CN110909463B - Active control and protection method and system for high-power millimeter wave gyrotron traveling wave tube - Google Patents

Abstract

The invention belongs to the technical field of high-power millimeter waves, and specifically provides a method and system for active control and protection of high-power millimeter-wave gyro traveling wave tubes based on neural network prediction, so as to overcome the shortcomings of passive control and protection of existing automated testing systems. The present invention establishes and trains a neural network prediction model by adding spark tags to historical sample data, and predicts spark tags for the test data collected in real time through the prediction model. During the time period T, the spark tag prediction value is continuous Th ₁ When the number of times is 1 or the number of times the predicted value of the ignition tag is 1 is greater than Th ₂ , the automatic control basic control and protection module adopts emergency shutdown processing to realize the active control and protection of the automatic test of the high-power millimeter-wave gyroscopic traveling wave tube, which can effectively The number of ignitions of the device is reduced, thereby improving the safety of the ignition protection of the device, reducing the occurrence rate of the event of damage to the gyroscopic traveling wave tube due to ignition, and having a good economic effect.

Description

A kind of active control and protection method and system of high-power millimeter wave gyroscopic traveling wave tube

technical field

The invention belongs to the technical field of high-power millimeter waves, and relates to an automatic testing system for high-power millimeter-wave gyroscopic traveling wave tubes; in particular, it relates to ignition protection technology and neural network prediction technology of high-power millimeter-wave gyroscopic traveling wave tubes, and provides a neural network-based prediction technology The active control and protection method and system of the high-power millimeter-wave gyroscopic traveling wave tube are used to predict whether there will be a spark under the current operating state of the device, and through the control of the control and protection module, the active control of the device spark is realized. Save.

Background technique

High-power millimeter-wave gyroscopic traveling wave tube has a wide range of applications in the fields of national defense, scientific research, and civil communication. It has the advantages of high power, wide frequency band, and high gain. Usually when the device is tested or working normally, when the voltage applied to the electric vacuum device exceeds a certain value, a spark or high-voltage breakdown will occur. The so-called spark is a spark of a certain color between the electrodes. Similar to the discharge, the sound of the discharge can be heard at the same time as the fire is fired. This sound is formed by the instantaneous deflation of the fire. Most of the gas emitted by the ignition is usually quickly absorbed by the electrodes and other gettering materials in the tube, and the remaining part forms a plasma discharge with a high current density between the electrodes. When the current density reaches a certain level, the electrodes will be short-circuited. , causing serious damage or even scrapping the device, resulting in economic losses; at the same time, repeated ignition will reduce the vacuum in the device and affect the performance of the device; therefore, it should be avoided that the device is in the state of ignition for a long time. The alarm should be timely and the power should be turned off. Electric vacuum devices have a long development cycle and high cost, so the control and protection system is particularly important.

The control and protection system in the existing automated test system is passive control and protection, which is based on real-time collection of device parameters to determine whether the parameters exceed the threshold set in advance. When the threshold is exceeded, it is considered that the device has an abnormality such as sparking and power off. Because the reasonable setting of the threshold requires a certain amount of experience, there will be a certain delay in the acquisition of device parameters, and this control and protection processing method cannot take measures before the device sparks. Take measures.

Based on this, the present invention provides a method and system for active control and protection of a high-power millimeter-wave gyroscopic traveling wave tube based on neural network prediction.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide an active control and protection method and system with learning ability in view of the shortcomings of the passive control and protection of the existing automated test systems in the above-mentioned background technology, so as to realize the active control and protection of high-power millimeter-wave gyroscopic traveling wave tubes. , reduce the damage to the device due to ignition.

To achieve the above object, the technical scheme adopted in the present invention is:

An active control and protection method for a high-power millimeter-wave gyroscopic traveling wave tube, comprising the following steps:

Step S1. Establish a neural network prediction model;

Step S11. Collect and establish a neural network prediction model training sample set;

The training sample data includes: magnetic field current (A), compensated magnetic field current (A), cathode voltage (KV), cathode current (A), titanium pump current (nA), ignition label: ignition status is 1, no The ignition state is 0; the ignition label is used to mark whether the current state is ignition;

Set the sampling frequency, sample the historical data, and obtain the training sample data; during the sampling process, the ignition labels corresponding to the first Q (3≤Q≤5) moments of the ignition time are marked as ignition state 1;

Step S12. According to the training sample set, a neural network prediction model is established by the BP neural network method;

