A method for precisely controlling the influence of a beam-dump system firing on a proton therapy process

Abstract

The application discloses a method for precisely controlling a proton treatment process, comprising the following steps: selecting a feeding voltage fluctuation amplitude greater than or equal to U0 as a late fault prediction standard; collecting data of a leakage current, a cavity vacuum degree and a feeding voltage value, and establishing a time sequence sample database; constructing and training an LSTM-based time sequence prediction model; collecting electrostatic deflection plate leakage current and vacuum degree data at a current t time point in real time; using the data collected at the current t time point as input of the prediction model, and predicting a feeding voltage fluctuation amplitude at a t+1 time point through the prediction model; after predicting that a fault is about to occur, issuing an instruction through a PLC to control a central area vertical deflection plate to shut off a beam current; and after predicting that the fault is eliminated, issuing a reset signal; through the combination of the two predictions, the application excludes factors causing instability of the beam current in the treatment process, and immediately resumes the treatment after the factors causing instability of the beam current are excluded, so that the proton treatment process is precisely controlled.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of accelerators, and particularly relates to a method for precisely controlling the influence of sparking of an extraction system on a proton treatment process. BACKGROUND

[0002] With the progress of technology, the development of economic foundation and the improvement of people's living standards, the demand for proton treatment is increasingly widespread. Internationally, the conventional proton treatment equipment needs a superconducting cyclotron to generate protons. The cyclotron is a core component of the proton treatment device, and integrates multiple systems, including a main magnet, a high frequency, a vacuum, an ion source, a beam extraction, a beam line, a treatment device, a beam monitoring and control system and the like.

[0003] The motion of the proton beam in the cyclotron can be divided into three stages: injection, acceleration and extraction. The electrostatic deflection plate is a key component of the extraction system of the cyclotron. The electrostatic deflection plate forms a strong electric field by using a high-voltage power supply to generate a high voltage, so that the beam trajectory is deviated, and then the beam is extracted to the beam line outside the accelerator, and then the beam is transported to the treatment head through the beam line.

[0004] The electrostatic deflection plate is composed of a cutting plate and a negative high-voltage electrode, and a very high uniformity electric field is generated between the two. When the electrostatic deflection plate is affected by the surrounding environmental factors and its own physical properties, high-voltage breakdown sparking discharge phenomenon occurs. When high-voltage sparking discharge occurs in the gap between the deflection plates, the leakage current increases, the stability of the cavity vacuum degree decreases, and the pressure drop in the loop increases, thereby affecting the electric field in the electrostatic deflection plate, so that the flow intensity of the extraction beam from the electrostatic deflection plate to the treatment head fluctuates, and the precision of the proton treatment process is affected.

[0005] The difficulty in realizing the precision of the proton treatment process lies in the contradiction between the instability of the beam flow intensity and the precision of the proton treatment. On the one hand, the instability of the beam flow intensity cannot be avoided because the environmental factors are irresistible. During the daily operation of several hours of treatment, high-voltage breakdown sparking phenomenon will inevitably occur due to environmental factors or sudden changes in its own physical properties. On the other hand, in the proton treatment process, it is necessary to accurately control the beam flow intensity and the beam loading time at each treatment position. The instability of the beam flow intensity will lead to poor irradiation effect at some point positions, thereby affecting the overall treatment effect. SUMMARY

[0006] The present application proposes a method for precisely controlling the influence of sparking of an extraction system on a proton treatment process to solve the problem of precision of the proton treatment process.

[0007] The present application proposes the following technical solutions to solve the technical problems:

[0008] A method for precisely controlling the influence of sparking of an extraction system on a proton therapy process, characterized by comprising the following steps:

[0009] Step one, selecting a feed-in voltage fluctuation amplitude greater than or equal to U0 as a later fault prediction standard;

[0010] Step two, entering the electrostatic deflection plate into a stabilized voltage working state;

