CN110956764A

CN110956764A - Buried line damage early warning device and method based on neural network

Info

Publication number: CN110956764A
Application number: CN201911220856.9A
Authority: CN
Inventors: 赵栓峰; 李卿; 郭卫
Original assignee: Xian University of Science and Technology
Current assignee: Shaanxi Wisdom Luheng Electronic Technology Co ltd
Priority date: 2019-12-03
Filing date: 2019-12-03
Publication date: 2020-04-03
Anticipated expiration: 2039-12-03
Also published as: CN110956764B

Abstract

The invention discloses a buried line damage early warning device and an early warning method based on a neural network.A monitoring pile is arranged along the direction of a buried line at a distance of 50 meters, and comprises an IMU inertia measurement unit with an SPI (serial peripheral interface), a CPU with the SPI, a 4G transmission module, an alarm and a switch; this external solar panel of monitoring stake, the battery is located inside the monitoring stake and is connected with solar panel and CPU, and solar panel installs on the top of monitoring stake, and the alarm is installed on the top of monitoring stake, and IMU installs on the inside plane of bottom of monitoring stake, and CPU and 4G transmission module install on the IMU next door, realize IMU and CPU's information transmission through the SPI interface. The invention introduces the deep neural network into the working condition of preventing the underground line from being damaged, realizes the monitoring of the construction of the trend area of the underground line, and has higher practical value for ensuring the integrity and the working state of the underground line.

Description

Buried line damage early warning device and method based on neural network

Technical Field

The invention relates to an early warning protection device for an underground line, which can prevent the underground line from being touched and damaged by mistake in construction.

Background

The buried cable can enhance the reliability of the urban power transmission network, eliminate the contradiction of line crossing, avoid visual pollution, improve the land utilization rate and the like, so that the buried cable is increasingly widely applied in society. Although the advantages of the underground line are very outstanding, there are many cases in daily life where the underground line is damaged due to construction.

Disclosure of Invention

The invention provides a buried line damage early warning device and method based on a neural network aiming at the problem of buried line damage caused by construction in a traditional buried line arrangement area.

The technical scheme of the invention is as follows: a buried line damage early warning device based on a neural network is characterized by comprising monitoring piles arranged along the direction of a buried line at a distance of 50 meters, wherein each monitoring pile comprises an IMU (inertial measurement unit) with an SPI (serial peripheral interface), a CPU (central processing unit) with an SPI, a 4G transmission module, an alarm and a switch;

this external solar panel of monitoring stake, the battery is located inside the monitoring stake and is connected with solar panel and CPU, and solar panel installs on the top of monitoring stake, and the alarm is installed on the top of monitoring stake, and IMU installs on the inside plane of bottom of monitoring stake, and CPU and 4G transmission module install on the IMU next door, realize IMU and CPU's information transmission through the SPI interface.

A buried line damage early warning method based on a neural network is characterized by comprising the following steps:

(1) acquiring acceleration data in three directions and angular velocity data around the three directions by the IMU to obtain vibration signals reflecting ground vibration vectors, accessing a data buffer of the IMU by the CPU2 through an SPI (serial peripheral interface) to read the vibration signals and send the vibration signals to a 4G communication module, transmitting the vibration signals to a server by the 4G communication module, and converting the vibration signals into a vibration frequency domain diagram after the server receives the vibration signals;

(2) inputting the training data set into the built residual error neural network for training to obtain a trained residual error neural network; the trained residual error neural network is utilized, a vibration frequency domain diagram on a server is used as a detection sample and is input into the residual error neural network for detection, and detection results are divided into two types: in a construction state and in an unfinished state; finally, the neural network outputs the two detection results to a server;

(3) the method comprises the steps that a server sends a detection result output by a residual error neural network to a monitoring pile, when a CPU of the monitoring pile receives construction data transmitted by the server, a circuit switch of an alarm is closed to enable the alarm to give an alarm, and the site and the coordinates of the buried line construction are sent to a person in charge and a unit in charge of the buried line in a short message mode; when the CPU of the monitoring pile receives the non-construction data transmitted by the server, the circuit switch of the alarm cannot be closed, and the IMU continues to acquire the vibration signal.

The invention has the advantages that:

(1) the method introduces the deep neural network into the working condition of preventing the damage of the buried line, is applied to the protection of the buried line for the first time, can quickly and effectively detect the construction state and has real-time property.

(2) Compared with the traditional buried wire warning pile, the warning and protection functions are greatly improved.

(3) The invention has full automation and real-time performance in the working process.

Drawings

FIG. 1 is an overall flow diagram of the method of the present invention.

Fig. 2 is a schematic view of the installation of the warning device.

In the figure: the system comprises 1-an IMU with an SPI interface, 2-a CPU with an SPI interface, 3-4G communication modules, 4-a storage battery, 5-a monitoring pile, 6-a solar panel, 7-an alarm and 8-a switch.

