CN111365624A - A kind of intelligent terminal and method for leakage detection of brine pipeline - Google Patents

Abstract

The invention relates to the technical field of halogen transmission pipeline detection, and discloses an intelligent terminal and method for leakage detection of a halogen transmission pipeline. The intelligent terminal includes an STM32F7 chip, a piezoelectric composite sensor, a filter circuit module, a high-precision A/D conversion circuit, and a GPS. module, external SDRAM module, SD card module, 4G communication module. The detection method includes acquiring the historical data set H; performing discrete S transform on it, and dividing it into training set Z and test set T; training and determining the LSTM model; The data is fed into an already trained LSTM model to predict if a leak will occur. Compared with the prior art, the present invention fully understands the data characteristics of the brine pipeline at a certain time through S transformation, and solves the time correlation between data through LSTM modeling, avoids artificially setting thresholds, and increases the accuracy of leakage judgment.

Description

A kind of intelligent terminal and method for leakage detection of brine pipeline

technical field

The present invention relates to the technical field of detection of halogen transportation pipelines, in particular to an intelligent terminal and method for leakage detection of brine transportation pipelines.

Background technique

With the increase of the service life of the pipeline, the leakage of the pipeline is increasing, and the leakage of the pipeline will not only cause serious pollution to the environment, but also bring huge economic losses to the enterprise. Therefore, it is of great significance to monitor the pipeline in real time, to determine the occurrence of faults in time, and to precisely locate the leak point.

At present, the detection methods of pipeline leakage mainly include: 1. Negative pressure wave method; 2. Infrasound wave method; 3. Distributed optical fiber early warning method, etc. When there is a small leak in the pipeline, the signal change is not obvious. When using these methods to detect tiny leaks, the problem of low detection accuracy is common.

SUMMARY OF THE INVENTION

Purpose of the invention: In view of the problems existing in the prior art, the present invention provides an intelligent terminal and method for detecting leakage of a halogen transmission pipeline that can solve the problem of low accuracy of the existing pipeline leakage detection algorithm.

Technical solution: The present invention provides an intelligent terminal for leakage detection of halogen transmission pipeline, including STM32F7 chip, piezoelectric composite sensor, filter circuit module, high-precision A/D conversion circuit, GPS module, external SDRAM module, SD card module , 4G communication module;

The piezoelectric composite sensor is used to detect the pressure signal and vibration signal inside the halogen transmission pipeline, and the collected analog signal is converted into a digital signal through a filter circuit module and a high-precision A/D conversion circuit, and is transmitted to the digital signal through SPI. The STM32F7 chip writes the collected data into the external SDRAM module. The STM32F7 chip analyzes the data, stores the suspected leaked signal in the SD card, and transmits it to the host computer through the 4G module.

Further, the intelligent terminal synchronously collects the vibration signal and pressure signal at a certain moment upstream and downstream of the halogen transmission pipeline through the second pulse signal of the GPS module, and uses the GPS module to add time stamps to the collected data.

Further, the high-precision A/D conversion circuit adopts ADS1274.

The invention also discloses a method for leak detection of a halogen transmission pipeline, comprising the following steps:

Sept1: Obtain the historical data set H of the pressure and vibration signals of the inner wall of the halogen pipeline;

Sept2: Perform discrete S-transform on the historical data set H, record the S-transformed data set D, and divide the S-transformed data set D into a training set Z and a test set T;

Sept3: Build the LSTM model, select the training set Z in Sept2 to train the LSTM model, and adjust the parameters until the network effect reaches the expected effect, and establish the LSTM model;

Sept4: Use the test set T in Sept2 as the input of the LSTM model to verify the accuracy of the model;

Sept5: Simultaneously sample the current vibration and pressure signals of the halogen pipeline, and perform S discrete transformation on the current sampled data;

Sept6: Input the current data after S transformation into the trained LSTM model to predict whether leakage occurs.

Preferably, the discrete form of the S-transform is as follows:

Among them, N is the total number of sampling points of the signal, T is the adoption period, X[kT](k=0,1,2...N-1) is the sampled signal, n is the sequence number of the nth point, m is the direction Frequency point for left translation, j is an imaginary unit.

