Intelligent station operation optimization control system based on big data analysis

Abstract

The invention relates to the technical field of new energy station monitoring, in particular to an intelligent station operation optimization control system based on big data analysis, which comprises an anomaly detection unit, an image recognition unit and a monitoring large screen unit, and further comprises a video monitoring unit, a data processing unit and a data processing unit, the video monitoring unit is used for carrying out video monitoring on a photovoltaic area and booster station equipment in a photovoltaic energy station to obtain a station monitoring video, and transmitting the station monitoring video to the anomaly detection unit. Through the combination of the video monitoring unit and the abnormity detection unit, the system can monitor the operation state of the photovoltaic station in real time and rapidly detect abnormal conditions, so that the safety and reliability of the station are effectively guaranteed, and through the station inspection unit, the inspection robot is utilized to carry out comprehensive inspection, so that the inspection efficiency is improved. Meter data and equipment state information can be efficiently obtained, the operation parameters of equipment can be rapidly and accurately extracted and analyzed in combination with the image recognition technology, the inspection efficiency is improved, and the labor cost is reduced.

Description

Technical Field

[0001] The present invention relates to the field of new energy station monitoring technology, and in particular to a smart station operation optimization control system based on big data analysis. Background Art

[0002] In order to solve the current prominent problems of repeated allocation of operation and maintenance resources, unclear division of labor, and low work efficiency in new energy stations, enterprises must develop in the direction of specialization, informatization, and intelligence. By building a smart remote control center to realize the new mode of "unmanned and less-manned" new energy stations, it is a powerful means for enterprises to enhance their competitiveness, and it is also the only way for enterprises to innovate their production management models. It is also an inevitable trend for the future development of my country's wind power and photovoltaic station operation industry. Summary of the invention

[0003] The technical problem to be solved by the present invention is to overcome the shortcomings of the above-mentioned prior art and provide a smart station operation optimization control system based on big data analysis.

[0004] The technical solution adopted to solve the above technical problems is: a smart station operation optimization control system based on big data analysis, including an abnormality detection unit, an image recognition unit and a monitoring large screen unit, and also includes: A video monitoring unit, which is used to perform video monitoring on the photovoltaic area and booster station equipment in the photovoltaic energy station to obtain a station monitoring video, and transmit the station monitoring video to the abnormality detection unit; A station inspection unit, the station inspection unit is used to conduct a full-field inspection of the photovoltaic energy station based on an inspection robot to obtain inspection data of the photovoltaic energy station, wherein the inspection data includes meter images and infrared images of booster station equipment, and transmit the inspection data of the photovoltaic energy station to an image recognition unit; A security management unit, which is used to manage vehicles entering and exiting the photovoltaic energy station based on intelligent access control to obtain vehicle entry and exit records, and transmit the vehicle entry and exit records to a monitoring large screen unit; An equipment management unit is used to predict the life of a sensor group based on maintenance records to obtain the remaining life of the sensor group, and obtain the current operating status of the sensor group based on the remaining life of the sensor group, wherein the sensor group includes various cameras in the photovoltaic energy station, and the current operating status of the sensor group is transmitted to the monitoring large screen unit.

[0005] Preferably, the anomaly detection unit receives the station monitoring video transmitted by the video monitoring unit, and extracts the latest station monitoring video frame sequence in the station monitoring video, performs anomaly detection on the latest station monitoring video frame sequence based on the trained anomaly detection model to obtain the anomaly label corresponding to the latest station monitoring video frame sequence, that is, obtains the anomaly label corresponding to the station monitoring video, and performs flame target detection on the latest station monitoring video frame sequence based on the trained target detection model to obtain the flame target frame corresponding to the latest station monitoring video frame sequence, performs image segmentation on the flame target frame to obtain a target flame image, obtains the flame level corresponding to the latest station monitoring video frame sequence based on the target flame image, that is, obtains the flame level corresponding to the station monitoring video, and transmits the anomaly label and flame level corresponding to the station monitoring video to the monitoring large screen unit.

[0006] Preferably, the image recognition unit receives the inspection data of the photovoltaic energy station transmitted by the station inspection unit, and performs image recognition on the meter image based on a trained image recognition model to obtain a meter count value corresponding to the meter image, performs temperature recognition on the infrared image of the booster station equipment to obtain a temperature value corresponding to the infrared image of the booster station equipment, and transmits the meter count value corresponding to the meter image and the temperature value corresponding to the infrared image of the booster station equipment to the monitoring large screen unit.