Fig. 3 is a residual network detection diagram.

Detailed Description

The present invention is described in further detail below with reference to the attached drawings.

The method comprises the following steps: early warning device installation

As shown in fig. 2. The utility model provides a buried line destroys early warning device based on neural network, comprises IMU (be inertia measurement unit) 1, a CPU2, 4G communication module 3, battery 4, monitoring stake 5, solar panel 6, alarm 7 and the switch 8 that have the SPI interface, one have the SPI interface. The monitoring piles 5 are arranged along the buried line run at a distance of 50 metres apart. Solar panel 6 is installed on the top of monitoring stake 5, is convenient for gather solar energy. Alarm 7 is installed on the top of monitoring stake 5, makes things convenient for the propagation of the sound of warning. And the information transmission between the IMU and the CPU is realized by adopting an SPI interface. The IMU (inertial measurement unit) 1 is installed on the plane inside the bottom end of the monitoring pile 5, so that the requirement of a base plane for installing a sensor is met, and measurement of acceleration in the horizontal direction and the vertical direction is conveniently obtained. The CPU with the SPI interface and the 4G communication module 3 are arranged beside the IMU (inertial measurement unit) 1, so that the data processing and transmission are ensured.

Step two: vibration signal acquisition

In order to acquire the vibration signal, the entire apparatus needs to be operated. The solar panels 6 convert solar energy into electrical energy during the day so that the entire apparatus remains in operation. And the collected redundant electric energy is transmitted to the storage battery 4 to be stored, and the storage battery 4 supplies energy to the whole device under the dark environment.

The IMU (i.e. the inertial measurement unit) 1 of the invention adopts a sensor of an MPU9250 model. The sensor can acquire acceleration data in three directions and angular velocity data around the three directions to obtain a vibration signal reflecting a ground vibration vector. The CPU accesses the data buffer of the IMU through the SPI interface to read the vibration signal. When the amplitude exceeds a certain threshold value, the CPU sends the vibration signal to the 4G communication module 3, the 4G communication module 3 establishes a communication relation with the server, and the vibration signal is sent to the server for processing and analysis. If the amplitude of the data does not exceed the set threshold value, the CPU controls the whole device to keep low energy consumption of the whole device when the device does not work or data is invalid, and the energy is saved.

Step two: data pre-processing on a server

The vibration signal transmitted by the 4G communication module 3 is transmitted to the server. At the server, a corresponding server program is set to monitor the communication of the port, and then when the 4G communication module 3 sends a communication request to the port, the data transmission can be completed. After the server receives the vibration signal, the vibration signal is preprocessed. The collected vibration signals are converted into a frequency domain diagram with frequency on the horizontal axis and frequency amplitude on the vertical axis. The vibration signal is converted into a frequency domain diagram so as to facilitate the detection and judgment of data by using a neural network.

Step three: residual neural network detection

The training steps of the neural network are as follows:

1. constructing a training data set

When a neural network is constructed, firstly, a training data set is needed, the training data set is used for extracting and storing features, and then, the features are compared with actual data to realize identification; to ensure the accuracy of the data set, we need a large number of samples. The data of vehicles passing by, excavator construction, road pressing of road rollers, hoeing, people meeting, piling and well drilling under various conditions are collected, and the total number of the collected data is 10000. We divided it into 2 groups, one is training in the presence of construction; the other is training without construction. Before training, labels need to be made on the vibration frequency domain graph, namely, the vibration frequency domain graph is divided into a vibration frequency domain graph during construction and a vibration frequency domain graph without construction. And then inputting the classified vibration frequency domain graph serving as a training data set into a residual error network for training.

2. Training residual neural network

Compared with the common convolutional neural network training for identification, the residual neural network has more accurate result, reduced complexity, reduced parameters and simpler optimization, and solves the problems of Didu diffusion gradient explosion and the like. Therefore, a residual neural network is selected and built for classifying the images, namely, whether construction exists or not is detected. As shown in fig. 3. The training process mainly comprises the following steps:

1) initialization of weight for residual error network

Here, an initialization method suitable for the ReLU activation function is mainly used:

wherein h is_i、w_iRespectively representing the height and width of the convolution kernel in the convolution layer, and d_iRepresenting the number of current convolution kernels.

2) The input data is propagated to the output value through the convolution layer, the pooling layer and the full connection layer

The forward propagation process of the convolution layer carries out convolution operation on input data through convolution kernel, collected image data form a local receptive field in a network, then convolution calculation is carried out on the local receptive field, namely, a weight matrix and a characteristic value of an image are weighted and added with an offset value, and then the local receptive field is output through a ReLU activation function;

wherein, the function formula of the ReLU activation function is as follows:

f(x)＝max(0,x)

the vibration frequency domain image extracted from the previous layer (convolution layer) is subjected to pooling operation of the pooling layer, so that the dimensionality of image characteristic data is reduced, and overfitting can be avoided;

and after the collected vibration frequency domain image is used as output to extract the characteristics of the convolution layer and the pooling layer, the extracted characteristics are transmitted to the full-connection layer, and classification is carried out through the full-connection layer to obtain a classification model and obtain a final result.

