AI video smoke and fire identification alarm system for photovoltaic field area

Abstract

The invention relates to the technical field of photovoltaic factory early warning, and discloses an AI video smoke and fire identification alarm system for a photovoltaic field, and the system comprises a video collection module which is used for collecting visible light video and infrared video data of the photovoltaic field; the data transmission module is used for transmitting the video data to the AI comprehensive management platform through an optical fiber ring network; the AI comprehensive management module is used for carrying out smoke and fire identification and intrusion detection on the transmitted video data by utilizing an artificial intelligence algorithm; the alarm module is used for sending alarm information to a management and control center and triggering a corresponding alarm response when the AI comprehensive management module detects smoke and fire or invasion; and the data storage module is used for storing visible light and infrared video data. By multiplexing the original network resources, the investment of newly built hardware facilities is reduced, the interconnection and intercommunication of the whole field area monitoring network are ensured, the overall deployment and operation of the system are optimized, the accident risk caused by the damage of the photovoltaic facilities by animals is reduced, and the normal operation of the photovoltaic field area is effectively ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic plant area early warning, and in particular to an AI video fireworks recognition alarm system for photovoltaic areas. Background Art

[0002] As the global demand for clean energy continues to increase, the scale of photovoltaic power stations is gradually expanding. A large number of photovoltaic modules, electrical equipment and power transmission and distribution systems are arranged in the photovoltaic field. Once a fire occurs, the fire spreads rapidly, which may cause equipment damage, large-scale power outages, and even casualties. At the same time, some animals or people mistakenly enter the photovoltaic plant area and easily damage the photovoltaic facilities in the plant area during their activities. Therefore, timely detection of fire and intrusion information and rapid response are important tasks to ensure the safe operation of photovoltaic power stations.

[0003] In some cities in my country, the photovoltaic power stations under the jurisdiction of photovoltaic power plants occupy a large area, the plots are scattered, the road conditions are complex, and they are located near forest areas. There are many weeds and trees outside the photovoltaic field, and fires are often caused by external factors, which poses a great safety hazard. Once a fire occurs, it cannot be quickly contained, which will cause huge losses to the equipment in operation in the entire field. The original video surveillance was deployed in each box transformer, which was limited by the installation location and had a large blind spot. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides an AI video fireworks recognition alarm system for photovoltaic areas, which solves the problem that there are many weeds and trees outside the photovoltaic area, and fires are often caused by external factors, posing a major safety hazard. Once a fire occurs, it cannot be quickly contained, causing huge losses to the equipment in operation in the entire area.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: an AI video fire and smoke recognition alarm system for photovoltaic fields, comprising:

[0006] Video acquisition module, used to collect visible light video and infrared video data of the photovoltaic field;

[0007] Data transmission module, which transmits video data to the AI ​​integrated management platform through the optical fiber ring network;

[0008] AI integrated management module, which uses artificial intelligence algorithms to perform fireworks recognition and intrusion detection on transmitted video data;

[0009] Alarm module: When the AI ​​integrated management module detects fireworks or intrusion, it sends alarm information to the control center and triggers the corresponding alarm response;

[0010] A data storage module for storing visible light and infrared video data and performing historical data retrieval based on time and event type;

[0011] The system linkage module realizes the linkage control between the AI ​​integrated management platform and the GIS+ platform to perform fire location, camera priority control and electronic fence management;

[0012] The simulation expulsion module works in conjunction with the intrusion detection unit to expel intruders based on the detection results.