[0007] Preferably, the monitoring large screen unit receives the vehicle entry and exit records transmitted by the security management unit, the current operating status of the sensor group transmitted by the equipment management unit, the abnormal labels and flame levels corresponding to the station monitoring video transmitted by the abnormal detection unit, the meter count values ​​corresponding to the meter image transmitted by the image recognition unit, and the temperature value corresponding to the infrared image of the booster station equipment, and adds the vehicle entry and exit records, the current operating status of the sensor group, the abnormal labels and flame levels corresponding to the station monitoring video, the meter count values ​​corresponding to the meter image, and the temperature value corresponding to the infrared image of the booster station equipment to the corresponding positions in the electronic map of the photovoltaic energy station to obtain a dynamic map of the photovoltaic energy station, and display the dynamic map on the electronic large screen.

[0008] Preferably, the abnormal labels include face abnormal labels, pedestrian abnormal labels, safety helmet abnormal labels, mask abnormal labels, insulating gloves abnormal labels, work clothes abnormal labels, short sleeve abnormal labels, action abnormal labels, personnel stay abnormal labels, personnel intrusion abnormal labels, personnel fall abnormal labels, limb crossing abnormal labels, climbing abnormal labels, gathering abnormal labels and leaving the post abnormal labels.

[0009] Preferably, predicting the life of the sensor group based on the maintenance record to obtain the remaining life of the sensor group includes: The degradation of the sensor group is defined to obtain a degradation definition formula of the sensor group, wherein the degradation definition formula of the sensor group is as follows: ; in, Indicates that the sensor group is The degree of degradation of the moment, Indicates that the sensor group is The natural degradation of time, Indicates that the sensor group is The maintenance degradation degree at the time, and , represents the initial degradation level of the sensor group, Indicates that the sensor group is The drift coefficient at time, The diffusion factor representing the degradation of the sensor group, represents the Brownian motion increment; A life prediction model of the sensor group is constructed based on the degradation definition of the sensor group, wherein the life prediction model of the sensor group is as follows: ; in, represents the average value of sensor group operational failures, represents the difference in the operational failure threshold of the sensor group, Indicates the time when the sensor group fails to operate. Indicates that the sensor group uses a failure threshold. represents the probability density function, represents the conversion parameter, represents the maintenance rate parameter of the sensor group, Indicates the remaining life of the sensor group.

[0010] Preferably, predicting the life of the sensor group based on the maintenance record to obtain the remaining life of the sensor group further includes: Solving the parameters in the life prediction model of the sensor group to obtain the probability density function, conversion parameter and maintenance rate parameter of the sensor group, wherein: The solution formula of the maintenance rate parameter is as follows: ; in, Indicates the sensor group The total time that the components have been running, Indicates the sensor group The total number of repairs of each component during the operating time, Indicates the total number of components in the sensor group; The solution formula of the probability density function is as follows: ; in, Indicates the sensor group Part No. The influence coefficient of the maintenance on the sensor group; The solution formula of the conversion parameter is as follows: ; The life prediction model of the sensor group is solved based on the probability density function of the sensor group, the conversion parameter and the maintenance rate parameter of the sensor group to obtain the remaining life of the sensor group.

[0011] Preferably, obtaining the flame level corresponding to the latest station monitoring video frame sequence based on the target flame image, that is, obtaining the flame level corresponding to the station monitoring video, includes: Calculate the centroid coordinates of the target flame image, wherein the calculation formula of the centroid coordinates of the target flame image is as follows: ; in, represents the centroid coordinates of the target flame image, , , and Indicates the area range in the target row and column directions. Indicates that the pixel coordinates in the target flame image are The pixel value of The center point of the target flame image is used as the reference point of the target flame image, and the distance between the reference point and the centroid is calculated. The calculation formula of the distance is as follows: ; in, represents the distance between the reference point and the centroid, Indicates the coordinates of the center point of the target flame image; The stability coefficient of the latest site monitoring video frame sequence is calculated based on the distance between the reference point and the centroid and the average distance of a preset number of target flame images in the latest site monitoring video frame sequence, wherein the calculation formula of the stability coefficient is as follows: ; in, Represents the stability coefficient of the latest station monitoring video frame sequence, Indicates the average distance of a preset number of target flame images in the latest station monitoring video frame sequence; The stability coefficient of the latest station monitoring video frame sequence is matched with a preset flame level table to obtain the flame level corresponding to the latest station monitoring video frame sequence, that is, to obtain the flame level corresponding to the station monitoring video.