[0011] Step three, collecting data of leakage current, cavity vacuum degree and feed-in voltage value of the electrostatic deflection plate in the stabilized voltage working state, and establishing a time sequence sample database; the data collection period T of the time sequence sample database is greater than or equal to the prediction model calculation time t 预测 + the time t of issuing a "pause" instruction by the PLC when a fault occurs PLC + the maximum value max of the two times: the high-voltage power supply voltage adjustment time t 电源 of the center zone opening and closing current and the response instruction time t 治疗头 of the treatment head, and the data sampling period T is less than or equal to the duration of a single sparking fault;

[0012] Step four, constructing a time sequence prediction model based on LSTM, inputting the sample database into the model, and obtaining an input-output model for sparking fault prediction of the high-voltage extraction electrode of the cyclotron through model training, model evaluation and model optimization; the input is the input data collected at the current t time point, and the output is the voltage fluctuation amplitude at the t+1 time point predicted by the model; the actual time is the data sampling period T*t, and t at the t time point is the time sequence number;

[0013] Step five, collecting real-time data of the leakage current and vacuum degree of the electrostatic deflection plate at the current t time point;

[0014] Step six, using the data collected at the current t time point as the input of the prediction model, and predicting the feed-in voltage fluctuation amplitude at the t+1 time point through the prediction model;

[0015] Step seven, whether the feed-in voltage fluctuation amplitude at the t+1 time point is greater than U0, if not, return to step six, and if yes, continue to step eight;

[0016] Step eight, after detecting that a fault is about to occur, issuing a fault signal, and issuing an instruction through the PLC to control the center zone vertical deflection plate to shut off the beam current and control the treatment head at the treatment end to pause the action while recording the beam pause time T*△t,△t being the total sum of time points experienced from pausing to resuming of the treatment, and the initial value being 0;

[0017] Step nine,△t=t+1 time point, judging whether the feed-in voltage fluctuation amplitude at the t+1 time point is greater than U0, if yes, repeating step nine, and if no, continuing to step ten;

[0018] Step ten, after detecting the failure elimination, a reset signal is sent out, the beam is opened by the PLC instruction control center vertical deflection plate, the treatment head at the treatment end is restored to work, and the pause time T*△t is fed back to the treatment head program, the pause time T*△t is the time from the failure to the failure elimination; the beam intensity in the working program of the treatment head is changed from f(T*t) to f(T*t-T*△t); wherein, T*t is the current time;

[0019] Further, the fluctuation amplitude of the selected feed-in voltage in step one is as follows:

[0020] 1) When the fluctuation amplitude of the feed-in voltage is greater than U1, the beam fluctuation of the treatment end beam reaches about 10%;

[0021] 2) When the leakage current is greater than a certain value, and the fluctuation amplitude of the feed-in voltage is greater than U2, the leakage current arc will cause damage to the high-voltage feed-in ball head part of the electrostatic deflection plate;

[0022] 3) The fluctuation amplitude of the feed-in voltage U0=min(U1,U2).

[0023] Further, the process of establishing the time sequence sample database in step three is as follows:

[0024] 1) Use a high-speed data acquisition card to collect the leakage current i, the cavity vacuum degree p and the feed-in voltage value U in the stable voltage working state as the original data sequence;

[0025] 2) The data cavity vacuum degree p and the feed-in voltage value U are processed to obtain the time sequence database, which includes time sequence number t, leakage current i, cavity vacuum degree change amount△p and feed-in voltage fluctuation amplitude△U;

[0026] 3) The data is divided into training set and test set according to the time sequence with 1:1 of the first half time sequence after the second half time sequence, which is used for training and testing of the prediction model;

[0027] 4) The leakage current i and the cavity vacuum degree change amount△p of the training set are used as input variables, and the feed-in voltage fluctuation amplitude△U is used as an output variable to form a time sequence sample database;

[0028] 5) All features of the training set and the test set are normalized;

[0029] 6) The data of the training set is converted into a supervised learning problem, and by converting into a supervised learning problem, the output variable at the t+1 time point is predicted by the input variable at the t time point.

[0030] Supplementary note 1

[0031] Each time point corresponds to an actual time, for example: if the t time point is 1, the time corresponding to the t time point = T*t, T = 150ms, if the time point = t+1, the time corresponding to the t+1 time point = T*2 = 300ms;

[0032] Further, the step four is to construct an LSTM-based time series prediction model, and the specific process is as follows:

[0033] 1) Build an LSTM model, which is divided into an input layer, a hidden layer and an output layer. There are 2 neurons in the input layer, 50 neurons in the initial setting of the hidden layer, and 1 neuron in the output layer.

