阅读说明：本技术 变电站自动巡检方法和平台 (Substation's automatic detecting method and platform ) 是由 陈作伟 李勋 吕启深 刘顺桂 杨强 徐健 张裕汉 于 2019-08-19 设计创作，主要内容包括：本发明涉及一种变电站自动巡检方法和平台。变电站自动巡检方法包括：向巡检机器人发送巡检路径,以使巡检机器人按照巡检路径到达巡检位置,采集巡检位置处的设备的噪声数据,对设备和噪声数据进行对应标识,以及获取当前巡检机器人的定位信息；接收设备的噪声数据以及巡检机器人的定位信息,并根据预设规则对噪声数据进行处理,得到设备运行时的时频图；根据定位信息确认设备对应的训练完成的卷积神经网络模型,以及将设备的时频图输入到与设备对应的训练完成的卷积神经网络模型,利用卷积神经网络模型判断设备是否发生故障,从而将故障检测问题转化为图像识别问题,避免了对人工以及知识技能的依赖,同时提高了巡检的效率和可靠性。(The present invention relates to a kind of substation's automatic detecting method and platforms.Substation's automatic detecting method includes: to send polling path to crusing robot, so that crusing robot reaches inspection position according to polling path, acquire the noise data of the equipment at inspection position, corresponding mark is carried out to equipment and noise data, and obtains the location information of current crusing robot；The noise data of receiving device and the location information of crusing robot, and noise data is handled according to preset rules, obtain time-frequency figure when equipment operation；The convolutional neural networks model that the corresponding training of equipment is completed is confirmed according to location information, and the time-frequency figure of equipment is input to the convolutional neural networks model that training corresponding with equipment is completed, judge whether equipment breaks down using convolutional neural networks model, to convert problem of image recognition for fault detection problem, it avoids to artificial and knowledge expertise dependence, while improving the efficiency and reliability of inspection.)

1. a kind of substation's automatic detecting method characterized by comprising

Polling path is sent to crusing robot, so that the crusing robot reaches inspection position, acquisition according to polling path The noise data of equipment at the inspection position carries out corresponding mark to the equipment and the noise data, and obtains The location information of presently described crusing robot；

The noise data of the equipment and the location information of the crusing robot are received, and is made an uproar according to preset rules to described Sound data are handled, and time-frequency figure when equipment operation is obtained；

The convolutional neural networks model that the corresponding training of the equipment is completed is confirmed according to the location information, and is set described Standby time-frequency figure is input to the convolutional neural networks model that training corresponding with the equipment is completed, and utilizes the convolutional Neural net Network model judges whether the equipment breaks down.

2. substation's automatic detecting method as described in claim 1, which is characterized in that described to be made an uproar according to preset rules to described Sound data are handled, and time-frequency figure when equipment operation is obtained, comprising:

Short Time Fourier Transform is carried out to the noise data, obtain and stores time-frequency figure when running with the equipment.

3. substation's automatic detecting method as described in claim 1, which is characterized in that confirming institute according to the location information Before the convolutional neural networks model for stating the corresponding training completion of equipment, further includes:

The history run noise data under the equipment normal condition and under malfunction is acquired, and respectively to each noise The operating status of the corresponding equipment of data is demarcated, and noise data sample is formed；

The noise data sample is subjected to Short Time Fourier Transform, obtains time-frequency figure corresponding with the noise data sample；

The time-frequency figure is inputted pre-set convolutional neural networks as image training sample to be trained, it is described to obtain The convolutional neural networks model that training is completed, and store.

4. substation's automatic detecting method as described in claim 1, which is characterized in that sending institute to the crusing robot Before stating polling path, further includes:

According to the design drawing of the substation or high definition satellite image, the position for each equipment for needing to detect in substation is obtained Confidence breath；

The location information of the equipment detected as needed and the time for carrying out Noise Acquisition, generate the polling path.

5. such as weighing substation's automatic detecting method described in claim 1, which is characterized in that further include:

When determining the device fails, the warning message for prompting user equipment failure is generated.

6. substation's automatic detecting method as claimed in claim 2, which is characterized in that further include:

Retraining and optimization are carried out using the time-frequency figure pair convolutional neural networks model corresponding with the equipment.