When the output result of the neural network does not accord with the expected value, namely when the image of the input construction state is compared and identified as the situation of non-construction, the back propagation is needed to be carried out, the error between the result and the expected value is solved, the error is returned layer by layer, the error of each layer is calculated, and then the weight value is updated. The main purpose of this process is to adjust the network weights (convolution kernels) by training the samples and the expected values. The essence of this process is: the vibration frequency domain image data passes through a convolution layer, a pooling layer and a full connection layer from an input layer to an output layer, and the transmission process of the image data among the layers inevitably causes data loss, thereby causing errors. The error value caused by each layer is different, so we need to find the total error of the network, and introduce the error into the network to find the proportion of each layer to the error.

The error E between the output a (n) of the output layer n and the target value is determined. The calculation formula is as follows:

E＝-(y-a(n))*f(z(n))

where f (z (n)) is the derivative function value of the excitation function.

Comparing the calculated error E with the set minimum allowable error epsilon, if E is less than epsilon, executing a test network to read the vibration frequency domain image, and judging whether construction exists; if E > epsilon, the convolution kernel and the sliding step length need to be readjusted, and the network needs to be retrained.

And then inputting the training data set into the built residual error neural network, and training the residual error neural network to obtain the trained residual error neural network.

The trained residual error neural network is utilized, a frequency domain graph on a server is used as a detection sample to be input into the neural network, and detection results are divided into two types: 1. the detection result conforms to the construction comparison model, namely the data is in a construction state. 2. The detection result accords with the comparison model which is not constructed, and is in the state of not construction. The final neural network will output both results to the server.

Step four: information feedback processing

And obtaining a neural network detection result through the third step. And if the detection result is the construction condition, the server warns the information. And when the alarm is given, the server can send the location and the coordinates of the buried line to a buried line responsible unit and a responsible person in a short message form. And the short message application is required to be created on the server for sending the information in the form of the short message, and the APP _ Key, the APP _ Secret and the APP access address are acquired (a channel is also required to be acquired for the international short message). For sending the domestic short message, the short message signature is required to be applied to obtain a signature channel number. Therefore, the short message can be sent. Meanwhile, the server sends construction data to the 4G communication module 3 and transmits the construction data to the CPU through the 4G communication module 3, and the CPU receives the data and then orders the alarm circuit switch 8 to be closed to communicate with the alarm circuit, so that the alarm 7 gives an alarm. If the detection result is not the construction condition, the server sends the detection result to the CPU through the 4G communication module 3, the CPU cannot close the circuit switch 8 of the alarm, and vibration data can be continuously collected.

The above-mentioned embodiments are merely illustrative of the present invention, and not restrictive, and other examples of the method are within the scope of the invention, for example, the urban garbage truck loading volume measurement, the measurement of road surface profile of road surface, or the measurement of raised portion volume, etc. are within the spirit and scope of the claims, and any modifications and changes made thereto are within the scope of the invention.

Claims

1. A buried line damage early warning device based on a neural network is characterized by comprising monitoring piles (5) which are arranged along the direction of a buried line at a distance of 50 meters, wherein each monitoring pile (5) comprises an IMU (1) with an SPI (serial peripheral interface), a CPU (2) with the SPI, a 4G transmission module (3), an alarm (7) and a switch (8);

this monitoring stake (5) external solar panel (6), battery (4) are located monitoring stake (5) inside and are connected with solar panel (6) and CPU (2), the top at monitoring stake (5) is installed in solar panel (6), the top at monitoring stake (5) is installed in alarm (7), IMU (1) is installed on the inside plane in bottom of monitoring stake (5), install on IMU (1) next door CPU (2) and 4G transmission module (3), realize IMU (1) and CPU (2)'s information transmission through the SPI interface.

2. The warning device for the damage of the buried line based on the neural network as claimed in claim 1, wherein the warning method of the warning device comprises the following steps:

(1) acquiring acceleration data in three directions and angular velocity data around the three directions through an IMU (inertial measurement Unit), obtaining vibration signals reflecting ground vibration vectors, accessing a data buffer of the IMU through an SPI (serial peripheral interface) by a CPU (Central processing Unit) to read the vibration signals and send the vibration signals to a 4G communication module, transmitting the vibration signals to a server by the 4G communication module, and converting the vibration signals into a vibration frequency domain graph after the server receives the vibration signals;