阅读说明：本技术 基于时序重构的雷达故障诊断方法 (Radar fault diagnosis method based on time sequence reconstruction ) 是由 刘井泉 解光耀 曾聿赟 张昊宇 刘正藩 秦楚晴 杨辉 于 2019-10-16 设计创作，主要内容包括：本发明公开了一种基于时序重构的雷达故障诊断方法,包括以下步骤：利用气象雷达系统的发射机分系统的历史状态数据和报警数据,以报警数据做标签,使用逐步回归法提取与发射机分系统故障相关的特征参数共n个,若特征参数发生显著变化,则发射机的健康状态发生变化；取特征参数中的相关系数最大的特征参数y作为重构模型的重构参数目标,使用长短时记忆网络LSTM模型搭建重构模型,利用除y以外的特征参数共n-1个对y进行拟合重构,得到重构值<Image he="65" wi="50" file="DDA0002235356960000011.GIF" imgContent="drawing" imgFormat="GIF" orientation="portrait" inline="no"></Image>对重构值<Image he="51" wi="31" file="DDA0002235356960000012.GIF" imgContent="drawing" imgFormat="GIF" orientation="portrait" inline="no"></Image>和实测值y的差值做基于概率的量化标准；对量化结果做时间区间统计,滤除误报警,得到最终诊断结果。(The invention discloses a radar fault diagnosis method based on time sequence reconstruction, which comprises the following steps: the method comprises the steps that historical state data and alarm data of a transmitter subsystem of a meteorological radar system are utilized, the alarm data are used as labels, a stepwise regression method is used for extracting n characteristic parameters related to faults of the transmitter subsystem, and if the characteristic parameters are remarkably changed, the health state of the transmitter is changed; taking the characteristic parameter y with the maximum correlation coefficient in the characteristic parameters as a reconstruction parameter target of a reconstruction model, building the reconstruction model by using a long-term memory network (LSTM) model, and performing fitting reconstruction on y by using n-1 characteristic parameters except y to obtain a reconstruction value For the reconstructed value Making a probability-based quantization standard for the difference value of the measured value y and the measured value y; the time interval statistics is carried out on the quantization result,and filtering false alarms to obtain a final diagnosis result.)

1. A radar fault diagnosis method based on time sequence reconstruction is characterized by comprising the following steps:

the method comprises the steps that historical state data and alarm data of a transmitter subsystem of a meteorological radar system are utilized, the alarm data are used as labels, a stepwise regression method is used for extracting n characteristic parameters related to faults of the transmitter subsystem, and if the characteristic parameters are remarkably changed, the health state of the transmitter is changed;

taking the characteristic parameter y with the maximum correlation coefficient in the characteristic parameters as a reconstruction parameter target of a reconstruction model, building the reconstruction model by using a long-term memory network (LSTM) model, and performing fitting reconstruction on y by using n-1 characteristic parameters except y to obtain a reconstruction value

For the reconstructed value

and carrying out time interval statistics on the quantized result, and filtering false alarms to obtain a final diagnosis result.

2. The time-series reconstruction-based radar fault diagnosis method according to claim 1, wherein the pair of reconstructed values

analyzing the distribution condition of the measured value y, and if a plurality of running states exist, superposing the plurality of Gaussian distributions;

assuming that the predicted value and the measured value are independent, if the predicted value obeys Gaussian distribution, assuming that the radar operating state is normal, the measured value also obeys the Gaussian distribution, and the difference value of the two obeys N (0,2 sigma ^2) distribution;

and carrying out normalization processing on the probability of the difference value to obtain a final quantization result:

3. The radar fault diagnosis method based on time sequence reconstruction as claimed in claim 1 or 2, wherein the time interval statistics is performed on the quantized result, the false alarm is filtered out, and the obtaining of the final diagnosis result further comprises:

setting a probability decision threshold η₀，η＜η₀Considering a fault state point of the transmitter;

and (5) performing time interval statistics by using m continuous time points, and determining that a fault occurs when the fault state point is greater than 0.3 m.

4. The radar fault diagnosis method based on time series reconstruction as claimed in claim 1 or 2, wherein the stepwise regression method is a forward-lead method, in particular: firstly, only adding one independent variable with the largest variation of the dependent variable into the model, then, trying to add another independent variable, checking whether the variation of the dependent variable which can be explained by the whole model is obviously increased, and repeatedly iterating until no independent variable exists and then conforming to the condition of adding the model.