[0012] Preferably, the anomaly detection model includes a feature extraction module, a feature enhancement module, a region of interest extraction module, a region of interest alignment module and a category classification module. The feature extraction module adopts a ResNet model and is used to extract image features of the latest station monitoring video frame sequence. The feature enhancement module adopts an improved cascade feature pyramid network and is used to enhance the image features. The region of interest extraction module adopts a convolutional neural network and is used to generate candidate regions of the latest station monitoring video frame sequence. The region of interest alignment module adopts a RoIAlign model and is used to fix the dimension of the image features after feature enhancement. The category classification module adopts a fully connected layer and is used to classify the image features after dimension fixation and feature enhancement.

[0013] Preferably, the feature enhancement module improves the cascade feature pyramid network through a feedback mechanism, and feeds back the fused feature map of each layer to the bottom-up backbone network of the same layer through the feedback connection module through the feedback mechanism, and fuses them into the feature map of the final output, and the feature enhancement module convolves the image features of the latest station monitoring video frame sequence through the backbone network to obtain feature maps C1-C6, performs dimensionality reduction operation on the feature maps C2-C6 through 1×1 convolution in an upsampling manner to obtain feature maps D2-D6, uses the feature map D6 as the feature map M6, and splices the feature map M6 with D5 to obtain the feature map M5, and repeats the above operation to obtain the feature map M2-M4, and eliminate the aliasing effect of the feature map M2-M6 through 3×3 convolution to obtain the feature map P2-P6, and feed back the feature map P2-P6 to the backbone network of the same layer through the feedback connection module, extract the fused feature map through the backbone network, and fuse and output the results of the two feature extractions through the fusion module, the feedback connection module includes 4 parallel branches, the first 3 parallel branches include a convolution layer and a Relu layer, the convolution kernel sizes are 1, 3 and 3 respectively, the void rates are 1, 3 and 6 respectively, and the padding is 0, 3 and 6 respectively, and the last parallel branch includes a global average pooling layer, a 1×1 convolution layer and a ReLu layer.

[0014] The beneficial effects of the present invention are as follows: (1) Through the combination of the video monitoring unit and the abnormality detection unit, the system can monitor the operating status of the photovoltaic station in real time and quickly detect abnormal situations, thereby effectively ensuring the safety and reliability of the station. By using the inspection robot to conduct comprehensive inspections through the station inspection unit, meter data and equipment status information can be efficiently obtained. Combined with image recognition technology, the operating parameters of the equipment can be quickly and accurately extracted and analyzed, thereby improving inspection efficiency and reducing labor costs. (2) Through the integration of the security management unit and the equipment management unit, the system can comprehensively manage the safety of the station and the operating status of the equipment. This information is visualized in the monitoring large-screen unit, providing an intuitive and dynamic The diagram is convenient for managers to quickly understand the overall operating status of the station, and can summarize and analyze data from different modules (such as abnormal labels, flame levels, meter counts and temperature values, etc.), providing managers with a scientific basis for decision-making, helping to optimize operation control and maintenance strategies, and extending the service life of equipment; (3) Through flame target detection and image segmentation, the system of the present invention can quickly identify and respond to potential fire risks, thereby greatly improving the safety of the station. In addition, combined with the management of vehicle entry and exit records, the overall security measures of the station can be enhanced. By predicting the remaining life of the sensor group, the system can achieve preventive maintenance, reduce the failure rate, and improve the operating efficiency of the system, thereby improving the economic benefits of the photovoltaic station. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 A schematic diagram of the system architecture of an overall system in an embodiment of the present invention.