The BP neural network is adopted, in which the number of input layer nodes N=5, the number of output layer nodes L=1, and the initial value of the number M of hidden layer nodes is 3; the loss function Loss is set as the mean square error, and the number of training is nb_epoch, according to the training samples The data is used to train the BP neural network, and the neural network prediction model is obtained;

Step S2. Predict whether the current state of the device will spark through the neural network prediction model, whether it needs to be warned and take corresponding treatment;

Step S21. Collect the device working state data in real time and store it in the database, and collect the collected magnetic field current (A), compensated magnetic field current (A), cathode voltage (KV), cathode current (A), and titanium pump current (nA) The data is input into the neural network prediction model, and the predicted value of the ignition label corresponding to the real-time state is obtained;

Step S22. During the time period T, when the predicted value of the ignition tag is 1 for consecutive Th ₁ times or the number of times the predicted value of the ignition tag is 1 is greater than Th ₂ , a warning is issued and an emergency shutdown process is taken.

It should be noted that, in the above step S22, the threshold Th ₁ , the threshold Th ₂ , and the preset time period T can all be set according to the actual application environment.

Further, the active control and protection method further includes:

Step S3. Use the magnetic field current (A), compensation magnetic field current (A), cathode voltage (KV), cathode current (A), titanium pump current (nA) obtained in step S2 as sample data, and according to the real ignition Add the spark tag to the state, and at the same time mark the spark tags corresponding to the first Q moments of the spark time as spark state 1, obtain a new training sample, and add the new training sample to the training sample set for use. The next time you train the neural network prediction model, revise the neural network prediction model.

An active control and protection system for a high-power millimeter-wave gyroscopic traveling wave tube, comprising: a data acquisition module, a data storage module, a data preprocessing module, a neural network prediction module, a remote control module and a basic control and protection module; wherein:

The data acquisition module is used to collect sample data required for establishing a neural network prediction model, including: magnetic field current (A), compensation magnetic field current (A), cathode voltage (KV), cathode current (A), titanium pump current ( nA) data and current waveform and spectrum data acquisition; it should be noted that the waveform and spectrum data are used to determine whether the device is in a firing state;

The data preprocessing module is used for preprocessing the original data collected by the data acquisition module, including: data format conversion, data merging, data deduplication, and abnormal value removal;

The data storage module, in the form of a database, is used to store the original data collected by the data acquisition module and the data preprocessed by the data preprocessing module;

The neural network prediction module is connected to the data preprocessing module, and after preprocessing, the data is input into the neural network prediction model, and the neural network prediction model outputs the ignition label prediction value;

The remote control module is located at the client, and within the time period T, when the predicted value of the ignition tag is 1 for consecutive Th ₁ times or the number of times the predicted value of the ignition tag is 1 is greater than Th ₂ , a warning is issued and an instruction is issued to the basic controller. security module;

The basic control and protection module is used to realize passive control and protection and active control and protection operations; wherein, the active control and protection is: after receiving an instruction from the remote control module, emergency shutdown processing is performed on the data acquisition module; the passive control Guarantee: When the data value collected by the data acquisition module exceeds the set threshold, the acquisition module will be emergency shut down.

The working principle of the present invention is:

In the present invention, select specific historical data, and add a spark tag indicating whether the device in the corresponding state has sparked (1 if spark occurs, 0 if no spark), and the first Q pieces of data in the actual spark state are all Marked as ignition state 1, the labeled data is used as sample data, and the neural network prediction model is established by the BP neural network method, that is, the neural network extracts the characteristics of the state parameters before ignition, and can predict the device according to the real-time input state. Whether there will be sparks at work. When the neural network prediction model predicts that the current state of the ignition label is 1, it may not actually start, but if the predicted ignition labels are all 1 continuously or have more than a certain number of 1s within the preset time period, There is reason to believe that there is a high probability that the device will spark in the next, and a warning will be issued and an emergency shutdown process will be taken; thus, early warning and active control and protection before the spark occurs are realized.

To sum up, the beneficial effects of the present invention are:

The present invention establishes and trains a neural network prediction model by adding spark tags to historical sample data, and predicts spark tags for the test data collected in real time through the prediction model. During the time period T, the spark tag prediction value is continuous Th ₁ When the number of times is 1 or the number of times the predicted value of the ignition tag is 1 is greater than Th ₂ , the automatic control basic control and protection module adopts emergency shutdown processing to realize the active control and protection of the automatic test of the high-power millimeter-wave gyroscopic traveling wave tube, which can effectively The number of ignitions of the device is reduced, thereby improving the safety of the ignition protection of the device, reducing the occurrence rate of the event of damage to the gyroscopic traveling wave tube due to ignition, and having a good economic effect.