7. a kind of substation's automatic detecting platform characterized by comprising

Crusing robot acquires the noise number of the equipment at the inspection position for reaching inspection position according to polling path According to, and the equipment and the noise data are identified, and obtain itself current location information；With

Earth station system is communicated to connect with the crusing robot, for receiving the noise data of the equipment and described patrolling Examine robot location information, the noise data is handled according to preset rules, obtain the equipment operation when Frequency is schemed, and confirms the convolutional neural networks model that the corresponding training of the equipment is completed according to the location information, will be described The convolutional neural networks model that the time-frequency figure input training corresponding with the equipment of equipment is completed, utilizes the convolutional Neural net Network model judges whether the equipment breaks down.

8. substation's automatic detecting platform as claimed in claim 7, which is characterized in that the crusing robot includes:

Motion-control module, for driving the crusing robot to reach the inspection position according to the polling path, and it is right The equipment and the noise data are identified, and obtain the current location information of the crusing robot；

Noise Acquisition module, for acquiring the noise data of the equipment at the inspection position；

Noise processed module is communicated to connect with the noise processed module, for carrying out the noise data and the equipment Corresponding mark；And

Communication module distinguishes communication link with the Noise Acquisition module, the noise processed module and the earth station system It connects, the motion-control module obtains the polling path by the communication module from the earth station system, described to make an uproar Sonication module sends the noise data to the earth station system by the communication module.

9. substation's automatic detecting platform as claimed in claim 8, which is characterized in that the noise processed module is also used to Before the noise data is sent to the earth station system, the noise data is clicked through according to the sampling of designated length Row segmentation and packing processing.

10. substation's automatic detecting platform as claimed in claim 8, which is characterized in that the earth station system includes:

Data processing module is communicated to connect with the communication module, for receiving the noise data of the equipment and described patrolling The location information of robot is examined, and the noise data is handled according to preset rules, is obtained when the equipment is run Time-frequency figure, and the polling path is sent to the crusing robot；

Noise storage and retrieval module, is electrically connected with the data processing module, for receiving and storing the location information, and The convolutional neural networks model that the corresponding training of the equipment is completed is determined according to the location information；

Failure analysis module is electrically connected with the noise storage and retrieval module, for receiving the time-frequency figure, and will enter into In the determining convolutional neural networks model, judge whether the equipment occurs event using the convolutional neural networks model Barrier；And

Patrol task planning module is electrically connected with the data processing module, for according to the Substation Design drawing or height Clear satellite image, obtains the location information of multiple equipment in substation, and the position of the equipment detected as needed Confidence breath and the time for carrying out Noise Acquisition, the polling path is generated, and be sent to the data processing module.

11. substation's automatic detecting platform as claimed in claim 10, which is characterized in that the noise storage and retrieval module It is also used to receive and store the time-frequency figure, and utilizes the time-frequency figure pair convolutional neural networks mould corresponding with the equipment Type carries out retraining and optimization.

12. substation's automatic detecting platform as claimed in claim 10, which is characterized in that the failure analysis module is also used to When determining the device fails, the warning message for prompting user equipment failure is generated.

13. substation's automatic detecting platform as claimed in claim 12, which is characterized in that the earth station system further includes police Module is reported, the alarm modules are electrically connected with the failure analysis module, for receiving the warning message, and according to the report Alert information generates buzzer.

Technical field

The present invention relates to the applicating and exploitation technical fields of crusing robot, more particularly to a kind of substation's automatic detecting Method and platform.

Background technique

Substation equipment is chronically under operating status, in order to ensure the safe and stable operation of electrical equipment, is found in time The defect or hidden danger of equipment need operations staff to carry out inspection, but heavy workload, low efficiency to field device, and testing result is past It is past to fall flat.Wherein noise data contains equipment running status information, can measure using noncontacting proximity sensor, It is more satisfactory accident analysis and detection means.Therefore noise detection technique is integrated on crusing robot, utilizes inspection Robot substitutes manual inspection, is conducive to save cost of labor, improves routing inspection efficiency and reliability.But it is based on conventional noise The method of signal analysis, which carries out abnormality detection, needs complicated expertise, therefore general employee may when analyzing noise signal It will appear fault, then detect reliability and detection efficiency is not high.

Summary of the invention

Based on this, it is necessary to a kind of substation's automatic detecting method and platform are provided, to improve the efficiency of inspection and reliable Property.