[0016] Figure numerals: 1. Video surveillance unit; 2. Abnormal detection unit; 3. Station inspection unit; 4. Image recognition unit; 5. Security management unit; 6. Monitoring large screen unit; 7. Equipment management unit. DETAILED DESCRIPTION

[0017] Embodiment 1, as Figure 1 As shown, the present invention proposes a smart station operation optimization control system based on big data analysis, including an abnormality detection unit 2, an image recognition unit 4 and a monitoring large screen unit 6, and also includes: The video monitoring unit 1 is used to perform video monitoring on the photovoltaic area and booster station equipment in the photovoltaic energy station to obtain the station monitoring video, and transmit the station monitoring video to the abnormality detection unit 2; The station inspection unit 3 is used to conduct a full-field inspection of the photovoltaic energy station based on the inspection robot to obtain the inspection data of the photovoltaic energy station, wherein the inspection data includes the meter image and the infrared image of the booster station equipment, and transmits the inspection data of the photovoltaic energy station to the image recognition unit 4; The security management unit 5 is used to manage vehicles entering and exiting the photovoltaic energy station based on the intelligent access control system, so as to obtain vehicle entry and exit records, and transmit the vehicle entry and exit records to the monitoring large screen unit 6; The equipment management unit 7 is used to predict the life of the sensor group based on the maintenance record to obtain the remaining life of the sensor group, and obtain the current operating status of the sensor group based on the remaining life of the sensor group, wherein the sensor group includes various cameras in the photovoltaic energy station, and transmit the current operating status of the sensor group to the monitoring large screen unit 6.

[0018] In the present invention, photovoltaic area refers to the area where photovoltaic modules are installed, which are used to convert solar energy into electrical energy; booster station refers to the facility responsible for boosting the low-voltage electrical energy generated by photovoltaic power generation to high-voltage electrical energy for transmission to the power grid; monitoring video refers to real-time video information collected by a camera; inspection robot refers to an automated equipment used to regularly check and monitor the status of equipment in the station; meter image refers to the reading image of photovoltaic equipment or other related equipment; infrared image refers to the image captured by an infrared camera, which is usually used to detect the thermal state of the equipment to identify overheating or failure; smart access control refers to the use of technical means (such as card swiping, fingerprint, license plate recognition, etc.) to control and record vehicles entering and leaving the station; maintenance record refers to data recording equipment maintenance, maintenance and fault information; remaining life refers to the predicted time that the sensor can still work normally under the current usage status.

[0019] In an optional embodiment, the anomaly detection unit 2 receives the station monitoring video transmitted by the video monitoring unit 1, and extracts the latest station monitoring video frame sequence in the station monitoring video, performs anomaly detection on the latest station monitoring video frame sequence based on the trained anomaly detection model to obtain the anomaly label corresponding to the latest station monitoring video frame sequence, that is, obtains the anomaly label corresponding to the station monitoring video, and performs flame target detection on the latest station monitoring video frame sequence based on the trained target detection model to obtain the flame target frame corresponding to the latest station monitoring video frame sequence, performs image segmentation on the flame target frame to obtain a target flame image, obtains the flame level corresponding to the latest station monitoring video frame sequence based on the target flame image, that is, obtains the flame level corresponding to the station monitoring video, and transmits the anomaly label and flame level corresponding to the station monitoring video to the monitoring large screen unit 6.

[0020] It should be noted that flame target detection refers to the process of detecting and identifying flames in video frames; the flame target frame refers to the detected flame position marked with a rectangular frame, usually composed of four coordinates (upper left corner and lower right corner); the target flame image refers to the image that only contains the flame area after image segmentation processing, removing the background information.

[0021] In an optional embodiment, the image recognition unit 4 receives the inspection data of the photovoltaic energy station transmitted by the station inspection unit 3, and performs image recognition on the meter image based on the trained image recognition model to obtain the meter count value corresponding to the meter image, and performs temperature recognition on the infrared image of the booster station equipment to obtain the temperature value corresponding to the infrared image of the booster station equipment, and transmits the meter count value corresponding to the meter image and the temperature value corresponding to the infrared image of the booster station equipment to the monitoring large screen unit 6.

[0022] It should be noted that the image recognition model includes but is not limited to a convolutional neural network, a residual network, a regional convolutional neural network and YOLO; the infrared radiation intensity is converted into a temperature value using Planck's law and Stefan-Boltzmann's law.