[0013] Preferably, the video acquisition module is mainly composed of a visible light camera, an infrared thermal imager, a motion compensation unit, an electronic image stabilization unit and a video ring network unit, which are respectively responsible for collecting different types of video information of the photovoltaic field. The visible light camera performs visual monitoring of the field under sufficient light during the day, while the infrared thermal imager uses thermal imaging technology to monitor at night or under bad weather conditions to ensure all-weather video acquisition. The motion compensation unit detects the motion state of the camera in real time and corrects the image displacement caused by camera movement or environmental vibration. The electronic image stabilization unit further optimizes the stability of the video, reduces the picture jitter caused by external factors, and ensures the clarity of the video data. The video ring network unit adopts a fiber ring network architecture and connects each monitoring camera node in the photovoltaic field through fiber optic Ethernet. The ring network structure can automatically switch to a backup path when any node fails or the optical fiber breaks to ensure continuous data transmission. The combination of the visible light and infrared acquisition systems of this module enables the system to seamlessly monitor the photovoltaic field 24 hours a day. The application of motion compensation and electronic image stabilization technology greatly reduces the image blur or offset caused by wind or equipment movement, improves the accuracy of data acquisition, and ensures the high quality of video images. Especially in scenarios such as fire or intrusion detection, clear images help to detect and respond to incidents in a timely manner.

[0014] Preferably, the data transmission module mainly adopts optical fiber transmission, combined with wireless transmission scheme. The optical fiber transmission unit transmits the high-definition video data collected in the field to the AI ​​integrated management platform in a high-speed and low-latency manner through the optical fiber ring network. For areas where it is inconvenient to lay optical cables, the system uses a wireless transmission unit to transmit remote signals through a wireless bridge powered by the power generated by the photovoltaic plant and the supporting energy storage battery group added by itself. Even if the power supply line is damaged due to a fire in the plant, the backup energy storage power supply can still be used to ensure that the system can continuously monitor the emitted signals. The signal conversion unit is responsible for converting the collected analog video signal into a digital signal for easy transmission and processing. The bandwidth management unit dynamically adjusts the data transmission bandwidth to ensure that a stable transmission rate can be maintained during the peak period of data traffic to avoid network congestion. By combining optical fiber and wireless transmission, the system can flexibly adapt to different field environments, especially remote areas or areas with difficulty in power supply can also be continuously monitored. The bandwidth management function improves the network adaptability of the system, so that a large amount of video data can be quickly and safely transmitted to the AI ​​management platform, effectively reducing delays and packet loss during video transmission.

[0015] Preferably, the AI ​​integrated management module realizes intelligent management of the site through a fire identification unit, an intrusion detection unit, a self-learning unit and a self-diagnosis unit. The fire identification unit uses a deep learning algorithm to identify smoke and flame characteristics by combining multi-source data of visible light and infrared images to determine whether there is a fire. The intrusion detection unit detects personnel and animal targets entering the site based on visual recognition technology. The self-learning unit can continuously optimize the accuracy of fire and intrusion identification through big data training. The self-diagnosis unit monitors the system operation status in real time, can automatically detect faults and provide diagnostic information, and prompt maintenance personnel. The AI ​​integrated management module significantly improves the system's detection efficiency and accuracy of fire and intrusion. Fire identification and intrusion detection are optimized through deep learning algorithms, reducing the problems of false alarms and missed alarms in traditional monitoring systems. At the same time, the self-learning function enables the system to have the ability to continuously improve and adapt to different environments and scene changes. The self-diagnosis function ensures the stable operation of the system, can quickly identify and solve problems, reduces system failure time, and improves operation efficiency.

[0016] Preferably, the simulation expulsion module is linked with the intrusion detection unit, and when a human or animal intrusion is detected, the corresponding expulsion means are activated. When a human intrusion is detected, the language prompt unit plays a pre-recorded language prompt through a speaker to warn and expel the intruder. For animal intrusion, the animal expulsion unit will scare away the invading animals by playing the sound of its natural enemy. When the noise interference unit is used to expel animals, it makes the animals feel uncomfortable by generating high-frequency or low-frequency noise, forcing them to leave. The environmental adaptation unit automatically adjusts the type and intensity of the expulsion means according to different types of intrusion targets and environmental changes to ensure the effectiveness of the expulsion measures. This module can effectively reduce manual intervention and respond immediately when an intrusion is detected through automated means to play an expulsion role. The language prompt function during human intrusion can reduce stray entry and potential dangers, while the animal expulsion function can more effectively expel invading animals and ensure the safety of photovoltaic field equipment by simulating the dual means of natural enemy sound and noise. In addition, the environmental adaptation unit can adjust the expulsion strategy according to real-time monitoring data to improve the flexibility and adaptability of the system.