The present invention provides a kind of substation's automatic detecting methods, comprising:

Polling path is sent to crusing robot, so that the crusing robot reaches inspection position according to polling path, The noise data for acquiring the equipment at the inspection position carries out corresponding mark to the equipment and the noise data, and Obtain the location information of presently described crusing robot；

The noise data of the equipment and the location information of the crusing robot are received, and according to preset rules to institute It states noise data to be handled, obtains time-frequency figure when equipment operation；

The convolutional neural networks model that the corresponding training of the equipment is completed is confirmed according to the location information, and by institute The time-frequency figure for stating equipment is input to the convolutional neural networks model that training corresponding with the equipment is completed, and utilizes the convolution mind Judge whether the equipment breaks down through network model.

It is described in one of the embodiments, that the noise data is handled according to preset rules, obtain described set Time-frequency figure when received shipment row, comprising:

Short Time Fourier Transform is carried out to the noise data, obtain and stores time-frequency figure when running with the equipment.

In one of the embodiments, in the convolution for confirming the corresponding training completion of the equipment according to the location information Before neural network model, further includes:

The history run noise data under the equipment normal condition and under malfunction is acquired, and respectively to each described The operating status of the corresponding equipment of noise data is demarcated, and noise data sample is formed；

The noise data sample is subjected to Short Time Fourier Transform, obtains time-frequency corresponding with the noise data sample Figure；

The time-frequency figure is inputted pre-set convolutional neural networks as image training sample to be trained, to obtain The convolutional neural networks model that the training is completed, and store.

In one of the embodiments, before sending the polling path to the crusing robot, further includes:

According to the design drawing of the substation or high definition satellite image, each equipment for needing to detect in substation is obtained Location information；

The location information of the equipment detected as needed and the time for carrying out Noise Acquisition, generate the inspection road Diameter.

Substation's automatic detecting method in one of the embodiments, further include:

When determining the device fails, the warning message for prompting user equipment failure is generated.

Substation's automatic detecting method in one of the embodiments, further include:

Retraining and optimization are carried out using the time-frequency figure pair convolutional neural networks model corresponding with the equipment.

Based on the same inventive concept, the embodiment of the invention also provides a kind of substation's automatic detecting platforms, comprising:

Crusing robot acquires making an uproar for the equipment at the inspection position for reaching inspection position according to polling path Sound data, and the equipment and the noise data are identified, and obtain itself current location information；With

Earth station system is communicated to connect with the crusing robot, for receiving noise data and the institute of the equipment The location information for stating crusing robot is handled the noise data according to preset rules, when obtaining equipment operation Time-frequency figure, and according to the location information confirm the equipment it is corresponding training complete convolutional neural networks model, will The convolutional neural networks model that the time-frequency figure input training corresponding with the equipment of the equipment is completed, utilizes the convolution mind Judge whether the equipment breaks down through network model.

The crusing robot includes: in one of the embodiments,

Motion-control module, for driving the crusing robot to reach the inspection position according to the polling path, And the equipment and the noise data are identified, and obtain the current location information of the crusing robot；

Noise Acquisition module, for acquiring the noise data of the equipment at the inspection position；

Noise processed module is communicated to connect with the noise processed module, is used for the noise data and the equipment Carry out corresponding mark；And

Communication module is led to respectively with the Noise Acquisition module, the noise processed module and the earth station system Letter connection, the motion-control module obtain the polling path, institute by the communication module from the earth station system It states noise processed module and the noise data is sent to the earth station system by the communication module.

The noise processed module in one of the embodiments, be also used to the noise data is sent to it is described Before earth station system, the noise data is split according to the sampled point of designated length and packing is handled.

The earth station system includes: in one of the embodiments,

Data processing module is communicated to connect with the communication module, for receiving noise data and the institute of the equipment The location information of crusing robot is stated, and the noise data is handled according to preset rules, obtains the equipment operation When time-frequency figure, and the polling path is sent to the crusing robot；

Noise storage and retrieval module, is electrically connected with the data processing module, for receiving and storing the positioning letter Breath, and the convolutional neural networks model that the corresponding training of the equipment is completed is determined according to the location information；

Failure analysis module is electrically connected with the noise storage and retrieval module, for receiving the time-frequency figure, and will be defeated Enter into the determining convolutional neural networks model, judges whether the equipment occurs using the convolutional neural networks model Failure；And

Patrol task planning module is electrically connected with the data processing module, for according to the Substation Design drawing Or high definition satellite image, obtain the location information of multiple equipment in substation, and the equipment detected as needed Location information and carry out Noise Acquisition time, generate the polling path, and be sent to the data processing module.