[0023] In an optional embodiment, the monitoring large screen unit 6 receives the vehicle entry and exit records transmitted by the security management unit 5, the current operating status of the sensor group transmitted by the equipment management unit 7, the abnormal labels and flame levels corresponding to the station monitoring videos transmitted by the abnormality detection unit 2, the meter count values ​​corresponding to the meter images transmitted by the image recognition unit 4, and the temperature values ​​corresponding to the infrared images of the substation equipment, and adds the vehicle entry and exit records, the current operating status of the sensor group, the abnormal labels and flame levels corresponding to the station monitoring videos, the meter count values ​​corresponding to the meter images, and the temperature values ​​corresponding to the infrared images of the substation equipment to the corresponding positions in the electronic map of the photovoltaic energy station to obtain a dynamic map of the photovoltaic energy station, and display the dynamic map on the electronic large screen.

[0024] Embodiment 2: A smart station operation optimization control system based on big data analysis proposed in the present invention, compared with embodiment 1, this embodiment also includes: abnormal labels including face abnormal labels, pedestrian abnormal labels, safety helmet abnormal labels, mask abnormal labels, insulating gloves abnormal labels, work clothes abnormal labels, short sleeve abnormal labels, action abnormal labels, personnel stay abnormal labels, personnel intrusion abnormal labels, personnel fall abnormal labels, limb crossing abnormal labels, climbing abnormal labels, gathering abnormal labels and leaving the post abnormal labels.

[0025] In an optional embodiment, predicting the life of the sensor group based on the maintenance record to obtain the remaining life of the sensor group includes: The degradation of the sensor group is defined to obtain a degradation definition formula of the sensor group, wherein the degradation definition formula of the sensor group is as follows: ; in, Indicates that the sensor group is The degree of degradation of the moment, Indicates that the sensor group is The natural degradation of time, Indicates that the sensor group is The maintenance degradation degree at the time, and , represents the initial degradation level of the sensor group, Indicates that the sensor group is The drift coefficient at time, The diffusion factor representing the degradation of the sensor group, represents the Brownian motion increment; The life prediction model of the sensor group is constructed based on the degradation definition of the sensor group, wherein the life prediction model of the sensor group is as follows: ; in, represents the average value of sensor group operational failures, represents the difference in the operational failure threshold of the sensor group, Indicates the time when the sensor group fails to operate. Indicates that the sensor group uses a failure threshold. represents the probability density function, represents the conversion parameter, represents the maintenance rate parameter of the sensor group, Indicates the remaining life of the sensor group.

[0026] In an optional embodiment, predicting the life of the sensor group based on the maintenance record to obtain the remaining life of the sensor group further includes: The parameters in the life prediction model of the sensor group are solved to obtain the probability density function, conversion parameter and maintenance rate parameter of the sensor group, where: The solution formula for the maintenance rate parameter is as follows: ; in, Indicates the sensor group The total time that the components have been running, Indicates the sensor group The total number of repairs of each component during the operating time, Indicates the total number of components in the sensor group; The solution formula for the probability density function is as follows: ; in, Indicates the sensor group Part No. The influence coefficient of the maintenance on the sensor group; The solution formula for the conversion parameter is as follows: ; The life prediction model of the sensor group is solved based on the probability density function, the conversion parameter and the maintenance rate parameter of the sensor group to obtain the remaining life of the sensor group.

[0027] In an optional embodiment, the flame level corresponding to the latest station monitoring video frame sequence is obtained based on the target flame image, that is, the flame level corresponding to the station monitoring video is obtained, including: Calculate the centroid coordinates of the target flame image, where the calculation formula of the centroid coordinates of the target flame image is as follows: ; in, represents the centroid coordinates of the target flame image, , , and Indicates the area range in the target row and column directions. Indicates that the pixel coordinates in the target flame image are The pixel value of The center point of the target flame image is used as the reference point of the target flame image, and the distance between the reference point and the center of mass is calculated. The distance calculation formula is as follows: ; in, represents the distance between the reference point and the centroid, Indicates the coordinates of the center point of the target flame image; The stability coefficient of the latest station monitoring video frame sequence is calculated based on the distance between the reference point and the centroid and the average distance of a preset number of target flame images in the latest station monitoring video frame sequence, where the calculation formula of the stability coefficient is as follows: ; in, Represents the stability coefficient of the latest station monitoring video frame sequence, It represents the average distance of a preset number of target flame images in the latest station monitoring video frame sequence; The stability coefficient of the latest station monitoring video frame sequence is matched with the preset flame level table to obtain the flame level corresponding to the latest station monitoring video frame sequence, that is, to obtain the flame level corresponding to the station monitoring video.