Description of drawings

FIG. 1 is a block diagram of the active control and protection system of the high-power millimeter-wave gyroscopic traveling wave tube of the present invention.

FIG. 2 is a schematic block diagram of a neural network prediction model in an embodiment of the present invention.

FIG. 3 is a structural diagram of a BP neural network in an embodiment of the present invention.

Detailed ways

In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings.

Example 1

This embodiment provides an active control and protection method for a high-power millimeter-wave gyroscopic traveling wave tube based on neural network prediction, including the following steps:

Step S1. Establish a neural network prediction model; specifically:

Step S11. Collect the sample data set required for establishing the neural network prediction model;

Required sample data include: magnetic field current (A), compensated magnetic field current (A), cathode voltage (KV), cathode current (A), titanium pump current (nA), ignition label: ignition is 1, no ignition is 0; the ignition label is used to mark whether the current state is ignited;

Further, considering the delay in data collection during actual ignition and the need to minimize the number of device ignitions, the first 5 states at the actual ignition time (the actual sample sampling rate is 0.5 seconds/time, that is, the first 2.5 data in seconds) are marked as ignition state 1; in this way, when the current state ignition label is predicted to be 1, it may not be really ignition, but if within a certain period of time, the predicted ignition label When all are 1 continuously or there are more than a certain number of 1s, there is reason to believe that the device will spark with a high probability in the next, so it is necessary to issue a warning and judge whether emergency shutdown processing is required according to the prediction result;

Step S12, according to the sample data, establish a neural network prediction model through the BP neural network method;

Specifically, in the present invention, a BP neural network (Back Propagation Networks—back propagation network) is used. The BP neural network is composed of nonlinear transfer function neurons, which can learn and store a large number of input-output mode mapping relationships. The learning rule is to use the gradient descent method to continuously adjust the connection weights of the network through backpropagation, so that the loss function of the neural network is as small as possible;

The number of input layer nodes N: depends on the number of attributes of the sample; input variables in the present invention: magnetic field current (A), compensation magnetic field current (A), cathode voltage (KV), cathode current (A), titanium pump current ( nA), so the input layer is determined to be 5 nodes;

The number of output layer nodes L: depends on the predicted number of nodes; in the present invention, the output variable: the ignition label (1 for ignition, 0 for unignited), so the output layer is determined to be 1 node;

The number of hidden layer nodes M: Generally, there are three empirical formulas as follows:

In this embodiment, the number of nodes in the input layer is N=5, and the number of nodes in the output layer is L=1, so the number of nodes in the hidden layer M=3 is taken as the initial value; the number of neural network layers is simply 3 layers. The structure is shown in Figure 3;

Here, the number of hidden layer nodes of the neural network can be dynamically adjusted according to the prediction results, and the prediction results can be dynamically adjusted according to the situation according to the number of different hidden layer nodes;

Then select the loss function Loss of the model as loss='mean_squared_error', that is, the mean squared error:

为网络预测值，n为样本个数；Among them, Y _i is the value of the spark label in the sample,

is the predicted value of the network, and n is the number of samples;

Select the optimization method as optimizer='adam', that is, adaptive moment estimation, and dynamically adjust the learning rate of each parameter by using the first-order moment estimation and second-order moment estimation of the gradient;

Then set the number of training nb_epoch;

Input the sample data collected in step S11 into the constructed neural network structure, and perform training to obtain a trained neural network model;

Step S2. Predict whether the current state of the device will spark through the neural network prediction model, whether it needs to be warned and control the basic control and protection module to take corresponding processing, as shown in Figure 2; the details are:

Step S21. Collect the working state of the device in real time, put the collected data on the server, and collect the collected magnetic field current (A), compensated magnetic field current (A), cathode voltage (KV), cathode current (A), titanium pump The current (nA) data is input into the trained neural network prediction model to obtain the prediction value of whether the device corresponding to the real-time state is on fire label;

In step S22, since the ignition label value in the sample data is to mark all the first 5 pieces of data in the actual ignition state as ignition state 1, when a single 1 is predicted to appear, it must be really ignition, but if it is in a certain state. During the time period T, when the predicted ignition tag value appears to be 1 continuously or exceeds a certain number of 1s, a warning is issued and an emergency shutdown process is determined according to the number of predicted ignition tags 1 .