[0034] 2) Input the training set data of the time series sample database into the LSTM model for training to obtain a training set data model for predicting the output data of the next time point based on the input data of the previous time point.

[0035] 3) Test the training set data model training result using the test set data of the time series sample database, and evaluate the accuracy of the training set data model through the test.

[0036] Further, the accuracy of the training set data model is evaluated in the process 3) of step four, and the specific steps are as follows:

[0037] 1) Calculate the error loss with the prediction result and the test set data;

[0038] 2) When the error loss exceeds the threshold value, adjust and optimize the training set data model.

[0039] The adjustment and optimization of the training set data model is specifically: the model is optimized by adjusting the number of hidden layer neurons to reduce the prediction error loss, and an input-output model for predicting the sparking fault of the high-voltage extraction electrode of the cyclotron is obtained. The input-output model for predicting the sparking fault of the high-voltage extraction electrode of the cyclotron after completion of the training can realize the prediction of the output variable feed-in voltage fluctuation amplitude △U at the t+1 time point by inputting the input variable leakage current i and the cavity vacuum degree change amount △p collected at the current t time point.

[0040] Further, the step eight has the following specific process:

[0041] 1) After detecting that a fault is about to occur, increase the voltage difference of the central vertical deflection plate through the PLC control center to form an electric field to make the central beam current off, and at the same time, the PLC controls the proton accelerator treatment head end to pause;

[0042] 2) Record the pause time T*△t(△t initial value is 0) from the occurrence of the fault to the elimination of the fault.

[0043] Further, the step ten specific process as follows:

[0044] 1) after detecting the failure elimination, by PLC delay a set time, control center area vertical deflection plate voltage difference drop to 0, so that the center area beam current stop off and open again;

[0045] 2) at the same time, PLC control proton accelerator treatment head end work begins, and the pause time T*△t feedback to the treatment head program, the beam current intensity in the treatment head work program is changed from f (T*t) to f (T*t-T*△t), that is, in the stable voltage state of electrostatic deflection plate, the treatment work is restored to normal at the moment of failure elimination;Said T*t is the current time;The pause time T*△t is the time from the failure to the failure elimination.

[0046] Further, the delay a set time is the data acquisition period T of the time sequence sample database minus the prediction model calculation time t 预测 - the time t of PLC issuing "pause" instruction when the failure occurs PLC - the maximum value max of two times.

[0047] Advantages and effects of the application

[0048] 1, the application combines "two predictions", uses a time series model to predict that a failure will occur at the t+1 time point, and pauses treatment at the t time point;Use a time series model to predict that the failure has been eliminated at the t+1 time point, and resume treatment at the t+1 time point, so as to exclude the factors of unstable beam current intensity in the treatment process, and immediately resume treatment after excluding the factors of unstable beam current intensity, so as to realize precise control of the proton treatment process.

[0049] 2, the application precisely designs the data acquisition period T of the sample database, so that the time before the failure occurs, PLC has enough time to issue "pause" instruction, the center area high voltage power supply has enough time to shut off the current, and the treatment head has enough time to respond to the instruction of PLC to stop treatment;Also make sure that after the failure is eliminated, PLC has enough time to issue "resume treatment" instruction, the center area high voltage power supply has enough time to start the current, and the treatment head has enough time to respond to the instruction of PLC to continue treatment. BRIEF DESCRIPTION OF DRAWINGS

[0050] Figure 1 The flow chart of the application for precisely controlling the proton treatment process.