[0028] In an optional embodiment, the anomaly detection model includes a feature extraction module, a feature enhancement module, a region of interest extraction module, a region of interest alignment module and a category classification module. The feature extraction module adopts a ResNet model and is used to extract image features of the latest station monitoring video frame sequence. The feature enhancement module adopts an improved cascade feature pyramid network and is used to enhance image features. The region of interest extraction module adopts a convolutional neural network and is used to generate candidate regions of the latest station monitoring video frame sequence. The region of interest alignment module adopts a RoIAlign model and is used to fix the dimension of the image features after feature enhancement. The category classification module adopts a fully connected layer and is used to classify the image features after dimension fixation and feature enhancement.

[0029] It should be noted that the ResNet model is a deep convolutional neural network architecture designed to solve the degradation problem in deep network training. ResNet introduces skip connections to make information more easily propagated in the network, allowing the model to learn at a deeper level. The core idea is to let the network learn the residual (the difference between input and output) to achieve better training results and performance; the cascaded feature pyramid network is a network architecture for target detection. By constructing a multi-scale feature pyramid, the network can effectively detect targets at different scales. The cascaded feature pyramid network further enhances the feature extraction capability and improves the model's recognition capability and accuracy for complex scenes through the cascade structure. The improved version may include optimization of the original FPN; the fully connected layer is a neural network layer in which each input node is connected to each output node. It is usually used in the last few layers of the network for the final classification or regression task.

[0030] In an optional embodiment, the feature enhancement module improves the cascade feature pyramid network through a feedback mechanism, and feeds back the fused feature map of each layer to the bottom-up backbone network of the same layer through the feedback connection module through the feedback mechanism, and fuses them into the feature map of the final output. The feature enhancement module convolves the image features of the latest station monitoring video frame sequence through the backbone network to obtain feature maps C1-C6, performs dimensionality reduction operation on the feature maps C2-C6 through 1×1 convolution in an upsampling manner to obtain feature maps D2-D6, uses feature map D6 as feature map M6, concatenates feature map M6 with D5 to obtain feature map M5, and repeats the above operation to obtain feature map C1-C6. The feature maps M2-M4 are taken, and the aliasing effect of the feature maps M2-M6 is eliminated through 3×3 convolution to obtain the feature maps P2-P6. The feature maps P2-P6 are fed back to the backbone network of the same layer through the feedback connection module. The fused feature maps are extracted through the backbone network, and the results of the two feature extractions are fused and output through the fusion module. The feedback connection module includes 4 parallel branches. The first 3 parallel branches include a convolution layer and a Relu layer. The convolution kernel sizes are 1, 3 and 3 respectively, the void rates are 1, 3 and 6 respectively, and the padding is 0, 3 and 6 respectively. The last parallel branch includes a global average pooling layer, a 1×1 convolution layer and a ReLu layer.

[0031] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited thereto, and various changes can be made within the knowledge scope of technicians in the relevant technical field without departing from the purpose of the present invention.

Claims

1. A smart station operation optimization control system based on big data analysis, comprising an abnormality detection unit (2), an image recognition unit (4) and a monitoring large screen unit (6), characterized in that: A video monitoring unit (1), the video monitoring unit (1) being used to perform video monitoring on the photovoltaic area and booster station equipment in the photovoltaic energy station to obtain a station monitoring video, and transmit the station monitoring video to the abnormality detection unit (2); A station inspection unit (3), the station inspection unit (3) being used to perform a full-site inspection of the photovoltaic energy station based on an inspection robot to obtain inspection data of the photovoltaic energy station, wherein the inspection data includes meter images and infrared images of booster station equipment, and the inspection data of the photovoltaic energy station is transmitted to an image recognition unit (4); A security management unit (5), the security management unit (5) is used to manage vehicles entering and exiting the photovoltaic energy station based on intelligent access control to obtain vehicle entry and exit records, and transmit the vehicle entry and exit records to a monitoring large screen unit (6); A device management unit (7), the device management unit (7) is used to predict the life of the sensor group based on the maintenance record to obtain the remaining life of the sensor group, and obtain the current operating status of the sensor group based on the remaining life of the sensor group, wherein the sensor group includes various cameras in the photovoltaic energy station, and transmits the current operating status of the sensor group to the monitoring large screen unit (6).