The noise storage and retrieval module is also used to receive and store the time-frequency figure in one of the embodiments, And retraining and optimization are carried out using the time-frequency figure pair convolutional neural networks model corresponding with the equipment.

The failure analysis module is also used to when determining the device fails in one of the embodiments, raw At for prompting the warning message of user equipment failure.

The earth station system further includes alarm modules in one of the embodiments, the alarm modules and the event Hinder analysis module electrical connection, generates buzzer for receiving the warning message, and according to the warning message.

The present invention provides a kind of substation's automatic detecting method and platforms.Wherein, substation's automatic detecting method includes: Polling path is sent to crusing robot, so that the crusing robot reaches inspection position according to polling path, described in acquisition The noise data of equipment at inspection position carries out corresponding mark to the equipment and the noise data, and obtains current The location information of the crusing robot；The noise data of the equipment and the location information of the crusing robot are received, And the noise data is handled according to preset rules, obtain time-frequency figure when equipment operation；According to the positioning The convolutional neural networks model that the corresponding training of equipment described in validation of information is completed, and the time-frequency figure of the equipment is input to The convolutional neural networks model that training corresponding with the equipment is completed is set using described in convolutional neural networks model judgement It is standby whether to break down.In the present invention, by converting time-frequency figure for noise data, the convolutional Neural of training completion is then utilized Network model analyzes time-frequency figure, realizes automatically extracting for image information feature, utilizes the time-frequency figure and volume of noise signal Fault detection problem is converted problem of image recognition by product neural network model, avoids when carrying out noise signal analysis to people The dependence of work and knowledge expertise, while improving the efficiency and reliability of inspection.

Detailed description of the invention

Fig. 1 is a kind of substation's automatic detecting method flow schematic diagram provided in an embodiment of the present invention；

Fig. 2 is convolutional neural networks structure provided in an embodiment of the present invention intention；

Fig. 3 shows noise data according to the result after the sampling number segmentation of designated length to be provided in an embodiment of the present invention It is intended to；

Fig. 4 is the effect picture of the noise data progress Short Time Fourier Transform after dividing；

Fig. 5 is a kind of electrical structure schematic diagram of substation's automatic detecting platform provided in an embodiment of the present invention.

Specific embodiment

In order to make the foregoing objectives, features and advantages of the present invention clearer and more comprehensible, with reference to the accompanying drawing to the present invention Specific embodiment be described in detail.Many details are explained in the following description in order to fully understand this hair It is bright.But the invention can be embodied in many other ways as described herein, those skilled in the art can be not Similar improvement is done in the case where violating intension of the present invention, therefore the present invention is not limited to the specific embodiments disclosed below.

Referring to Figure 1, the embodiment of the invention provides a kind of substation's automatic detecting methods, comprising:

Step S110 sends polling path to crusing robot, so that the crusing robot is reached according to polling path Inspection position acquires the noise data of the equipment at the inspection position, corresponds to the equipment and the noise data Mark, and obtain the location information of presently described crusing robot；

Step S120 receives the noise data of the equipment and the location information of the crusing robot, and according to pre- If rule handles the noise data, time-frequency figure when equipment operation is obtained；

Step S130 confirms the convolutional neural networks mould that the corresponding training of the equipment is completed according to the location information Type, and the time-frequency figure of the equipment is input to the convolutional neural networks model that training corresponding with the equipment is completed, benefit Judge whether the equipment breaks down with the convolutional neural networks model.

It is appreciated that deep learning method automatic learning characteristic and can carry out categorised decision, realize that fault detection is automatic Change.Refer to Fig. 2, convolutional neural networks (CNN) are one of most popular deep learning models in recent years, are had in image recognition There is significant advantage.Convolutional neural networks include input layer (input layer), convolutional layer (convolutional layer), pond Change layer (pooling layer), full articulamentum (fully-connected layer), output layer (output layer) etc., Middle convolutional layer and pond layer are most important feature extraction steps, can be realized the automatic study and extraction of image information feature. When input is two-dimensional time-frequency figure, exports and preset label for equipment state (such as to represent equipment normal for output 0, and output 1 is represented and set It is standby to break down).In the present embodiment, 2 convolution-pond layer in convolutional neural networks structure is to being used for feature extraction, and one Full articulamentum connects entirely for feature, is followed by softmax classifier, the output of classifier is then used as diagnostic result.It need to refer in particular to Out, distinct device may need to construct the convolutional neural networks model of different structure in actual design.