[0017] Preferably, the alarm module is composed of a fire alarm unit, an intrusion alarm unit, an alarm response unit and an alarm information upload unit. When the fire alarm unit detects a fire, it prompts the fire through sound or signal light, and automatically uploads the fire information to the GIS+ platform. The intrusion alarm unit triggers an alarm when it detects a human or animal intrusion, and links related equipment to respond, such as turning on the lighting and focusing the camera. The alarm response unit takes corresponding measures according to different levels of alarm events to ensure that the emergency can be handled in time. The alarm information upload unit uploads the alarm information to the GIS+ platform for recording and display, so that the operator of the control center can view and analyze it. This module can issue a warning and link other equipment for emergency response as soon as a fire or intrusion event occurs, greatly shortening the accident response time. Through alarm upload, managers can monitor the status of the site in real time and take timely countermeasures to ensure the safety of the site. The refinement of the alarm level can help operators take appropriate response measures according to different threat levels, thereby reducing accident losses.

[0018] Preferably, the system linkage module realizes linkage with the GIS+ platform through the fire location unit, the camera control unit, the electronic fence management unit and the data interaction unit. The fire location unit uses the azimuth and pitch angle information provided by the monitoring system to accurately determine the location of the fire, and uploads the longitude and latitude data to the GIS platform. When a fire or intrusion occurs, the camera control unit automatically adjusts the camera to focus on the hot spot area and tracks the target in real time. The electronic fence management unit monitors the status of the electronic fence around the site, triggers an alarm and links other devices when an intrusion occurs. The data interaction unit is responsible for bidirectional interaction between the data of the monitoring system and the GIS+ platform to ensure real-time synchronization and display of the data. The system linkage module realizes the efficient operation of the monitoring system through deep integration with the GIS+ platform. The fire location and camera control functions improve the response speed to emergencies and can quickly locate and monitor the accident site. At the same time, the electronic fence management strengthens the protection of the site boundary and provides early warning before intrusion. Bidirectional data interaction ensures the efficient linkage between the monitoring platform and the GIS system, so that the real-time and accuracy of the data are guaranteed.

[0019] Preferably, the data storage module is used to store video data, audio data, captured images, and fire and intrusion related alarm data. The design of this module must have sufficient storage capacity and efficient data retrieval capabilities to ensure that the system can store a large amount of unstructured data for a long time and can quickly respond to historical data query needs. The module also supports structured management of different types of data, providing a reliable foundation for subsequent data analysis, historical tracing, and security report generation.

[0020] The present invention provides an AI video fire and smoke recognition alarm system for photovoltaic fields. It has the following beneficial effects:

[0021] 1. The present invention makes full use of the video ring network of the photovoltaic field by adding a panoramic video camera data module, and directly connects the panoramic video data to the AI ​​comprehensive management platform. This not only saves resources and avoids duplicate construction, but also improves the system integration. The system uses a fiber optic ring network and wireless transmission to ensure that data can be transmitted to the AI ​​management platform in a high-speed and stable manner, achieving seamless integration with existing monitoring equipment in the photovoltaic field, and greatly reducing overall construction and maintenance costs. By reusing existing network resources, the present invention reduces the investment in new hardware facilities, while ensuring the interconnection of the entire field monitoring network and optimizing the overall deployment and operation of the system.

[0022] 2. The AI ​​integrated management platform of the present invention has the functions of fire identification and intrusion monitoring, and can make real-time judgments on fire conditions (such as fire points) in the area covered by the camera and issue alarms. At the same time, the platform can also intelligently identify the intrusion behavior of large animals such as humans, cattle and sheep, and automatically trigger the corresponding alarm mechanism. The electronic fence function is conveniently set up to realize the linkage between entry and exit monitoring and alarm to ensure the safety of the site. Through the introduction of deep learning algorithms, the system has greatly improved the accuracy of fire and intrusion detection, and combined with the multi-source fusion of infrared thermal imaging and visible light data, the fire identification has extremely high accuracy in both day and night environments. The intrusion identification and expulsion function of large animals reduces the risk of accidents caused by animals destroying photovoltaic facilities and effectively ensures the normal operation of the photovoltaic field.