5. The radar fault diagnosis method based on time series reconstruction as claimed in claim 1 or 2, wherein the stepwise regression method is a backward elimination method, and specifically comprises: putting all variables into a model, then trying to remove one independent variable from the model, checking whether the variation of the explanation dependent variable of the whole model has obvious variation or not, and removing the variable which reduces the explanation quantity to the minimum; and repeating iteration until no independent variable meets the condition of elimination.

6. The radar fault diagnosis method based on time series reconstruction as claimed in claim 1 or 2, wherein the stepwise regression method is a two-way elimination method, and specifically comprises: firstly, adding an independent variable which explains the maximum variation of the dependent variable into the model, then trying to add another independent variable, checking all the variables in the whole model, if the dependent variable is remarkably increased, retaining the independent variable, eliminating the variable with insignificant effect, and repeating iteration until an optimal variable combination is finally obtained.

7. The radar fault diagnosis method based on time sequence reconstruction as claimed in claim 1 or 2, wherein the process of building the reconstruction model by using the long-time memory network LSTM model is as follows:

forgetting to remember door gamma_fReading the non-Linear activation of the last cell a^<t-1>And input data x of the current LSTM cell^<t>Outputting a value between 0 and 1 to each of the states c of the cells in the LSTM^<t-1>Wherein 1 represents "completely retained", and 0 represents "completely discarded";

input gate gamma_uFor the sigmoid layer, which is used to determine the value to be updated, a new candidate vector is created using the tanh layer

output gate gamma_oDetermining the output value, running sigmoid layer gamma to determine which part of LSTM cell state will be output, and outputting cell state c^<t>Treatment with tanh gives a composition of between-1 and 1Value and sum it with the output gate Γ_oThe outputs of (a) are multiplied to finally output a part of the determined output.

Technical Field

The invention belongs to the technical field of radar systems, and particularly relates to a radar fault diagnosis method based on time sequence reconstruction.

Background

The natural weather radar system in use at present is a relatively complex electronic system, and all electronic parameters are not physically connected, so that the electronic parameters reflected by faults are not physically connected, and the fault diagnosis and prediction of the natural weather radar system cannot be carried out by utilizing the traditional expert experience.

Disclosure of Invention

In view of the above technical problems, the present invention is directed to providing a radar fault diagnosis method based on time sequence reconstruction.

In order to solve the technical problems, the invention adopts the following technical scheme:

a radar fault diagnosis method based on time sequence reconstruction comprises the following steps:

the method comprises the steps that historical state data and alarm data of a transmitter subsystem of a meteorological radar system are utilized, the alarm data are used as labels, a stepwise regression method is used for extracting n characteristic parameters related to faults of the transmitter subsystem, and if the characteristic parameters are remarkably changed, the health state of the transmitter is changed;

taking the characteristic parameter y with the maximum correlation coefficient in the characteristic parameters as a reconstruction parameter target of a reconstruction model, building the reconstruction model by using a long-term memory network (LSTM) model, and performing fitting reconstruction on y by using n-1 characteristic parameters except y to obtain a reconstruction value

For the reconstructed valueMaking a probability-based quantization standard for the difference value of the measured value y and the measured value y;

and carrying out time interval statistics on the quantized result, and filtering false alarms to obtain a final diagnosis result.

Preferably, the pair of reconstructed values

Making a probability-based quantification of the difference from the measured value y further comprises:

analyzing the distribution condition of the measured value y, and if a plurality of running states exist, superposing the plurality of Gaussian distributions;

assuming that the predicted value and the measured value are independent, if the predicted value obeys Gaussian distribution, assuming that the radar operating state is normal, the measured value also obeys the Gaussian distribution, and the difference value of the two obeys N (0,2 sigma ^2) distribution;

and carrying out normalization processing on the probability of the difference value to obtain a final quantization result:

if the predicted value is equal to the measured value, η is equal to 1, and the more the predicted value deviates from the measured value, the closer η is to 0.

Preferably, the time interval statistics of the quantization result and the filtering of false alarm to obtain the final diagnosis result further comprise:

setting a probability decision threshold η₀，η＜η₀Considering a fault state point of the transmitter;

and (5) performing time interval statistics by using m continuous time points, and determining that a fault occurs when the fault state point is greater than 0.3 m.