Therefore, by converting time-frequency figure for noise data in the present invention, the convolutional Neural net of training completion is then utilized Network model analyzes time-frequency figure, realizes automatically extracting for image information feature, utilizes the time-frequency figure and convolution of noise signal Fault detection problem is converted problem of image recognition by neural network model, avoids when carrying out noise signal analysis to artificial And the dependence of knowledge expertise, while improving the efficiency and reliability of inspection.In addition, being mentioned using contactless noise transducer The safety of data acquisition has been risen, the influence to equipment is reduced.

In addition, carrying out corresponding mark to the equipment and the noise data, noise data is needed according to acquisition position Mark is carried out with the time, to select corresponding convolutional neural networks model to be detected, and facilitates historical data archiving and inspection Rope.

It is described in one of the embodiments, that the noise data is handled according to preset rules, obtain described set Time-frequency figure when received shipment row, comprising:

Short Time Fourier Transform is carried out to the noise data, obtain and stores time-frequency figure when running with the equipment.

It is appreciated that the basic thought of Short Time Fourier Transform is: signal being divided into many small time intervals, then uses Fu In leaf transformation analysis it is each when compartment, to determine the frequency existing for each time interval, these frequencies it is overall with regard to table Show how frequency spectrum changes in time.In the present embodiment, by Short Time Fourier Transform, frequency is converted by noise data The X-Y scheme that rate changes over time reduces fault detection difficulty to convert problem of image recognition for data analysis problems, It increases economic efficiency.

In one of the embodiments, in the convolution for confirming the corresponding training completion of the equipment according to the location information Before neural network model, further includes:

The history run noise data under the equipment normal condition and under malfunction is acquired, and respectively to each described The operating status of the corresponding equipment of noise data is demarcated, and noise data sample is formed；

The noise data sample is subjected to Short Time Fourier Transform, obtains time-frequency corresponding with the noise data sample Figure；

The time-frequency figure is inputted pre-set convolutional neural networks as image training sample to be trained, to obtain The convolutional neural networks model that the training is completed, and store.

It is appreciated that the normal condition of each equipment of acquisition and the noise data under malfunction, and it is respectively formed time-frequency Convolutional neural networks are trained and are optimized in advance using the time-frequency figure by figure, the setting ginseng of adjustable convolutional neural networks Number, so that trained convolutional neural networks model may be implemented fast and accurately to restrain in the time-frequency figure for receiving input, Guarantee that exporting result has high accuracy simultaneously.

Further, since collected noise data includes ambient noise, crusing robot is sent to by noise data Before earth station system, it is also necessary to carry out preparatory noise reduction process appropriate.Also, noise data is needed according to acquisition position and Time carries out mark, to select corresponding convolutional neural networks to be detected, and facilitates historical data archiving and retrieval.

In the present embodiment, Fig. 3 and Fig. 4 are referred to, wherein Fig. 3 is the noise data figure of exemplary each collection point, and Fig. 4 is Noise data is carried out to the effect picture of Short Time Fourier Transform.By convolutional neural networks module be carried to automatic detecting platform it Before, it is with noise signal when failure and corresponding to each noise data sample setting manually to acquire each key equipment normal operation Label, including equipment identity and working condition (such as: normal condition 0, malfunction 1) are used for successive depths convolution The learning training of neural network.Then, the sampling number that the noise data sample of each equipment presses designated length respectively is divided (such as 1024 sampled points are a segment), and carry out Short Time Fourier Transform and obtain corresponding time-frequency figure as image training Sample is sent into pre-set convolutional neural networks and is trained, excellent to carry out parameter to preset convolutional neural networks model Change, obtain suitable model parameter, so that building has the model of fault identification ability.