Preferably, the specific steps of the S transformation are as follows:

Step1.1: Collect the pressure signal X[kT] of the inner wall of the brine pipeline;

Step1.2: Perform fast Fourier transform on the pressure signal X[kT] to get

Step1.3: When n=0, go to Step1.4, and execute Step1.4 and Step1.5; when n is not 0, for a given frequency point n, calculate the FFT of the Gaussian window function:

(j→m,m为频率点)，并转Step1.6；

(j→m, m is the frequency point), and go to Step1.6;

Step1.4: Calculate the S transform S[kt,0] of the time series corresponding to the given time point k according to the formula of n=0 (k=0,1,2,...,N-1 represents the time sampling point);

Step1.5: Set k=k+1, repeat Step1.4 until the S-transformation of all points is completed, and end the S-transformation;

向左平移m个频率点得到频谱函数

Step1.6: Put in Step1.2

Shift m frequency points to the left to get the spectral function

再进行反傅里叶变换，即可得到频率点n对应的S变换谱

Step1.7: Convolve the Gaussian window function after Fourier transform and the shifted spectral function to get

Then perform the inverse Fourier transform to obtain the S-transform spectrum corresponding to the frequency point n

Step 1.8: Let n=n+1, and repeat Step1.6 and Step1.7 until the S-transformation of all frequency points is calculated.

Preferably, the LSTM model formula includes:

1) Forget gate: Conditionally select which information is discarded from the current unit, the formula is as follows:

f _t =σ(W _f .[h _t-1 ,X _t ]+b _f )

where f _t ∈ [0,1], 1 means "completely preserved", 0 means "completely discarded", where h _t-1 is the output of the LSTM at the previous moment, X _t is the current input of the cell, W _f is the weight matrix of the forget gate, b _f is the bias, σ is the activation function, usually the Sigmoid function is used, that is

2) Input gate: Conditionally decide what information to store in the cell, the formula is as follows:

i _t =σ(W _i .[h _t-1 ,X _t ]+ _bi )

表示单元状态更新值，i_t控制

的哪些特征用于更新当前的状态，从而生成新的状态

Among them, the input gate i _t is generated by X _t and h _t-1 through the Sigmoid function calculation, i _t is a vector between [0, 1] like f _t ; the other is composed of X _t and h _{t- 1} A vector generated by the tanh activation function

Represents the unit state update value, it _controls

which features of is used to update the current state to generate a new state

3) Output gate: conditionally determine which information needs to be output, and output the information; the formula is as follows:

O _t =σ(W _o .[h _t-1 ,X _t ]+b _o )

h _t =O _t *tanh(C _t )

Among them, a sigmoid layer is run to determine which part of the cell state will be output, then the cell state is processed through tanh to get a value between -1 and 1, and it is multiplied by the output of the sigmoid gate, In the end, only the part that we are sure to output will be output.

Preferably, in the Sept3, a cross-entropy loss function is used to describe the difference between the actual output and the expected output, and the stochastic gradient descent method is used to minimize the cross-entropy loss function, and the parameters of the LSTM model are adjusted until the model meets the requirements, and its cross The entropy loss function formula is:

表示t时刻输卤管道发生泄漏的实际概率，Z为训练集，z为训练集中的一个数据，p(y_t|h_t)表示模型预测的概率，即当输卤管道发生泄漏时的概率为： p(y_t|h_t)＝softmax(θh_t+b)，其中

θ＝(θ₁，θ₂...θ_Z)为权重矩阵，b为偏置，设“1”标记为发生泄漏，“0”表示管道未发生泄漏。in,

Indicates the actual probability of leakage of the brine pipeline at time t, Z is the training set, z is a data in the training set, p(y _t |h _t ) represents the probability predicted by the model, that is, the probability of leakage of the brine pipeline is : p(y _t |h _t )=softmax(θh _t +b), where

_θ =(θ ₁ , θ ₂ .