Preferably, the stepwise regression method is a forward introduction method, specifically: firstly, only adding one independent variable with the largest variation of the dependent variable into the model, then, trying to add another independent variable, checking whether the variation of the dependent variable which can be explained by the whole model is obviously increased, and repeatedly iterating until no independent variable exists and then conforming to the condition of adding the model.

Preferably, the stepwise regression method is a backward elimination method, and specifically includes: putting all variables into a model, then trying to remove one independent variable from the model, checking whether the variation of the explanation dependent variable of the whole model has obvious variation or not, and removing the variable which reduces the explanation quantity to the minimum; and repeating iteration until no independent variable meets the condition of elimination.

Preferably, the stepwise regression method is a bidirectional elimination method, and specifically includes: firstly, adding an independent variable which explains the maximum variation of the dependent variable into the model, then trying to add another independent variable, checking all the variables in the whole model, if the dependent variable is remarkably increased, retaining the independent variable, eliminating the variable with insignificant effect, and repeating iteration until an optimal variable combination is finally obtained.

Preferably, the process of building the reconstruction model by using the long-term memory network LSTM model is as follows:

forgetting to remember door gamma_fReading the non-Linear activation of the last cell a^<t-1>And input data x of the current LSTM cell^<t>Outputting a value between 0 and 1 to each of the states c of the cells in the LSTM^<t-1>Wherein 1 represents "completely retained", and 0 represents "completely discarded";

input gate gamma_uFor the sigmoid layer, which is used to determine the value to be updated, a new candidate vector is created using the tanh layer

Updated vector C^<t>The input door and the forgetting door are jointly determined;

output gate gamma_oDetermining the output value, running sigmoid layer gamma to determine which part of LSTM cell state will be output, and outputting cell state C^<t>Processing by tanh to obtain a value between-1 and combining it with the output gate Γ_oThe outputs of (a) are multiplied to finally output a part of the determined output.

The invention has the following beneficial effects: the method has the advantages that fault diagnosis is carried out aiming at faults of the transmitter subsystem of the natural weather radar system, possible faults can be predicted in advance, pre-alarming is carried out, and false alarms can be filtered under the condition that the existing radar frequently and ineffectively alarms.

Drawings

FIG. 1 is a flowchart illustrating steps of a radar fault diagnosis method based on time sequence reconstruction according to an embodiment of the present invention;

fig. 2 is a schematic structural diagram illustrating a radar fault diagnosis method based on time sequence reconstruction according to an embodiment of the present invention, in which an operation state is a superposition of a plurality of gaussian distributions;

FIG. 3 is a schematic flow chart of an LSTM algorithm of a radar fault diagnosis method based on time sequence reconstruction according to an embodiment of the present invention;

fig. 4 is a schematic structural diagram of an LSTM of the radar fault diagnosis method based on time sequence reconstruction according to the embodiment of the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

Referring to fig. 1, a flowchart illustrating steps of a radar fault diagnosis method based on time sequence reconstruction according to an embodiment of the present invention is shown, which includes the following steps:

s10, using the historical state data and alarm data of the radar transmitter subsystem, using the alarm data as labels, using stepwise regression method to extract n characteristic parameters related to the fault of the transmitter subsystem, considering that if the characteristic parameters are changed significantly, the health state of the transmitter is changed, because the characteristic parameters are the key information extracted from the state variables of the system, the key information can accurately represent the state of the system, if the operation state of the system is changed, the key information is directly reflected on the change of the state variables, and the characteristic parameters show the change more significantly.

The historical status data may include radar status data for sites throughout the country of the historical year, with data points collected at intervals, each data point containing a number of variable parameters. The alarm data comprises alarm time and alarm classification numbers. In a specific application example, the characteristic parameters may include: horizontal channel antenna peak power, transmitter peak power, transmission to antenna power ratio, horizontal channel antenna power zeroing, transmitter power zeroing, reflectivity desired value, short pulse system calibration constant, long pulse system calibration constant, speed desired value 4, speed measured value 4, KD calibration desired value, KD calibration measured value, horizontal channel pre-filter power, horizontal channel post-filter power.

In S10, the stepwise regression method may adopt a Forward introduction method (Forward selection), a Backward elimination method (Backward elimination), and a Bidirectional elimination method (Bidirectional elimination). The step-by-step regression method is further described below.

If the forward-introduction method is used, firstly, only one independent variable which can explain the maximum variation of the dependent variable is added into the model, then, another independent variable is added in an attempt, whether the variation of the dependent variable which can be explained by the whole model is obviously increased or not is checked (F-test, t-test and the like), and iteration is repeated until no independent variable is added and the condition of adding the model is met.