[0051] Figure 2 The application of time series number change amount is shown in the diagram. DETAILED DESCRIPTION

[0052] The design principle of the present application

[0053] 1. The innovation of the present application: The innovation lies in the combination of time series prediction model and fine proton therapy. The conventional method only predicts the occurrence of failure with the model, but only predicting the occurrence of failure cannot be used for fine proton therapy. Because when the model predicts the occurrence of failure at the t+1 time point at the t time point, it is also the time when the treatment stops. Stopping treatment in the middle of treatment cannot complete the entire treatment process, which not only cannot achieve the purpose of fine treatment but also terminates the treatment. The present application not only predicts the occurrence of failure at the t+1 time point with the time series model, but also predicts the elimination of failure at the t+1 time point. When it is known that the failure at the t+1 time point will occur at the t time point, the PLC controller turns off the treatment head at the t+1 time point. When it is predicted that the failure at the t+1 time point is eliminated, the PLC controller completes the treatment of the treatment head at the t+1 time point. The prediction of the occurrence of failure at the t+1 time point and the prediction of the elimination of failure at the t+1 time point are accompanied by each other. First, the occurrence of failure is predicted, and then the elimination of failure is predicted. This time sequence cannot be reversed, otherwise, the prediction of the elimination of failure at the t+1 time point is inaccurate. In the standard definition of failure, the selection of the voltage fluctuation amplitude greater than or equal to U0 as the failure prediction standard depends on the need for treatment effect of the accelerator treatment end and the need for protection of the electrostatic deflection plate high-voltage feed ball.

[0054] 2. Several key points of the present application

[0055] One of the key points: the model not only predicts when the failure occurs, but also predicts when the failure is eliminated;

[0056] The second key point: the elimination of failure is "in advance", and the recovery of treatment is "immediate". The said "in advance" elimination of failure is to eliminate the failure at the t time point instead of the t+1 time point. Because the failure has already occurred at the t+1 time point, it is useless to eliminate it again. The said "immediate" recovery of treatment is to recover the treatment at the t+1 time point instead of the t time point, because the failure has not been completely eliminated at the t time point, and the treatment cannot be recovered when the failure has not been completely eliminated.

[0057] The third key point is: A, selection of model training input data: the present application takes the leakage current and the vacuum degree as the input data for model training. The principle is: when sparking occurs, leakage current will be generated, but the measurement of leakage current is the measurement of a certain point, which may have one-sidedness of local measurement, so the measurement of vacuum degree is added. When sparking occurs, the vacuum degree inside the entire vacuum chamber will change, but the change of vacuum degree has multiple causes, not only the change of vacuum degree caused by sparking. Therefore, leakage current and vacuum degree are measured together. B, selection of model training output data: the present application takes the change of input voltage as the output data for model training. The change of leakage current and vacuum degree is the cause of the change of voltage, and the change of voltage is the result of the change of leakage current and vacuum degree. Therefore, the change of input voltage is taken as the output data for model training.

[0058] The fourth key point is: the establishment of time series database. The data of time series database is not the originally collected data, but the data after a part of data is arranged, that is, the data after difference, for example, the leakage current i of time series database is directly collected data, and the cavity vacuum degree p and the input voltage value U are respectively the data after difference;

[0059] The fifth key point is: determination of data collection period T. The data collection period T cannot be too large or too small. It should be combined with the calculation time of the model and the response time of multiple hardware in the proton therapy system. Specifically: the prediction model calculation time, the PLC “pause” instruction time when the fault occurs, the high-voltage power supply voltage adjustment time of the central zone opening and closing current, the treatment head response instruction time, and the single sparking fault duration. Among these multiple times, it is difficult to determine which time as the data collection time, and it must be considered comprehensively. The data collection period T of the time series sample database of the present application is greater than or equal to the prediction model calculation time t 预测 + the PLC “pause” instruction time t PLC + the maximum value of the two times; the two times are: the high-voltage power supply voltage adjustment time t 电源 of the central zone opening and closing current and the treatment head response instruction time t 治疗头 , and the data sampling period T is less than or equal to the single sparking fault duration;

[0060] Based on the above principle, the present application relates to a method for precisely controlling the influence of sparking of the extraction system on the proton therapy process, as shown in Figure 1 、 Figure 2 , which is characterized by comprising the following steps:

[0061] Step 1: selecting a feeding voltage fluctuation amplitude greater than or equal to U0 as a later fault prediction standard;

[0062] Step 2: the electrostatic deflection plate enters a steady voltage working state;