[0023] 3. The AI ​​integrated management platform of the present invention and the control center (GIS+) platform are highly linked, so that after detecting a fire or intrusion incident, the system can synchronously transmit the location information and video surveillance data to the GIS+ platform, and accurately display it on the map. Especially for fire situations, the system can quickly determine the longitude and latitude of the fire point based on the camera's viewing angle and the location of the fire source, which is convenient for remote managers to conduct intuitive viewing and emergency response. At the same time, the data interaction function between the electronic fence and the GIS+ platform makes the alarm and display of intrusion incidents more timely and accurate. Through this linkage, the system improves the visualization of fire and intrusion incidents, helps managers make timely and accurate decisions, reduces the risk of accident spread, and significantly shortens the emergency response time, improving the overall site management efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 is a system architecture diagram of the present invention;

[0025] Figure 2 It is a schematic diagram of a video acquisition module of the present invention;

[0026] Figure 3 It is a schematic diagram of a data transmission module of the present invention;

[0027] Figure 4 This is a schematic diagram of the AI ​​comprehensive management module of the present invention;

[0028] Figure 5 It is a schematic diagram of a fire identification unit of the present invention;

[0029] Figure 6 It is a schematic diagram of an intrusion detection unit of the present invention;

[0030] Figure 7 It is a schematic diagram of an alarm module of the present invention;

[0031] Figure 8 It is a schematic diagram of an alarm module of the present invention;

[0032] Fig. 9 It is a schematic diagram of the system linkage module of the present invention;

[0033] Fig.10 It is a schematic diagram of the simulation expulsion module of the present invention;

[0034] Fig.11 It is a system flow chart of the present invention. DETAILED DESCRIPTION

[0035] The following will be combined with the drawings in the specification of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0036] Example:

[0037] Please see attached Figure 1 -Attached Fig.10 The embodiment of the present invention provides an AI video fire and smoke recognition alarm system for photovoltaic fields, comprising:

[0038] The video acquisition module is the basis of system monitoring. Combining visible light and infrared camera technology, it can provide stable monitoring data in all weather conditions. The acquired images are not only used for subsequent fire identification and intrusion detection, but also provide data support for other modules. Therefore, the design of the video acquisition module needs to ensure image clarity, stability of the acquisition equipment, and wide-area monitoring coverage.

[0039] In this embodiment, the specific implementation of the video acquisition module is as follows:

[0040] Visible light acquisition unit: Use a network camera with high-definition resolution (at least 2 million pixels), which has a 40x optical zoom function, can capture distant targets, and has an automatic focus function to ensure image clarity under different lighting conditions. The camera also supports automatic white balance adjustment to adapt to the day and night changes and weather conditions of the site.

[0041] Infrared acquisition unit: uses an uncooled vanadium oxide (VOx) thermal imaging detector with a resolution of 400×300 pixels to generate thermal imaging videos. The detector captures thermal radiation through the infrared spectrum (8-14 microns) and provides a clear image of temperature distribution. The thermal imaging system is also combined with a fixed-focus lens to ensure effective monitoring in dark or harsh weather conditions.

[0042] Motion compensation unit: Uses a set of gyroscopes and accelerometers to detect the camera's motion in real time. When the device is detected to be shaking, the motion compensation algorithm adjusts the camera's internal parameters to ensure that the image remains stable. The algorithm calculates the displacement difference of each frame of the image and generates a compensation matrix to correct the image jitter caused by device vibration.

[0043] Electronic image stabilization unit: The image stabilization function is achieved through image processing algorithms. Electronic image stabilization uses the SURF algorithm to identify static feature points in the video, and by calculating the relative position changes between the feature points, adjusts the alignment position of the image frame in real time, eliminates small jitters, and ensures image stability.