In one of the embodiments, before sending the polling path to the crusing robot, the substation Automatic detecting method further include:

According to the design drawing of the substation or high definition satellite image, each equipment for needing to detect in substation is obtained Location information；

The location information of the equipment detected as needed and the time for carrying out Noise Acquisition, generate the inspection road Diameter.

In the present embodiment, according to Substation Design drawing or high definition satellite image, substation's critical electrical equipment position is obtained Confidence is ceased and is labeled, and generates polling path by earth station system, polling path is indicated in the form of coordinate points, is concurrently set In the signal acquisition time of each key equipment location point, generates the location information for including the equipment for needing inspection and noise is believed Number acquisition time, so that crusing robot is acquired within the defined Noise Acquisition time and set accordingly according to the patrol task Standby noise.

Substation's automatic detecting method in one of the embodiments, further include:

Step S140 generates the warning message for prompting user equipment failure when determining the device fails.

In the present embodiment, by generating the alarm signal for prompting user equipment failure when determining device fails Breath, can remind in time user to check and repair equipment, and reduction failure as far as possible influences.

Substation's automatic detecting method in one of the embodiments, further include:

Retraining and optimization are carried out using the time-frequency figure pair convolutional neural networks model corresponding with the equipment.

It is appreciated that since the quantity of the time-frequency figure of the image pattern as training convolutional neural networks is limited, because This is possible to the optimized parameter that convolutional neural networks model is unable to get according to limited sample number, therefore subsequent using work In collected noise data, further parameter optimization is carried out to the convolutional neural networks model, to improve convolutional Neural net The convergence rate and accuracy of network model, and then improve the efficiency and reliability of inspection.

Based on the same inventive concept, Fig. 5 is referred to, it is flat that the embodiment of the invention also provides a kind of substation's automatic detectings Platform, including crusing robot 510 and earth station system 520.

The crusing robot 510 is used to reach inspection position according to polling path, acquires setting at the inspection position Standby noise data, and the equipment and the noise data are identified, and obtain itself current location information.

The earth station system 520 is communicated to connect with the crusing robot 510, for receiving the noise number of the equipment Accordingly and the location information of the crusing robot 510, the noise data is handled according to preset rules, is obtained described Time-frequency figure when equipment is run, and the convolutional Neural net that the corresponding training of the equipment is completed is confirmed according to the location information Network model, the convolutional neural networks model that the time-frequency figure input training corresponding with the equipment of the equipment is completed, utilizes The convolutional neural networks model judges whether the equipment breaks down.

In the present embodiment, by converting time-frequency figure for noise data, the convolutional neural networks of training completion are then utilized Model analyzes time-frequency figure, realizes automatically extracting for image information feature, and then whether judge equipment according to characteristics of image There are failures.As it can be seen that fault detection problem is converted image by the time-frequency figure and convolutional neural networks model using noise signal Identification problem not only avoids when carrying out noise signal analysis to artificial and knowledge expertise dependence, while improving and patrolling The efficiency and reliability of inspection.

The crusing robot 510 includes motion-control module 511, Noise Acquisition module in one of the embodiments, 512, noise processed module 513 and communication module 514.

The motion-control module 511 is used to drive the crusing robot 510 to reach according to the polling path described Inspection position, and the equipment and the noise data are identified, and to obtain the crusing robot 510 current Location information.

The Noise Acquisition module 512 is used to acquire the noise data of the equipment at the inspection position.

The noise processed module 513 is communicated to connect with the noise processed module 513, for by the noise data with The equipment carries out corresponding mark.

The communication module 514 and the Noise Acquisition module 512, the noise processed module 513 and the ground System of standing 520 communicates to connect respectively, and the motion-control module 511511 is from the earth station by the communication module 514 The polling path is obtained in system 520, the noise processed module 513 is to the earth station by the communication module 514 System 520 sends the noise data.

It is appreciated that can include motion-control module 511511, Noise Acquisition module by the crusing robot 510 512, noise processed module 513 is integrated on a single die, to reduce the volume and type of crusing robot 510.The communication Module includes serial line interface, is connected between the crusing robot and earth station system by serial ports.In addition, each module can also be single Solely setting, wherein Noise Acquisition includes sound pick-up outfit, and sound pick-up outfit signal acquisition time according to specified in polling path adopts Collect the noise that relative device generates.The noise processed module 513 then carries out collected noise data with corresponding equipment Mark, so that bottom surface station system can judge its corresponding equipment according to the mark, and then converts time-frequency figure for the noise data It is input in corresponding convolutional neural networks model, judges the current state of equipment using convolutional neural networks module.Specifically set In meter, the noise processed module 513 includes filter circuit, such as LC passive filter circuit, by filter circuit to noise data It is filtered, eliminates the ambient noise in the noise data.