[0063] Step three, data collection of leakage current, cavity vacuum degree and feed-in voltage value of electrostatic deflection plate in steady voltage working state, and establishment of time sequence sample database; data collection period T of time sequence sample database ≥ prediction model calculation time t 预测 + PLC sends "pause" instruction time t when fault occurs PLC + The maximum of two times max; the two times are: high-voltage power supply voltage regulation time t of center zone on-off current 电源 And treatment head response instruction time t 治疗头 , and data sampling period T ≤ single sparking fault duration time;

[0064] Step four, constructing a time sequence prediction model based on LSTM, inputting the sample database into the model, and obtaining the input-output model of the sparking fault prediction of the high-voltage extraction electrode of the cyclotron through model training, model evaluation and model optimization; the input is the input data collected at the current t time point, and the output is the voltage fluctuation amplitude predicted by the model at the t+1 time point; the actual time is data sampling period T*t, and t at the t time point is the time sequence number;

[0065] Step five, real-time collection of electrostatic deflection plate leakage current and vacuum degree data at the current t time point;

[0066] Step six, using the data collected at the current t time point as the input of the prediction model, and predicting the feed-in voltage fluctuation amplitude at the t+1 time point through the prediction model;

[0067] Step seven, whether the feed-in voltage fluctuation amplitude at the t+1 time point is greater than U0, if not, return to step six, if yes, continue to step eight;

[0068] Step eight, after detecting that the fault is about to occur, sending a fault signal, and sending an instruction through PLC to control the center zone vertical deflection plate to shut off the beam current and control the treatment head at the treatment end to pause the action while recording the beam current pause time T*△t,△t is the total time point sum experienced from pausing to recovering, and the initial value is 0;

[0069] Step nine,△t=t+1 time point, judge whether the feed-in voltage fluctuation amplitude at the t+1 time point is greater than U0, if yes, loop step nine, if no, continue to step ten;

[0070] Step ten, after detecting the fault elimination, a reset signal is sent out, the beam is opened by the PLC sending instructions to control the central area vertical deflection plate, the treatment head at the treatment end is restored to work, and the pause time T*△t is fed back to the treatment head program, the pause time T*△t is the time from the occurrence of the fault to the elimination of the fault; the beam intensity in the working program of the treatment head is changed from f(T*t) to f(T*t-T*△t); wherein, T*t is the current time;

[0071] Further, the fluctuation amplitude of the selected feed-in voltage in step one is as follows:

[0072] 1) When the fluctuation amplitude of the feed-in voltage is greater than U1, the beam load fluctuation of the treatment end beam reaches about 10%;

[0073] 2) When the leakage current is greater than a certain value, corresponding to the fluctuation amplitude of the feed-in voltage greater than U2, the leakage current arc will cause damage to the high-voltage feed-in ball head part of the electrostatic deflection plate;

[0074] 3) The fluctuation amplitude of the feed-in voltage U0=min(U1,U2).

[0075] Further, the process of establishing the time sequence sample database in step three is as follows:

[0076] 1) Use a high-speed data acquisition card to collect the leakage current i, the cavity vacuum degree p and the feed-in voltage value U in the stable voltage working state as the original data sequence;

[0077] 2) The data cavity vacuum degree p and the feed-in voltage value U are processed to obtain the time sequence database, which includes time sequence number t, leakage current i, cavity vacuum degree change amount △p and feed-in voltage fluctuation amplitude △U;

[0078] 3) The data is divided into training set and test set according to the time sequence with 1:1 of the first half time sequence after the second half time sequence, which is used for training and testing of the prediction model;

[0079] 4) The leakage current i and the cavity vacuum degree change amount △p of the training set are used as input variables, and the feed-in voltage fluctuation amplitude ΔU is used as an output variable to form a time sequence sample database;

[0080] 5) Normalize all features of the training set and the test set;

[0081] 6) The data of the training set is converted into a supervised learning problem, and by converting into a supervised learning problem, the output variable at the t+1 time point is predicted by the input variable at the t time point.

[0082] Supplementary note 1:

[0083] Each time point corresponds to an actual time, for example: if the t time point is 1, the time corresponding to the t time point is T*t, T = 150 ms, if the time point is t+1, the time corresponding to the t+1 time point is T*2 = 300 ms;

[0084] Further, the step four is to construct an LSTM-based time series prediction model, and the specific process is as follows:

[0085] 1) Build an LSTM model, which is divided into an input layer, a hidden layer and an output layer. There are 2 neurons in the input layer, 50 neurons in the initial setting of the hidden layer, and 1 neuron in the output layer.