Further, the active control and protection method further includes:

Step S3, take the magnetic field current (A), compensation magnetic field current (A), cathode voltage (KV), cathode current (A), titanium pump current (nA) obtained in step S2 as sample data, and combine the actual ignition Whether to add spark tag data to revise the neural network prediction model;

Specifically, in each test, the process data of each state parameter of the device operation is collected by the data acquisition module, stored in the original database in the database, and the label data of whether to fire is added to the original data, and the program control is written. Every time a spark actually occurs, the labels corresponding to the first 5 states corresponding to the actual spark time are marked as spark state 1 (the actual sample sampling rate is 0.5 seconds/time, that is, the data within the first 2.5 seconds), as a sample for the next prediction model training. With the operation of the automated test system, the model training samples continue to increase, and the prediction model is constantly revised. It has the ability to continuously improve the prediction accuracy, and has the ability to self-learn. According to the mathematical model, it can predict whether the device will be in the next time under the current state. , the ignition occurs, so that before the actual ignition occurs, early warning information can be issued, and when necessary (the number of 1 predicted results in a certain period of time exceeds a certain threshold), the basic control and protection module is automatically controlled to take emergency shutdown processing to achieve The active control and protection of automatic testing of high-power millimeter-wave gyroscopic traveling wave tubes can effectively reduce the number of ignitions of the device, thereby improving the safety of device ignition protection and reducing the occurrence of events that damage the gyroscopic traveling wave tube due to ignition. good economic effect.

Example 2

On the basis of Embodiment 1, the present invention provides an active control and protection system for a high-power millimeter-wave gyroscopic traveling wave tube, as shown in FIG. 1 , including:

Data acquisition module, data storage module, data preprocessing module, neural network prediction module, remote control module and basic control and protection module; of which:

The data acquisition module collects sample data required for establishing a neural network prediction model, including magnetic field current (A), compensation magnetic field current (A), cathode voltage (KV), cathode current (A), and titanium pump current (nA) data and current waveform acquisition, spectrum data acquisition, etc. Among them, whether the device is on fire is not only related to the first five attribute values, but also related to the current waveform and spectrum data, so in a more comprehensive analysis, it is also necessary to collect waveforms and spectrums data;

The basic control and protection module is connected with the data acquisition module, and mainly realizes two functions: when it is found through the data acquisition module that the device has actually sparked abnormally, that is, when the collected data value exceeds the set threshold, the acquisition module Perform emergency shutdown processing, which belongs to passive control and protection; when the data passes through the neural network prediction module and predicts that the device will have a spark abnormality within a certain period of time in the current state, the remote module sends an instruction to control the basic control and protection module. Perform emergency shutdown processing on the acquisition module, which belongs to active control and protection;

The data storage module, in the form of a database, is used to store the collected test data and the data processed by the data preprocessing module;

Specifically, in each test, the process data of each state parameter of the device operation is collected by the data acquisition module and stored in the original database in the storage module database. The original data is added with the label data of whether to fire, and the program control is written. , each time a spark actually occurs, the labels corresponding to the first 5 states corresponding to the actual spark time are marked as spark state 1 (the actual sample sampling rate is 0.5 seconds/time, that is, the data within the first 2.5 seconds). ), as the sample for the next prediction model training;

The data preprocessing module completes the preprocessing of the collected raw data: including data format conversion, data merging, data deduplication, removal of abnormal values, etc., and re-stores the processed data in the database;

The neural network prediction module connected with the data preprocessing module is suitable for establishing a neural network prediction model according to the sample data combined with the label value data of whether the device is on fire, and the data uploaded by the data acquisition module is processed as the test data, as the neural network. Predict the input of the model to get a prediction about whether the system will fire in the current state of operation;

The remote control module is located at the client, and when the neural network prediction module predicts that the current state has a high probability of sparking, a corresponding warning is issued, and an emergency shutdown process is taken by controlling the basic control and protection module;

Specifically, the prediction model is used to predict the test data collected in real time. When the prediction result is 1 (fire), the device has not been fired yet, and the predicted fire labels are continuous within the preset time period. When it is 1 or there are more than a certain number of 1s, an instruction is issued to control the basic control and protection module to take emergency shutdown processing.

