CN111124015A

CN111124015A - Intelligent wind power inspection video monitoring method

Info

Publication number: CN111124015A
Application number: CN201911361782.0A
Authority: CN
Inventors: 姜浩; 张瑞香; 刘洪志
Original assignee: Laixi Branch Of Huadian Shandong New Energy Co Ltd
Current assignee: Laixi Branch Of Huadian Shandong New Energy Co Ltd
Priority date: 2019-12-26
Filing date: 2019-12-26
Publication date: 2020-05-08

Abstract

The invention relates to a method for monitoring and checking the real-time state of a wind turbine, in particular to an intelligent wind power inspection video monitoring method, which is characterized by comprising the following steps: s1, recording a video of the wind turbine generator in real time by one or more high-definition camera devices; s2, transmitting the video to a graphic algorithm workstation; s3, accurately identifying instruments and indicator lamps of the air outlet machine control cabinet, the SVG control cabinet, the GIS control cabinet, the main transformer conservator, the humiture instrument and other parts through a machine vision algorithm realized in the workstation; and S4, when the abnormal exceeding state is detected, alarming is timely sent out, and the alarm is pushed to the mobile terminal for display. The invention can realize real-time state monitoring and automatic inspection of the wind turbine generator, and the system automatically alarms when finding abnormality, thereby greatly reducing the workload of on-site maintenance and inspection and reducing the potential safety hazard of personnel.

Description

Intelligent wind power inspection video monitoring method

Technical Field

The invention relates to a method for monitoring and checking the real-time state of a wind turbine generator, in particular to an intelligent wind power inspection video monitoring method based on a deep learning convolutional neural network.

Background

With the rapid development of new energy in China, wind power as the main force of new energy brings difficulty to the management of a wind power plant due to the prominent characteristics of large number of units, multiple faults, remote location, severe environment, few personnel and the like, and the problems of frequent unit accidents, serious damage to most components, high management cost, difficult profit and the like mainly exist. In the traditional power plant inspection, the following difficulties can be encountered in the inspection process: the problems of difficulty in avoiding inspection omission, difficulty in recording inspection conditions, difficulty in standard inspection operation and isolated data are solved.

The inspection of the wind turbine generator is mainly performed inside and outside the engine room, the engine room mainly comprises an electrical system and a mechanical system, and the engine room mainly comprises blades. The electrical system focuses on the temperature, voltage, current, frequency and other indexes of the electrical equipment, and the mechanical system focuses on the oil temperature, oil level, vibration and other indexes. In the prior art, the inspection method of the interior of the engine room is manual inspection, and through visual inspection, sound inspection, measurement, detection and the like, the electrical part is mainly acquired through background data of electrical equipment, but due to the fact that the inspection time interval is long, the monitoring sensing equipment cannot be found in time through means after abnormality occurs, and major accidents are easily caused.

Disclosure of Invention

The invention provides an intelligent wind power inspection video monitoring method which can realize real-time state monitoring and automatic inspection of a wind turbine generator and automatically alarm when a system finds abnormality; the auxiliary maintenance device is installed, automatic maintenance can be achieved according to the actual state on site, the failure time of the equipment is reduced, and the service life of the equipment is prolonged. The tower climbing times of field maintainers are reduced, the field maintenance inspection workload is greatly reduced, and the potential safety hazard of personnel is reduced. Unmanned intelligent inspection of the wind turbine generator is comprehensively realized, and major accidents of the wind turbine generator are avoided.

The invention specifically adopts the technical scheme that:

an intelligent wind power inspection video monitoring method based on a deep learning convolutional neural network is characterized by comprising the following steps:

s1, recording a video of the wind turbine generator in real time by one or more high-definition camera devices;

s2, transmitting the video to a graphic algorithm workstation;

s3, accurately identifying instruments and indicator lamps of the air outlet machine control cabinet, the SVG control cabinet, the GIS control cabinet, the main transformer conservator, the humiture instrument and other parts through a machine vision algorithm realized in the workstation;

(1) marking input picture data, carrying out position calibration on a required instrument and an indicator lamp on a required picture, and extracting coordinates of the upper left corner and the lower right corner (X1, Y1, X2 and Y2);

(2) the picture is convoluted and pooled (which is equivalent to compressing the picture), and instruments, indicator lights and the like become a plurality of smaller pixels along with the continuous compression of the picture, so that the unique pixel characteristics of the instruments, indicator lights and the like after being compressed are embodied, and thus the instruments and indicator lights of different types can be distinguished more easily after being compressed;

(3) performing regression of label coordinates according to labels of different instruments, indicator lights and the like and characteristics of the labels on a compressed pixel level, and finding out the positions of the instruments, the indicator lights and the like;

(4) through a coordinate LOSS equation, continuous iteration is performed, and optimization adjustment of parameters is performed, so that the value of the LOSS equation is reduced to the minimum, wherein the coordinate LOSS equation is as follows:

(5) and finally, inputting pictures of the instrument to be detected, the indicator light and the like into the trained model for recognition, and finding out the positions of the instrument and the indicator light.

