuit current at time k.

10. The method of claim 8, wherein the method further comprises: The update rule of the model parameters by the high-performance processor is as follows: ； In the above formula, is the adaptive gain at time k, is the adaptive gain at time k+1, η is an online learning type adaptive gain, and J is an error in a preset time, which can be expressed by the following formula: ； In the formula, N is the period of model update, and the model is updated once after collecting N current values.

Images