In addition, carrying out corresponding mark to the equipment and the noise data, noise data is needed according to acquisition position Mark is carried out with the time, to select corresponding convolutional neural networks model to be detected, and facilitates historical data archiving and inspection Rope.

In the present embodiment, to feature extraction is used for, one connects 2 convolution-pond layer in convolutional neural networks structure entirely It connects layer to connect entirely for feature, is followed by softmax classifier, the output of classifier is then used as diagnostic result, such as output 0 to represent Equipment is normal, and output 1 represents equipment failure.In particular, distinct device may need to construct in actual design The convolutional neural networks model of different structure.

The noise processed module 513 in one of the embodiments, is also used to the noise data being sent to institute Before stating earth station system 520, the noise data is split according to the sampled point of designated length and packing is handled.

It is appreciated that by by the noise data be split according to the sampled point of designated length and packing handle, so Earth station system 520 is sent to by communication module 514 again afterwards, so that earth station system 520 can directly make an uproar according to what is received Sound data carry out Short Time Fourier Transform.In addition, collected noise data includes ambient noise, need to carry out appropriate pre- First noise reduction process.Also, noise data is needed to carry out mark according to acquisition position and time, to select corresponding convolution mind It is detected through network, and facilitates historical data archiving and retrieval.

The earth station system 520 includes data processing module 521, noise storage and inspection in one of the embodiments, Rope module 522, failure analysis module 523 and patrol task planning module 524.

The data processing module 521 is communicated to connect with the communication module 514, for receiving the noise number of the equipment Accordingly and the location information of the crusing robot 510, and the noise data is handled according to preset rules, obtains institute Time-frequency figure when equipment operation is stated, and the polling path is sent to the crusing robot 510.

The noise storage and retrieval module 522 is electrically connected with the data processing module 521, for receiving and storing Location information is stated, and the convolutional neural networks model that the corresponding training of the equipment is completed is determined according to the location information.

The failure analysis module 523 is electrically connected with the noise storage and retrieval module 522, for receiving the time-frequency Figure, and will enter into the determining convolutional neural networks model, it is set using described in convolutional neural networks model judgement It is standby whether to break down.

The patrol task planning module 524 is electrically connected with the data processing module 521, for according to the substation Design drawing or high definition satellite image, obtain the location information of multiple equipment in substation, and detect as needed The location information of the equipment and the time for carrying out Noise Acquisition, the polling path is generated, and be sent to the data processing Module 521.

The noise storage and retrieval module 522 is also used to receive and store the time-frequency in one of the embodiments, Figure, and retraining and optimization are carried out using the time-frequency figure pair convolutional neural networks model corresponding with the equipment.

The failure analysis module 523 is also used to when determining the device fails in one of the embodiments, Generate the warning message for prompting user equipment failure.

The earth station system 520 further includes alarm modules 525 in one of the embodiments, the alarm modules 525 It is electrically connected with the failure analysis module 523, generates buzzing for receiving the warning message, and according to the warning message Sound.It is appreciated that staff possibly can not see the fault cues signal on monitoring interface, this reality in time for some reason It applies example and generates buzzer when receiving the warning message by the alarm modules 525, remind staff currently to set in time It is standby to be in malfunction, troubleshooting processing is carried out to equipment in time convenient for staff.

Each technical characteristic of embodiment described above can be combined arbitrarily, for simplicity of description, not to above-mentioned reality It applies all possible combination of each technical characteristic in example to be all described, as long as however, the combination of these technical characteristics is not deposited In contradiction, all should be considered as described in this specification.

The embodiments described above only express several embodiments of the present invention, and the description thereof is more specific and detailed, but simultaneously It cannot therefore be construed as limiting the scope of the patent.It should be pointed out that coming for those of ordinary skill in the art It says, without departing from the inventive concept of the premise, various modifications and improvements can be made, these belong to protection of the invention Range.Therefore, the scope of protection of the patent of the invention shall be subject to the appended claims.