(6) Converting the identified indicator lights into HSV color spaces, distinguishing the on-off states of the indicator lights with different colors and the indicator lights with different colors in the HSV color spaces, and prompting the alarm state once the indicator lights are found not to be in accordance with the states under normal operation;

(7) the method comprises the steps of identifying the rotation angle of an identified instrument, detecting the switch outline by using an outline detection method, then calculating the current position rotation angle of the switch, and prompting an alarm state once the existing state gear of the switch is found to be inconsistent with the state under normal operation.

And S4, when the abnormal exceeding state is detected, alarming is timely sent out, and the alarm is pushed to the mobile terminal for display.

The method for monitoring the running state of the wind driven generator is combined with a video monitoring technology based on a deep learning convolutional neural network, the running state of the wind driven generator is accurately judged and judged by monitoring the states of an instrument and an indicator lamp through a high-definition network camera, and if the state of the equipment is abnormal, an alarm is sent out and pushed to corresponding workers to be processed in time, so that safety accidents are prevented.

Compared with the prior art, the invention has the advantages that:

1) the invention adopts a non-contact detection method, so that the detection equipment cannot be collided and abraded, cannot be influenced by extreme weather, reduces the consumption cost and improves the detection real-time property.

2) The method for deeply learning the convolutional neural network by the high-definition network camera is combined with a video analysis technology, so that the running state of the wind driven generator is monitored, the running state of the equipment is accurately judged, and real-time alarm is given once the abnormality of the equipment is detected.

3) The abnormal condition of the fan equipment is reduced, thereby being beneficial to the safety production and obtaining greater economic benefit.

4) The fan equipment abnormal state monitoring system based on video analysis is compact in structure and high in integration degree, can replace traditional manual inspection, greatly reduces the field inspection force, saves a large amount of manpower, and reduces the working strength of remote field inspection.

Drawings

FIG. 1 is a block flow diagram of a monitoring method of the present invention;

FIG. 2 is a schematic diagram of the convolutional neural network structure of the present invention.

Detailed Description

As shown in figure 1, the method comprises the steps of recording a video of a wind turbine generator in real time through high-definition camera equipment, transmitting the video to a graphic algorithm workstation, accurately identifying the equipment state through a machine vision algorithm in the workstation, and timely sending an alarm if the abnormal state is monitored.

As shown in fig. 2, the method of the present invention is that, images such as meters and indicator lights are convoluted and pooled (which is equivalent to compressing the images), and the meters and indicator lights show unique pixel characteristics of the meters and indicator lights after being compressed along with the continuous convolution of the images, so that the meters and indicator lights of different types can be distinguished more easily after being compressed; and then, according to the labels of different instruments, indicator lights and the like and the characteristics of the labels on the compressed pixel level, the regression processing of label coordinates is carried out by utilizing full connection, the position coordinates of the instruments, the indicator lights and the like are found, and the output layer outputs the detected position coordinate information of the instruments and the indicator lights.

Claims

1. An intelligent wind power inspection video monitoring method is characterized by comprising the following steps:

s2, transmitting the video to a graphic algorithm workstation;

and S3, accurately identifying instruments and indicator lamps of the air outlet machine control cabinet, the SVG control cabinet, the GIS control cabinet, the main transformer conservator, the temperature and humidity instrument and the like through a machine vision algorithm realized in the workstation.

2. The intelligent wind power inspection video monitoring method according to claim 1, wherein the machine vision algorithm in the step S2 is a deep learning convolutional neural network-based machine algorithm.

3. The intelligent wind power inspection video monitoring method according to claim 1, wherein the identification method in the step S3 includes the steps of:

(1) labeling input picture data: on the picture, the position of the instrument and the indicator lamp to be identified is calibrated, and the coordinates of the upper left corner and the lower right corner are extracted (X1, Y1, X2 and Y2);

(2) the picture is convoluted and pooled, the instrument and the indicator light become a plurality of smaller pixels along with the continuous compression of the picture, and the unique pixel characteristics of the instrument and the indicator light after being compressed are embodied, so that the instruments and the indicator lights of different types can be distinguished more easily after being compressed;

(3) performing regression of label coordinates according to labels of different instruments and indicator lamps and characteristics of the labels on the compressed pixel level, and finding out the positions of the instruments and the indicator lamps;

(5) finally, inputting pictures of the instrument to be tested and the indicator lamp into the trained model for recognition, and finding out the positions of the instrument and the indicator lamp;