Beneficial effects:

1. The present invention can fully understand the three-dimensional characteristics of the time-frequency-mode data of the brine pipeline at a certain moment through the S transformation, as the input of the LSTM model, so that the model can better learn the characteristics of the data, thereby increasing the model judgment. accuracy.

2. The present invention uses LSTM to model the pressure signal and vibration signal inside the halogen pipeline, and solves the time correlation between the data.

3. In the prior art, a threshold value needs to be set for the leakage judgment of the brine pipeline, and the detection method of the present invention can avoid artificially setting the threshold value and increase the accuracy of leakage judgment.

Description of drawings

1 is a schematic block diagram of a leak detection device of the present invention;

Fig. 2 is the circuit connection diagram of the intelligent terminal of the present invention;

Fig. 3 is the overall block diagram of the present invention;

Fig. 4 is the S transform flow chart of the present invention;

Fig. 5 is the LSTM model flow chart of the present invention;

Fig. 6 is the simulation data diagram of the present invention;

FIG. 7 is a data diagram after S-transformation of the present invention.

Detailed ways

The present invention will be described in detail below with reference to the accompanying drawings.

As shown in Figure 1, an intelligent terminal for leakage detection of halogen transmission pipeline includes STM32F7 chip, piezoelectric composite sensor, filter circuit module, high-precision A/D conversion circuit, GPS module, external SDRAM module, SD card module, 4G communication module. The intelligent terminal synchronously collects the vibration signal and pressure signal at a certain moment upstream and downstream of the halogen transmission pipeline through the second pulse signal of the GPS module, and uses the GPS module to add time stamps to the collected data, which is convenient for subsequent data processing. The collected analog data is removed by the filter circuit module to remove clutter interference, and the filtered analog signal is converted to analog-to-digital by the A/D conversion circuit.

是否产生下降沿。若产生下降沿，则说明数据已准备好，此时开始传输数据。该智能终端将这些带有时间戳的、未被分析的数字信号暂时存储在外扩的SDRAM中，以缓解STM32F7的计算压力，从而可以在该智能终端的STM32F7中进行简单的数据分析。对于该智能终端分析后疑似泄漏的信号，存储到SD 卡中。同时，该智能终端利用4G模块将采集到的输卤管道上下游数据上传到云端，将大量的数据进行汇总分析。该智能终端的电路连接图如图1、图2所示。The high-precision A/D conversion circuit adopts ADS1274. ADS1274 is a 24-bit successive approximation analog-to-digital converter, which contains four AD conversion circuits. ADS1274 and STM32F7 perform data transmission through SPI. The STM32F7 chip first accepts the PPS interrupt signal of the GPS through the capture function of the timer. When the timer captures the rising edge, the data ready signal of the ADS1274 is detected in the PPS interrupt processing function.

Whether to generate a falling edge. If a falling edge occurs, it means that the data is ready, and the data transmission starts at this time. The intelligent terminal temporarily stores these time-stamped, unanalyzed digital signals in the externally expanded SDRAM to relieve the computing pressure of the STM32F7, so that simple data analysis can be performed in the STM32F7 of the intelligent terminal. The signals suspected to be leaked after analysis by the smart terminal are stored in the SD card. At the same time, the intelligent terminal uses the 4G module to upload the collected upstream and downstream data of the halogen pipeline to the cloud, and summarizes and analyzes a large amount of data. The circuit connection diagrams of the intelligent terminal are shown in Figure 1 and Figure 2.

Figure 2 is a filter circuit diagram of the circuit. What the present invention needs to analyze is the AC signal generated by the internal vibration of the halogen transmission pipeline, so the filtering and amplifying circuit of the first part amplifies the AC signal, and the DC signal is used as the carrier signal and remains unchanged. The second part is the differential amplifier circuit. For the DC signal, the differential amplifier circuit is a common-mode input, and the voltage at the output terminal is 0, which avoids the interference of the DC voltage and also amplifies the required AC signal.