[0044] Video ring network unit: It adopts a fiber ring network architecture and connects each surveillance camera node in the photovoltaic field through fiber Ethernet. The ring network structure can automatically switch to the backup path when any node fails or the fiber breaks to ensure continuous data transmission. At the same time, the video ring network adopts a dual-ring redundant design, and each monitoring node is connected to the ring network through two independent fiber optic lines. The main ring is used for normal video data transmission. When the main ring fails, the system automatically switches to the backup ring network for data transmission to ensure the high availability and reliability of the system. In addition, a fiber optic switch is configured on each camera node in the photovoltaic field, and the switch is configured as a logical ring network topology according to the geographical distribution and monitoring requirements of the field. Each node is both a data receiver and a data forwarder, forming a multi-hop transmission mechanism to ensure that each node can receive panoramic video data.

[0045] The various units of the video acquisition module are closely integrated, and the fusion of infrared and visible light data provides the system with an all-weather monitoring basis, while motion compensation and image stabilization technology further improve the quality and reliability of the data.

[0046] After the video acquisition is completed, the data transmission module transmits the collected data to the AI ​​integrated management platform for processing through the efficient video ring network. At the same time, in order to adapt to the complex environment of the photovoltaic field, the camera data transmission of the panoramic video requires both wired solutions and redundant designs for wireless transmission. Through the combination of optical fiber and wireless bridges, the system ensures the stability and real-time performance of data transmission.

[0047] In this embodiment, the specific implementation of the data transmission module is as follows:

[0048] Optical fiber transmission unit: A fiber ring network is laid in the field, and the Gigabit fiber transmission protocol is used for high-speed transmission of video data. Each camera node is connected to the optical fiber network through an optoelectronic conversion device to ensure real-time transmission of high-definition video data. The network adopts a redundant design to ensure that data can continue to be transmitted through other paths in the event of a break in some optical fiber links.

[0049] Wireless transmission unit: In remote areas where optical fiber cannot be laid, the system uses a wireless bridge based on the Wi-Fi 6 protocol, combined with a solar power supply system, to ensure long-term data transmission. The bridge frequency is above 5GHz and supports multi-channel transmission, ensuring low interference and high data transmission rate.

[0050] Signal conversion unit: Use analog video signal converter to convert analog signals collected by camera equipment into IP signals that comply with digital transmission protocols. The converter has real-time signal conversion function to ensure no delay during signal conversion and supports concurrent processing of multiple signals.

[0051] Bandwidth management unit: A bandwidth management system is deployed in the AI ​​integrated management platform, using the WFQ weighted fair queuing algorithm to dynamically allocate network bandwidth based on video traffic load. During peak video traffic periods, the system prioritizes bandwidth for video data to avoid congestion of other low-priority data.

[0052] The optical fiber and wireless transmission units of the data transmission module back up each other, ensuring that even if the optical fiber link fails, the system can continue to work through wireless transmission. The dynamic adjustment of the bandwidth management unit ensures the real-time nature of data transmission.

[0053] The AI ​​integrated management module is the brain of the system, responsible for receiving data and performing intelligent analysis. Through functions such as fire identification, intrusion detection, self-learning and self-diagnosis, the system can automatically process the collected data and take countermeasures. The design of this module ensures the intelligence and self-optimization capabilities of the system.

[0054] In this embodiment, the specific implementation of the AI ​​comprehensive management module is as follows:

[0055] Fire identification unit: Based on deep learning algorithms, the training model uses standard data sets for fire detection, including smoke and flame features. Multi-layer convolution is used to extract fire precursor features in the image, especially the diffusion speed of smoke and color changes of flames. The system also combines thermal imaging data to provide early warning of fires through abnormal temperature changes;

[0056] The deep learning algorithm formula is as follows:

[0057]

[0058] Where: y(i,j) represents the value of the i-th row and j-th column position of the convolution output;

[0059] x(i+m,j+n) represents the pixel value of the input image, i and j are coordinates, m and n are the biases of the convolution kernel.

[0060] displacement;

[0061] k(m,n) represents the weight of the convolution kernel (filter);

[0062] M and N represent the width and height of the convolution kernel respectively.