[0086] 2) Input the training set data of the time series sample database into the LSTM model for training to obtain a training set data model based on the input data of the previous moment to predict the output data of the next moment.

[0087] 3) Test the training set data model training result using the test set data of the time series sample database, and evaluate the accuracy of the training set data model through the test.

[0088] Further, the accuracy of the training set data model in the process 3) of step four is evaluated, and the specific steps are as follows:

[0089] 1) Calculate the error loss with the prediction result and the test set data.

[0090] 2) When the error loss exceeds the threshold value, adjust and optimize the training set data model.

[0091] The adjustment and optimization of the training set data model is specifically: the model is optimized by adjusting the number of hidden layer neurons to reduce the prediction error loss, and an input-output model for predicting the spark fault of the high-voltage extraction electrode of the cyclotron is obtained. The input-output model for predicting the spark fault of the high-voltage extraction electrode of the cyclotron after completion of the training can realize the prediction of the output variable feed-in voltage fluctuation amplitude △U at the t+1 time point by inputting the input variable leakage current i and the cavity vacuum degree change amount △p collected at the current t time point.

[0092] Further, the step eight has the following specific process:

[0093] 1) After detecting that a fault is about to occur, increase the voltage difference of the central vertical deflection plate through the PLC control center to form an electric field to make the central beam current off, and at the same time, the PLC controls the proton accelerator treatment head end to pause;

[0094] 2) Record the pause time T*△t(△t is the initial value of 0) from the occurrence of the fault to the elimination of the fault.

[0095] Further, the step ten specific process as follows:

[0096] 1) After detecting the failure elimination, the control center area vertical deflection plate voltage difference drop to 0 by PLC delay a set time, so that the center area beam current stop shut down re start;

[0097] 2) At the same time, PLC control proton accelerator treatment head end work start, and will pause time T*△t feedback to the treatment head program, the beam intensity of the treatment head work program from f(T*t) to f(T*t-T*△t), namely in the static deflection plate steady voltage work state at the time of failure elimination treatment work to restore normal; said T*t for the current time; said pause time T*△t is from the failure to the failure elimination time.

[0098] Further, the delay a set time is the data acquisition period T of timing sample database-predictive model calculation time t 预测 -PLC issued "pause" instruction time t PLC - The maximum value of two time max.

[0099] Example one: the calculation of sampling period and delay time

[0100] 1) The calculation of sampling period: the first step: sum of predictive model calculation time + PLC control reaction time + center area open and close current high voltage power supply voltage regulation time: if the predictive model calculation time is 100 ms, PLC control reaction time is 10 ms, and the high voltage power supply voltage regulation time of center area open and close current is 20 ms, then the predictive model calculation time + PLC control reaction time + center area open and close current high voltage power supply voltage regulation time = 130 mm; The second step: determine the sampling period: because the single spark failure time is usually greater than 200 ms, so take the sampling period between 130 mm and 200 ms, this embodiment takes the sampling period = 150 mm≤ single spark failure duration 200 ms; Because the high voltage power supply voltage regulation time of center area open and close current t 电源 and the treatment head response instruction time t 治疗头 time comparison, the high voltage power supply voltage regulation time of center area open and close current t 电源 is the maximum value. Therefore, after comparison, the high voltage power supply voltage regulation time of center area open and close current 20 ms is adopted.

[0101] The reason for choosing the high voltage power supply voltage regulation time of center area open and close current 20 ms is that compared with the treatment head response instruction time t 治疗头 , the high voltage power supply voltage regulation time of center area open and close current 20 ms is larger, and the treatment head response instruction time t治疗头 The time is less than the high-voltage power supply voltage regulation time of the center area opening and closing current.