The invention also discloses a method for leak detection of a halogen transmission pipeline, the overall flow chart of which is shown in Figure 3 , the total number of sampling points of the signal is set as N, and the adoption period is set as T. The detection method mainly includes the following steps:

Sept1: Obtain the historical data set H of the pressure and vibration signals of the inner wall of the halogen pipeline.

The pressure signal of the inner wall of the pipeline is obtained by the piezoelectric composite sensor (the vibration signal is analyzed in the same way), and the sampled signal is set as X[kT] (k=0,1,2...N-1).

Sept2: Perform discrete S transformation on the historical data set H, record the S transformed data set D, and divide the S transformed data set D into training set Z (70% of the total data) and test set T (30% of the total data) %)

The discrete form of the S-transform is as follows:

Among them, N is the total number of sampling points of the signal, T is the adoption period, X[kT](k=0,1,2...N-1) is the sampled signal, n is the sequence number of the nth point, m is the direction Frequency point for left translation, j is an imaginary unit.

Discrete S-transformation is performed on all collected signals (historical data set H), and the S-transformed data set is recorded.

The specific steps of S transformation are shown in Figure 4:

Step2.1: Collect the pressure signal X[kT] on the inner wall of the data pipeline.

Step2.2: Perform fast Fourier transform on X[kT] of the input signal to get

Step2.3: When n=0, go to Step2.4, and execute Step2.4 and Step2.5; when n is not 0, for a given frequency point n, calculate the FFT of the Gaussian window function:

(j→m,m为频率点)，并跳转到Step2.6。

(j→m, m is the frequency point), and jump to Step2.6.

Step2.4: Calculate the S-transform S[Kt,0] of the time series corresponding to the given time point k according to the formula of n=0 (k=0,1,2,...,N-1 represents the time sampling point).

Step2.5: Set k=k+1, and repeat Step2.4 until the S transformation of all points is completed.

向左平移m个频率点得到

Step2.6: Put in Step2.2

Shift m frequency points to the left to get

再进行反傅里叶变换，即可得到频率点n对应的S变换谱

Step2.7: Convolve the Gaussian window function after Fourier transform and the shifted spectral function to get

Then perform the inverse Fourier transform to obtain the S-transform spectrum corresponding to the frequency point n

Step2.8: Let n=n+1, and repeat Step2.6 and Step2.7 until the S-transformation of all frequency points is calculated.

Sept3: After S-transformation of N signal points, a complex matrix is obtained, and the LSTM model is built using the matrix. The training set Z in Sept2 is selected to train the LSTM model, and the parameters are adjusted until the network effect reaches the expected effect, and the LSTM model is established.

Sept4: Use the test set T in Sept2 as the input of the LSTM model to verify the accuracy of the model.

Sept5: Simultaneously sample the current vibration and pressure signals of the halogen pipeline, and perform S discrete transformation on the current sampled data.

Sept6: Input the current data after S transformation into the trained LSTM model to predict whether leakage occurs.

The above LSTM model, its formula includes:

1) Forget gate: Conditionally select which information is discarded from the current unit, the formula is as follows:

f _t =σ(W _f .[h _t-1 ,X _t ]+b _f )

where f _t ∈ [0,1], 1 means "completely preserved", 0 means "completely discarded", where h _t-1 is the output of the LSTM at the previous moment, X _t is the current input of the cell, W _f is the weight matrix of the forget gate, b _f is the bias, σ is the activation function, usually the Sigmoid function is used, that is

2) Input gate: Conditionally decide what information to store in the cell, the formula is as follows:

i _t =σ(W _i .[h _t-1 ,X _t ]+ _bi )

表示单元状态更新值，i_t控制

的哪些特征用于更新当前的状态，从而生成新的状态

Among them, the input gate i _t is generated by X _t and h _t-1 through the Sigmoid function calculation, i _t is a vector between [0, 1] like f _t ; the other is composed of X _t and h _{t- 1} A vector generated by the tanh activation function

Represents the unit state update value, it _controls

which features of is used to update the current state to generate a new state

3) Output gate: conditionally determine which information needs to be output, and output the information; the formula is as follows:

O _t =σ(W _o .[h _t-1 ,X _t ]+b _o )

h _t =O _t *tanh(C _t )

Among them, a sigmoid layer is run to determine which part of the cell state will be output, then the cell state is processed through tanh to get a value between -1 and 1, and it is multiplied by the output of the sigmoid gate, In the end, only the part that we are sure to output will be output.