2. According to claim 1, a smart station operation optimization control system based on big data analysis is characterized in that: The anomaly detection unit (2) receives the station monitoring video transmitted by the video monitoring unit (1), extracts the latest station monitoring video frame sequence in the station monitoring video, performs anomaly detection on the latest station monitoring video frame sequence based on the trained anomaly detection model to obtain an anomaly label corresponding to the latest station monitoring video frame sequence, that is, obtains the anomaly label corresponding to the station monitoring video, and performs flame target detection on the latest station monitoring video frame sequence based on the trained target detection model to obtain a flame target frame corresponding to the latest station monitoring video frame sequence, performs image segmentation on the flame target frame to obtain a target flame image, obtains the flame level corresponding to the latest station monitoring video frame sequence based on the target flame image, that is, obtains the flame level corresponding to the station monitoring video, and transmits the anomaly label and flame level corresponding to the station monitoring video to the monitoring large screen unit (6).

3. According to claim 2, a smart station operation optimization control system based on big data analysis is characterized in that: The image recognition unit (4) receives the inspection data of the photovoltaic energy station transmitted by the station inspection unit (3), and performs image recognition on the meter image based on the trained image recognition model to obtain the meter count value corresponding to the meter image, performs temperature recognition on the infrared image of the booster station equipment to obtain the temperature value corresponding to the infrared image of the booster station equipment, and transmits the meter count value corresponding to the meter image and the temperature value corresponding to the infrared image of the booster station equipment to the monitoring large screen unit (6).

4. According to claim 3, a smart station operation optimization control system based on big data analysis is characterized in that: The monitoring large screen unit (6) receives the vehicle entry and exit records transmitted by the security management unit (5), the current operating status of the sensor group transmitted by the equipment management unit (7), the abnormal label and flame level corresponding to the station monitoring video transmitted by the abnormal detection unit (2), the meter count value corresponding to the meter image and the temperature value corresponding to the infrared image of the booster station equipment transmitted by the image recognition unit (4), and adds the vehicle entry and exit records, the current operating status of the sensor group, the abnormal label and flame level corresponding to the station monitoring video, the meter count value corresponding to the meter image and the temperature value corresponding to the infrared image of the booster station equipment to the corresponding positions in the electronic map of the photovoltaic energy station to obtain a dynamic map of the photovoltaic energy station, and displays the dynamic map on the electronic large screen.

5. According to claim 2, a smart station operation optimization control system based on big data analysis is characterized in that: The abnormal labels include face abnormal labels, pedestrian abnormal labels, safety helmet abnormal labels, mask abnormal labels, insulating gloves abnormal labels, work clothes abnormal labels, short sleeve abnormal labels, action abnormal labels, personnel stay abnormal labels, personnel intrusion abnormal labels, personnel fall abnormal labels, limb crossing abnormal labels, climbing abnormal labels, gathering abnormal labels and leaving the post abnormal labels.

6. According to the smart station operation optimization control system based on big data analysis according to claim 1, it is characterized in that: Predicting the life of the sensor group based on the maintenance record to obtain the remaining life of the sensor group includes: The degradation of the sensor group is defined to obtain a degradation definition formula of the sensor group, wherein the degradation definition formula of the sensor group is as follows: ； in, Indicates that the sensor group is The degree of degradation of the moment, Indicates that the sensor group is The natural degradation of time, Indicates that the sensor group is The maintenance degradation degree at the time, and , represents the initial degradation level of the sensor group, Indicates that the sensor group is The drift coefficient at time, The diffusion factor representing the degradation of the sensor group, represents the Brownian motion increment; A life prediction model of the sensor group is constructed based on the degradation definition of the sensor group, wherein the life prediction model of the sensor group is as follows: ； in, represents the average value of sensor group operational failures, represents the difference in the operational failure threshold of the sensor group, Indicates the time when the sensor group fails to operate. Indicates that the sensor group uses a failure threshold. represents the probability density function, represents the conversion parameter, represents the maintenance rate parameter of the sensor group, Indicates the remaining life of the sensor group.