[0102] 2) Delay time calculation: if the sampling period is 150 ms, the prediction model calculation time is 100 ms, the PLC control reaction time is 10 ms, the high-voltage power supply voltage regulation time of the center area opening and closing current is 20 ms, and the single spark failure time is usually greater than 200 ms, then the delay is a set time = sampling period 150 ms - prediction model calculation time 100 ms - PLC control reaction time 10 ms - high-voltage power supply voltage regulation time of the center area opening and closing current 20 ms = 20 ms. 电源 20 ms, the single spark failure time is usually greater than 200 ms, then the delay is a set time = sampling period 150 ms - prediction model calculation time 100 ms - PLC control reaction time 10 ms - high-voltage power supply voltage regulation time of the center area opening and closing current 20 ms = 20 ms.

[0103] It should be emphasized that the above specific embodiments are only an explanation of the present application, and are not a limitation of the present application. Those skilled in the art can make modifications to the above embodiments without creative contribution after reading the present specification, but as long as they are within the scope of the claims of the present application, they are protected by the patent law.

Claims

1. A method of precisely controlling the impact of a beam-on system on a proton therapy procedure, the method comprising: determining a beam-on system parameter; and adjusting the beam-on system parameter based on the determined beam-on system parameter. The method comprises the following steps: Step one, selecting a feed-in voltage fluctuation amplitude greater than or equal to U0 as a later fault prediction standard; Step two, the electrostatic deflection plate enters a stable voltage working state; Step three, data collection of leakage current, cavity vacuum degree and feed-in voltage value of the electrostatic deflection plate in steady voltage working state, and establishment of time sequence sample database; data collection period T of the time sequence sample database ≥ prediction model calculation time t 预测 + PLC "pause" instruction time t when the fault occurs PLC + The maximum value max of the two times: the high-voltage power supply voltage regulation time t of the center area on-off current 电源 And the treatment head response instruction time t 治疗头 , and the data sampling period T ≤ the duration of a single sparking fault; Step four, constructing a time series prediction model based on LSTM, inputting a sample database into the model, and obtaining an input-output model for the spark fault prediction of the high-voltage extraction electrode of the cyclotron through model training, model evaluation, and model optimization; the input is the input data collected at the current t time point, and the output is the voltage fluctuation amplitude at the t+1 time point predicted by the model; the actual time is the data sampling period T*t, and t at the t time point is the time sequence number; Step five, real-time collection of the electrostatic deflection plate leakage current and vacuum degree data at the current t time point; Step six, using the data collected at the current t time point as the input of the prediction model to predict the feed-in voltage fluctuation amplitude at the t+1 time point through the prediction model; Step seven, whether the feed-in voltage fluctuation amplitude at the t+1 time point is greater than U0, if not, return to step six, and if yes, continue to step eight; Step eight, after detecting that a fault is about to occur, a fault signal is sent out, the PLC sends out an instruction to control the central area vertical deflection plate to shut off the beam current, control the treatment head at the treatment end to pause the action, and record the beam current pause time T*△t at the same time; △t is the total sum of time points experienced by the treatment from pausing to resuming, and the initial value is 0; Step nine, △t=t+1 time point, whether the feed-in voltage fluctuation amplitude at the t+1 time point is greater than U0, if yes, repeat step nine, and if no, continue to step ten; Step ten, after detecting that the fault is eliminated, a reset signal is sent out, the PLC sends out an instruction to control the central area vertical deflection plate to open the beam current, control the treatment head at the treatment end to resume work, and at the same time, feed back the pause time T*△t to the treatment head program; the pause time T*△t is the time from the occurrence of the fault to the elimination of the fault; the beam current intensity in the working program of the treatment head changes from f(T*t) to f(T*t-T*△t); wherein, T*t is the current time.

2. The method of claim 1, wherein the method is used to precisely control the effect of the system's sparking on the proton therapy process. The selected feed-in voltage fluctuation amplitude in step one has the following specific principles: 1) when the feed-in voltage fluctuation amplitude is greater than U1, the beam fluctuation of the treatment end beam reaches about 10%; 2) when the leakage current is greater than a certain value, and the corresponding feed-in voltage fluctuation amplitude is greater than U2, the leakage current arc will cause damage to the high-voltage feed-in ball head part of the electrostatic deflection plate; 3) the feed-in voltage fluctuation amplitude U0=min(U1,U2).