A "1" is set to mark a leak, and a "0" means that the pipe is not leaking. The model parameters are adjusted by minimizing the cross-entropy loss by stochastic gradient descent until the accuracy of the model meets the requirements. Its loss function formula is:

表示t时刻输卤管道发生泄漏的实际概率，Z为训练集，z为训练集中的一个数据，p(y_t|h_t)表示模型预测的概率，即当输卤管道发生泄漏时的概率为： p(y_t|h_t)＝soft max(θh_t+b)，其中

θ＝(θ₁，θ₂...θ_Z)为权重矩阵，b为偏置。in,

Represents the actual probability of leakage of the brine pipeline at time t, Z is the training set, z is a data in the training set, p(y _t |h _t ) represents the probability predicted by the model, that is, the probability of leakage of the brine pipeline is : p(y _t |h _t )=soft max(θh _t +b), where

θ=(θ ₁ , θ ₂ . . . θ _Z ) is the weight matrix, and b is the bias.

Fig. 6 is the simulation diagram of the halogen transmission pipeline, adding a white noise signal to the collected signal to simulate the noise of the halogen transmission pipeline. Figure 7 is a two-dimensional contour map after S-transformation of the simulation data, that is, the time-frequency-mode graph of the brine pipeline. Finally, the S-transformed matrix is used as the input of the LSTM model. When the output is "1", the table leaks, and when the output is "0", it means that no leakage occurs.

The above-mentioned embodiments are only intended to illustrate the technical concept and features of the present invention, and the purpose is to enable those who are familiar with the art to understand the content of the present invention and implement it accordingly, and cannot limit the protection scope of the present invention. All equivalent transformations or modifications made according to the spirit of the present invention should be covered within the protection scope of the present invention.

Claims (8)

1. An intelligent terminal for detecting leakage of a brine conveying pipeline is characterized by comprising an STM32F7 chip, a piezoelectric composite sensor, a filter circuit module, a high-precision A/D conversion circuit, a GPS module, an external SDRAM module, an SD card module and a 4G communication module;

the piezoelectric type composite sensor is used for detecting pressure signals and vibration signals inside a halogen conveying pipeline, analog signals collected by the piezoelectric type composite sensor are converted into digital signals through a filter circuit module and a high-precision A/D conversion circuit, the digital signals are transmitted to an STM32F7 chip in an SPI mode, collected data are written into an external SDRAM module, the STM32F7 chip analyzes the data, suspected leaked signals are stored in an SD card, and the signals are transmitted to an upper computer through a 4G module.

2. The intelligent terminal for detecting the leakage of the brine transportation pipeline as claimed in claim 1, wherein the intelligent terminal synchronously acquires the vibration signal and the pressure signal at a certain time upstream and downstream of the brine transportation pipeline through a pulse per second signal of the GPS module, and the GPS module is used for adding a time stamp to the acquired data.

3. The intelligent terminal for detecting the leakage of the halogen transmission pipeline according to claim 1, wherein the high-precision A/D conversion circuit adopts ADS 1274.

4. A method for detecting leakage of a halogen conveying pipeline is characterized by comprising the following steps:

sept 1: acquiring a historical data set H of pressure and vibration signals of the inner wall of the brine conveying pipeline;

sept 2: performing discrete S transformation on the historical data set H, recording a data set D after the S transformation, and dividing the data set D after the S transformation into a training set Z and a test set T;

sept 3: building an LSTM model, selecting a training set Z in Sept2 to train the LSTM model, adjusting parameters until the network effect reaches the expected effect, and determining the LSTM model;

sept 4: taking a test set T in Sept2 as the input of an LSTM model, and verifying the accuracy of the model;

sept 5: synchronously sampling current vibration and pressure signals of the brine conveying pipeline, and performing S discrete transformation on current sampling data;

sept 6: and inputting the current data after S transformation into the well-trained LSTM model to predict whether leakage occurs.