The above descriptions are only specific embodiments of the present invention, and any feature disclosed in this specification, unless otherwise stated, can be replaced by other equivalent or alternative features with similar purposes; all the disclosed features, or All steps in a method or process, except mutually exclusive features and/or steps, may be combined in any way.

Claims (3)

1. An active control and protection method for a high-power millimeter wave gyrotron traveling wave tube comprises the following steps:

step S1, establishing a neural network prediction model;

step S11, collecting and establishing a neural network prediction model training sample set;

the training sample data comprises: magnetic field current (a), compensation magnetic field current (a), cathode voltage (KV), cathode current (a), titanium pump current (nA), ignition tag: the ignition state is 1, and the non-ignition state is 0; the ignition tag is used for marking whether the current state is ignited or not;

setting sampling frequency, and sampling historical data to obtain training sample data; in the sampling process, before the ignition moment occursQThe ignition labels corresponding to all the moments are marked as the ignition state 1, wherein the number of the ignition labels is not less than 3Q≤5；

Step S12, establishing a neural network prediction model by a BP neural network method according to the training sample set;

using BP neural network, in which the number of nodes of the input layerNNumber of output layer nodes =5L=1, number of hidden layer nodesMThe initial value is 3; setting a loss functionLossTraining the BP neural network according to training sample data to obtain a neural network prediction model, wherein the training times nb _ epoch are mean square errors;

step S2, predicting whether the current state of the device will strike fire or not through the neural network prediction model, and whether warning is needed or not and corresponding processing is adopted;

step S21, acquiring working state data of the device in real time, storing the working state data in a database, inputting the acquired data of the magnetic field current (A), the compensation magnetic field current (A), the cathode voltage (KV), the cathode current (A) and the titanium pump current (nA) into a neural network prediction model, and acquiring a predicted value of the ignition label corresponding to the real-time state;

step S22 time periodTWhen the predicted value of the ignition label is continuousTh ₁The number of times of the order of 1 or the predicted value of the ignition tag of 1 is greater thanTh ₂When so, a warning is issued and emergency shutdown processing is taken.

2. The active control and protection method for the high-power millimeter wave gyrotron traveling wave tube according to claim 1, wherein the active control and protection method further comprises:

step S3, the magnetic field current (A), the compensation magnetic field current (A), the cathode voltage (KV) and the cathode obtained in the step S2Taking data of current (A) and titanium pump current (nA) as sample data, adding a sparking tag according to the real sparking state, and simultaneously adding the data before the sparking momentQAnd marking the ignition labels corresponding to each moment as the ignition state 1 to obtain new training samples, and adding the new training samples into the training sample set for training the neural network prediction model next time to correct the neural network prediction model.

3. An active control and protection system of a high-power millimeter wave gyrotron traveling wave tube comprises: the system comprises a data acquisition module, a data storage module, a data preprocessing module, a neural network prediction module, a remote control module and a basic control and protection module; wherein:

the data acquisition module is used for acquiring sample data required by establishing the neural network prediction model, and comprises: acquiring data of magnetic field current (A), compensation magnetic field current (A), cathode voltage (KV), cathode current (A), titanium pump current (nA) and current waveform and frequency spectrum data; it should be noted that the waveform and spectrum data are used to determine whether the device is in a sparking state;

the data preprocessing module is used for preprocessing the original data acquired by the data acquisition module, and comprises: converting data format, merging data, removing data duplicate and removing abnormal value;

the data storage module is in a database form and is used for storing the original data acquired by the data acquisition module and the data preprocessed by the data preprocessing module;

the neural network prediction module is connected with the data preprocessing module, preprocessed data are input into the neural network prediction model, and the neural network prediction model outputs a predicted value of the ignition tag;

the remote control module is located at the client side and used for time periodTWhen the predicted value of the ignition label is continuousTh ₁The number of times of the order of 1 or the predicted value of the ignition tag of 1 is greater thanTh ₂When the system is used, a warning is sent out, and an instruction is sent out to the basic control and protection module;

the basic control and protection module is used for realizing passive control and protection and active control and protection operation; wherein, the active control is: after receiving a remote control module instruction, carrying out emergency shutdown processing on the data acquisition module; the passive control is as follows: and when the data value acquired by the data acquisition module exceeds a set threshold value, carrying out emergency shutdown processing on the acquisition module.

Images