3. The method of claim 1, wherein the method is used to precisely control the effect of the system's sparking on the proton therapy process. The specific process of establishing the time sequence sample database in step three is as follows: 1) using a high-speed data acquisition card to collect the leakage current i, the cavity vacuum degree p, and the feed-in voltage value U in the stable voltage working state as the original data sequence; 2) performing difference processing on the data cavity vacuum degree p and the feed-in voltage value U to obtain a time sequence database, which includes the time sequence number t, the leakage current i, the cavity vacuum degree change △p, and the feed-in voltage fluctuation amplitude △U; 3) The data is divided into training set and test set according to time sequence, with the first half of the time sequence as the training set and the second half of the time sequence as the test set, for training and testing of the prediction model; 4) The leakage current i and the cavity vacuum degree change amount Δp of the training set are taken as input variables, and the input voltage fluctuation amplitude ΔU is taken as an output variable to form a time sequence sample database; 5) Normalization processing is performed on all features of the training set and the test set; 6) The data of the training set is converted into a supervised learning problem, and the output variable at the t+1 time point is predicted by the input variable at the t time point through the conversion into a supervised learning problem.

4. The method of claim 1, wherein the method is used to precisely control the effect of the system's sparking on the proton therapy process. The step four is to construct an LSTM-based time sequence prediction model, and the specific process is as follows: 1) Build an LSTM model, which is divided into an input layer, a hidden layer and an output layer; 2 neurons in the input layer, initially set 50 neurons in the hidden layer, and 1 neuron in the output layer; 2) The training set data of the time sequence sample database is input into the LSTM model for training to obtain a training set data model for predicting the output data at the next time point based on the input data at the previous time point; 3) The test set data of the time sequence sample database is used to test the training result of the training set data model, and the accuracy of the training set data model is evaluated through the test.

5. The method of claim 4, wherein the method is used to precisely control the effect of the system's sparking on the proton therapy process. The accuracy of the training set data model is evaluated in the process 3) of step four, and the specific steps are as follows: 1) Calculate the error loss with the prediction result and the test set data; 2) When the error loss exceeds the threshold, adjust and optimize the training set data model.

6. The method of claim 5, wherein the method is used to precisely control the effect of the system's sparking on the proton therapy process. The training set data model is adjusted and optimized, specifically: the model is optimized by adjusting the number of neurons in the hidden layer to reduce the prediction error loss, and an input-output model for predicting the spark fault of the high-voltage extraction electrode of the cyclotron is obtained, which can realize the prediction of the output variable input voltage fluctuation amplitude ΔU at the t+1 time point by inputting the input variables leakage current i and cavity vacuum degree change amount Δp collected at the current t time point.

7. The method of claim 1, wherein the method is used to precisely control the effect of the system's sparking on the proton therapy process. The specific process of step eight is as follows: 1) After detecting that the fault is about to occur, increase the voltage difference of the central vertical deflection plate through the PLC control center to form an electric field to make the central beam current turn off, and at the same time, the PLC controls the proton accelerator treatment head end to pause; 2) Record the pause time T*△t from the occurrence of the fault to the elimination of the fault, and the initial value of △t is 0.

8. The method of claim 1, wherein the method is used to precisely control the effect of the system's sparking on the proton therapy process. The specific process of step ten is as follows: 1) After detecting that the fault is eliminated, the voltage difference of the central vertical deflection plate is reduced to 0 through the PLC after a set time to make the central beam current stop turning off and restart; 2) At the same time, PLC controls the work of the proton accelerator treatment head end to start, and feeds back the suspension time T*△t to the treatment head program, and the beam current in the work program of the treatment head is changed from f(T*t) to f(T*t-T*△t), that is, the treatment work is restored to normal at the moment of failure elimination in the steady state of the electrostatic deflection plate; the T*t is the current time; the suspension time T*△t is the time from the failure to the failure elimination.

9. The method of claim 8, wherein the method is used to precisely control the effect of the system's sparking on the proton therapy process. The delay of a set time is the data collection period T of the timing sample database - prediction model calculation time t 预测 - The time t when the PLC issues a "pause" command when a fault occurs PLC - The maximum of the two times max.

Images