7. The intelligent station operation optimization control system based on big data analysis according to claim 6 is characterized in that: Predicting the life of the sensor group based on the maintenance record to obtain the remaining life of the sensor group also includes: Solving the parameters in the life prediction model of the sensor group to obtain the probability density function, conversion parameter and maintenance rate parameter of the sensor group, wherein: The solution formula of the maintenance rate parameter is as follows: ； in, Indicates the sensor group The total time that the components have been running, Indicates the sensor group The total number of repairs of each component during the operating time, Indicates the total number of components in the sensor group; The solution formula of the probability density function is as follows: ； in, Indicates the sensor group Part No. The influence coefficient of the maintenance on the sensor group; The solution formula of the conversion parameter is as follows: ； The life prediction model of the sensor group is solved based on the probability density function of the sensor group, the conversion parameter and the maintenance rate parameter of the sensor group to obtain the remaining life of the sensor group.

8. The intelligent station operation optimization control system based on big data analysis according to claim 2 is characterized in that: Acquiring the flame level corresponding to the latest station monitoring video frame sequence based on the target flame image, that is, obtaining the flame level corresponding to the station monitoring video, includes: Calculate the centroid coordinates of the target flame image, wherein the calculation formula of the centroid coordinates of the target flame image is as follows: ； in, represents the centroid coordinates of the target flame image, , , and Indicates the area range in the target row and column directions. Indicates that the pixel coordinates in the target flame image are The pixel value of The center point of the target flame image is used as the reference point of the target flame image, and the distance between the reference point and the centroid is calculated. The calculation formula of the distance is as follows: ； in, represents the distance between the reference point and the centroid, Indicates the coordinates of the center point of the target flame image; The stability coefficient of the latest site monitoring video frame sequence is calculated based on the distance between the reference point and the centroid and the average distance of a preset number of target flame images in the latest site monitoring video frame sequence, wherein the calculation formula of the stability coefficient is as follows: ； in, Represents the stability coefficient of the latest station monitoring video frame sequence, Indicates the average distance of a preset number of target flame images in the latest station monitoring video frame sequence; The stability coefficient of the latest station monitoring video frame sequence is matched with a preset flame level table to obtain the flame level corresponding to the latest station monitoring video frame sequence, that is, to obtain the flame level corresponding to the station monitoring video.

9. The intelligent station operation optimization control system based on big data analysis according to claim 2 is characterized in that: The anomaly detection model includes a feature extraction module, a feature enhancement module, a region of interest extraction module, a region of interest alignment module and a category classification module. The feature extraction module adopts a ResNet model and is used to extract image features of the latest station monitoring video frame sequence. The feature enhancement module adopts an improved cascade feature pyramid network and is used to enhance the image features. The region of interest extraction module adopts a convolutional neural network and is used to generate candidate regions of the latest station monitoring video frame sequence. The region of interest alignment module adopts a RoIAlign model and is used to fix the dimension of the image features after feature enhancement. The category classification module adopts a fully connected layer and is used to classify the image features after dimension fixation and feature enhancement.

10. The intelligent station operation optimization control system based on big data analysis according to claim 9 is characterized in that: The feature enhancement module improves the cascade feature pyramid network through a feedback mechanism, and feeds back the fused feature map of each layer to the bottom-up backbone network of the same layer through a feedback connection module through the feedback mechanism, and fuses them into the feature map of the final output. The feature enhancement module convolves the image features of the latest station monitoring video frame sequence through the backbone network to obtain feature maps C1-C6, performs dimensionality reduction operation on the feature maps C2-C6 through 1×1 convolution in an upsampling manner to obtain feature maps D2-D6, uses the feature map D6 as the feature map M6, and concatenates the feature map M6 with D5 to obtain the feature map M5, and repeats the above operation to obtain the feature map M2 -M4, and eliminate the aliasing effect of the feature map M2-M6 through 3×3 convolution to obtain the feature map P2-P6, and feed back the feature map P2-P6 to the backbone network of the same layer through the feedback connection module, extract the fused feature map through the backbone network, and fuse and output the results of the two feature extractions through the fusion module, the feedback connection module includes 4 parallel branches, the first 3 parallel branches include a convolution layer and a Relu layer, the convolution kernel sizes are 1, 3 and 3 respectively, the void rates are 1, 3 and 6 respectively, and the padding is 0, 3 and 6 respectively, and the last parallel branch includes a global average pooling layer, a 1×1 convolution layer and a ReLu layer.

Images