If a backward elimination method is used, as opposed to a forward introduction method, all variables are put into a model, then an attempt is made to eliminate one of the independent variables from the model, and whether the variation of the whole model interpretation dependent variable has significant variation or not is checked, so that the variable with the least reduction in interpretation amount is eliminated; and repeating iteration until no independent variable meets the condition of elimination.

If the bidirectional elimination method is used, in order to combine the forward introduction method and the backward elimination method, an independent variable which is used for independently explaining the maximum variation of a dependent variable is added into a model, then another independent variable is added in an attempt, all variables in the whole model are checked, if the dependent variable is remarkably increased, the independent variable is reserved, the variable with the insignificant effect is eliminated, and iteration is repeated until an optimal variable combination is finally obtained.

And S20, taking the characteristic parameter y (generally, the peak power of the transmitter or the peak power of the horizontal channel antenna) with the maximum correlation coefficient in the characteristic parameters as a reconstruction parameter target of the reconstruction model. Constructing a reconstruction model by using a Long short term Memory Network (LSTM) model, and performing fitting reconstruction on y by using n-1 characteristic parameters except the characteristic parameter y to obtain a reconstruction value

。

As shown in fig. 3 and 4, when the reconstruction model is constructed using the LSTM model, time-series state data X ═ X is input⁽¹⁾,…,X^(m)Outputting a reconstructed parameter value Y ═ Y through a long-time memory network LSTM⁽¹⁾,…,Y^(m)}. Wherein the LSTM comprises forgetting to remember the gate: gamma-shaped_f＝σ(W_f[a^<t-1>，x^<t>]+b_f) (ii) a Input gate (update gate): gamma-shaped_u＝σ(W_u[a^<t-1>，x^<t>]+b_u) (ii) a Updating the candidate value:an output gate: gamma-shaped_o＝σ(W_o[a^<t-1>，x^<t>]+b_o) (ii) a Updating the value:

nonlinear activation: a is^<t>＝Γ_o×tanh c^<t>. The calculation process is as follows:

the forgetting gate Γ f will read the nonlinear activation a of the last cell^<t-1>And input data x of the current LSTM cell^<t>Outputting a value between 0 and 1 to each of the states c of the cells in the LSTM^<t-1>The numbers in (1). 1 means "complete retention" and 0 means "complete discard".

Input gate gamma_uIs a sigmoid layer that decides what values are to be updated. Then, a new candidate value vector is created using a tanh layer

. Updated vector c^<t>Determined by the input gate and the forgotten gate.

Output gate gamma_oThe output value is determined by first running a sigmoid layer Γ to determine which portion of the LSTM cell state will be output. Next, the cell state c^<t>Is processed by tanh (to obtain a value between-1 and 1) and is coupled to the output gate Γ_oWill eventually output only that portion of the determined output.

S30, to the reconstruction value

And the difference value of the measured value y is used as a probability-based quantification standard.

In a specific application example, S30 may further include:

and i, analyzing the distribution of y, and if a plurality of operating states exist, superposing a plurality of Gaussian distributions. As shown in fig. 2, the distribution of the peak power of the transmitter is a superposition of two gaussian distributions, less than 300 obeying the (1) distribution and more than 300 obeying the (2) distribution.

And ii, assuming that the predicted value and the measured value are independent, if the predicted value is subjected to (1) distribution, and assuming that the radar operating state is normal, the measured value is also subjected to (1) distribution, and the difference value of the two is subjected to N (0,2 sigma ^2) distribution.

And iii, carrying out normalization processing on the probability of the difference value to obtain a final quantization result:

if the predicted value is equal to the measured value, η is equal to 1, and the more the predicted value deviates from the measured value, the closer η is to 0.

And S40, carrying out time interval statistics on the quantized result, and filtering false alarms to obtain a final diagnosis result.

In a specific application example, S40 may further include:

setting a probability decision threshold η₀，η＜η₀Deeming the transmitter to be faultyStatus point

And ii, taking m continuous time points as time interval statistics, and considering that a fault occurs when the fault state point is greater than 0.3 m.

It is to be understood that the exemplary embodiments described herein are illustrative and not restrictive. Although one or more embodiments of the present invention have been described with reference to the accompanying drawings, it will be understood by those of ordinary skill in the art that various changes in form and details may be made therein without departing from the spirit and scope of the present invention as defined by the following claims.