5. The method of claim 4, wherein the discrete form of the S transformation is as follows:

where N is the total number of sampling points of the signal, T is the sampling period, X [ kT ] (k is 0,1,2 … N-1) is the sampled signal, N is the number of the nth point, m is the frequency point shifted to the left, and j is an imaginary unit.

6. The method for detecting the leakage of the brine transportation pipeline according to claim 5, wherein the specific steps of S transformation are as follows:

step1.1: collecting pressure signal X [ kT ] of the inner wall of the halogen conveying pipeline;

step1.2: for pressure signal X [ kT ]]Performing fast Fourier transform to obtain

Step1.3: when n is equal to 0, turning to step1.4, and executing step1.4 and step 1.5; when n is not 0, for a given frequency point n, the FFT of the gaussian window function is calculated:

and turning to Step1.6;

step1.4: calculating S transform S [ kt,0] of a time series corresponding to a given time point k according to an equation of N ═ 0 (k ═ 0,1,2, …, N-1 denotes time sampling points);

step1.5: making k equal to k +1, repeating Step1.4 until S transformation of all the points is completed, and ending the S transformation;

step1.6: the product obtained in Step1.2

The frequency spectrum function is obtained by translating m frequency points to the left

Step1.7: performing convolution on the Gaussian window function after Fourier transform and the spectrum function after translation to obtain

Then, inverse Fourier transform is carried out to obtain an S transform spectrum corresponding to the frequency point n

Step 1.8: let n be n +1, repeat step1.6, step1.7 until S transform of all frequency points is calculated.

7. The method of claim 4, wherein the LSTM model formula comprises:

1) forget the door: conditionally choose which information to discard from the current cell, the formula is as follows:

f_t＝σ(W_f.[h_t-1,X_t]+b_f)

wherein f is_t∈[0,1]1 means "complete retention", 0 means "complete discard", wherein h_t-1Representing the output, X, of the last instant LSTM_tIndicating the current input of the cell, W_fWeight matrix for forgetting gate, b_fFor biasing, σ is an activation function, usually a Sigmoid function is chosen, i.e.

2) An input gate: conditionally deciding which information to store in the cell, the formula is as follows:

i_t＝σ(W_i.[h_t-1,X_t]+b_i)

wherein, the input gate i_tIs composed of X_tAnd h_t-1Generated by calculation of Sigmoid function, i_tSame as f_tLikewise is one between [0,1 ]]The vector of (a); the other is formed by X_tAnd h_t-1A vector generated by the tanh activation function

Represents the cell state update value, i_tControl of

Is used to update the current state, thereby generating a new state

3) An output gate: conditionally deciding which information needs to be output, and outputting the information; the formula is as follows:

O_t＝σ(W_o.[h_t-1,X_t]+b_o)

h_t＝O_t*tanh(C_t)

wherein a Sigmoid layer is run to determine which part of the cell state will be output, then the cell state is processed through tanh to get a value between-1 and 1, and it is multiplied by the output of the Sigmoid gate, and finally only the part that we determine the output will be output.

8. The method of claim 7, wherein the Sept3 is characterized by that the difference between the actual output and the expected output is characterized by a cross entropy loss function, and the cross entropy loss function is minimized by using a stochastic gradient descent method, and the LSTM model is parametrically adjusted until the model meets the requirement, and when "1" is marked as leakage, and "0" indicates that no leakage occurs in the pipeline, the cross entropy loss function is formulated as:

wherein,

the actual probability of leakage of the brine conveying pipeline at the time t is shown, Z is a training set, Z is data in the training set, and p (y)_t|h_t) The probability of model prediction is represented, namely the probability when the halogen conveying pipeline leaks is as follows: p (y)_t|h_t)＝softmax(θh_t+ b) of

θ＝(θ₁，θ₂...θ_Z) B is the bias.

Images