[0063] The result after convolution is transformed nonlinearly through the activation function to introduce nonlinear features:

[0064] z(i,j)=max(0,y(i,j))

[0065] Case 1: Normal fire identification operation

[0066] Assume that the input image x(i,j) contains flame features. When inputting the convolution layer, the convolution kernel k(m,n) is trained to be accurate enough to correctly detect the edge, shape and brightness of the flame. After multiple layers of convolution operations, the network extracts the flame features, and in the final classification layer, the Softmax function is used to determine the fire category (the probability of the "flame" category is the highest).

[0067] The final output is:

[0068] In the convolution result of y(i,j), the values ​​at multiple positions correspond to higher probabilities of flame features;

[0069] The system determines that the video frame contains a fire and triggers a corresponding alarm response.

[0070] Case 2: Problematic fire identification operation

[0071] If the input image contains a lot of noise or reflections (for example, the image is too bright due to sunlight reflection), the convolution kernel k(m,n) may mistakenly mistake these noises for fire features. At this time, the convolution operation generates erroneous high-value features, that is, the values ​​of some positions in the convolution result y(i,j) are too high, resulting in false alarms. After being processed by the activation function, the network eventually incorrectly determines that there is a fire in the image and triggers an unnecessary alarm.

[0072] in this case:

[0073] The output of y(i,j) contains false alarm information (non-fire areas are identified as fire);

[0074] The system falsely reported a fire even though there was no fire.

[0075] Intrusion detection unit: Based on the target detection algorithm, this unit can analyze video data in real time and identify potential intrusion targets. By comparing with the preset safety area of ​​the site, it can determine the occurrence of intrusion. Intrusion detection supports both human and animal identification, and combines convolutional neural networks to classify target behaviors and accurately identify illegal intrusions;

[0076] The target detection algorithm expression is:

[0077]

[0078] Where: S 2 is the number of grids into which the image is divided;

[0079] B is the number of bounding boxes predicted for each grid;

[0080] x i ,y i Represents the center coordinates of the predicted bounding box;

[0081] w i ,h i Represents the width and height of the predicted bounding box;

[0082] C i is the confidence level, indicating whether there is a target object in the grid;

[0083] are the true values ​​respectively;

[0084] is an indicator, which is 1 when there is an object in grid i and bounding box j, otherwise it is 0;

[0085] λ coord and λ noobj are the weights of each part in the loss function.

[0086] Case 1: Normal intrusion detection operation

[0087] In the intrusion detection module, assume that a herd of cattle invades the farm. The model divides the input image into grids and predicts the target and its bounding box in each grid through a neural network. Due to the large size of cattle, the model can accurately identify their location and output the corresponding bounding box coordinates (x i ,y i , w i ,h i ) At the same time, the confidence C i A higher value means that there is indeed a target in the grid.

[0088] The results are:

[0089] The bounding box of the target is closer to the true position, and the positioning error in the loss function is smaller;

[0090] With high confidence, the system identified the cattle intrusion and sounded the alarm.

[0091] Case 2: Problematic Intrusion Detection Run

[0092] If there is a lot of background noise in the image, such as the shaking of grass and trees caused by wind, the model may mistake these dynamic features for target intrusion, causing the model to incorrectly output a high-confidence bounding box. In this case, the position of the bounding box does not match the actual target, and the confidence may be too high, resulting in a false positive.

[0093] The results are:

[0094] The target's bounding box deviates greatly from the actual situation, and the system mistakenly identifies it as an intrusion;

[0095] The confidence level is not accurate and may trigger a false alarm even though there is actually no cattle or human intrusion.

[0096] Self-learning unit: Through reinforcement learning algorithms, the system can continuously learn from historical fire and intrusion data and optimize model parameters. This unit analyzes the deviation between the system's recognition results and the actual situation through feedback loops, and dynamically adjusts the model's hyperparameters to improve future recognition accuracy;

[0097] The reinforcement learning algorithm formula is as follows:

[0098]

[0099] Q(s t ,a t ) means that at time step t, the state s t Take action a t The Q value obtained;

[0100] α is the learning rate, which controls the update amplitude;

[0101] r t is the immediate reward after the system performs an action at time step t;

[0102] γ is the discount factor, which indicates the influence of future rewards on current decisions;

[0103] Indicates that in the next state s t+1 In the above example, take the optimal action a ′ The maximum Q value that can be obtained;

[0104] By continuously updating the Q value, the system can make the best decision on the path with the greatest long-term reward.

[0105] Self-learning process

[0106] 1. Assume that in the initial state s0 of the system, a new fire event is discovered and the system selects an identification method a0;

[0107] 2. The system gives an immediate reward r0 for the result of the recognition method. If the recognition result is accurate (e.g., timely alarm and fire confirmation), the reward value r0 is high;

[0108] 3. The model updates the Q value Q(s0,a0) of the state and action;

[0109] 4. In the next state s1, the recognition method a1 selected by the system will be influenced by previous experience and gradually optimized;

[0110] 5. After multiple iterations, the system can gradually learn the best identification and response measures to be taken in different scenarios.

[0111] result:

[0112] The system gradually identifies fire characteristics and selects the optimal detection method, so that the accuracy of fire identification continues to improve.

[0113] Self-diagnosis unit: Through the system SNMP protocol to achieve network device management, this unit can monitor the operating status of each hardware device in the system in real time (CPU, memory, network connection, etc.). When a device is abnormal, the self-diagnosis unit will generate a diagnosis report in time and notify the operation and maintenance personnel.

[0114] The various units of the AI ​​comprehensive management module work closely together to analyze data through intelligent algorithms, while continuously optimizing the recognition model through a self-learning mechanism. The self-diagnosis function ensures that the system responds promptly to abnormal situations, ensuring the continuous and efficient operation of the system.

[0115] The simulated expulsion module is responsible for expelling intruders according to the identification results of the intrusion detection unit. Through sound reminders, noise interference and other methods, the module realizes the automatic expulsion function without human intervention, effectively reducing the potential threats in the photovoltaic field.

[0116] In this embodiment, the specific implementation of the simulation driving module is as follows:

[0117] Language prompt unit: For human intruders, the system plays a preset warning message through the speaker. This unit is linked with the intrusion detection unit. When an illegal intruder is identified, it plays a voice prompt such as "You have entered the restricted area, please leave immediately". The voice file is pre-set through the AI ​​integrated management platform, and the choice of voice is related to the type of intrusion, time and other related parameters.

[0118] Animal Repelling Unit: When an animal invasion is detected, the system plays the sounds of the invading animal's natural enemies to repel them. For bird invasions, the system plays the calls of raptors; for large animals such as cattle and sheep, the system plays the sounds of large carnivores such as wolves. These sounds are dynamically retrieved through the system's built-in sound library and matched according to the type of invading animal.

[0119] Noise Interference Unit: For animals that cannot be driven away by verbal prompts or the sounds of natural enemies, the system uses high-frequency noise interference, using an ultrasonic generator to emit high-frequency noise that is not easy for humans to detect, making the animals feel uncomfortable and leave. The system will adjust the frequency and intensity of the noise according to the distance and type of the invading animal to ensure its driving effect.

[0120] Environmental adaptation unit: This unit detects environmental changes in the photovoltaic field in real time and adjusts the expulsion strategy according to parameters such as weather and time. At night or on rainy days, the volume of the language prompt will be appropriately lowered; for large animal intrusions, the noise frequency and duration of the sound will also be adjusted accordingly.

[0121] The simulation expulsion module realizes the function of automatic expulsion according to the classification of the target and the change of the environment through linkage with the intrusion detection unit. This module reduces manual intervention and improves the expulsion efficiency through appropriate expulsion strategies.

[0122] The alarm module is responsible for sending alarm signals to the control center or management personnel when a fire or intrusion is detected, and executing the corresponding emergency response. The module has a multi-level alarm function and can take different response measures according to different threat levels.

[0123] In this embodiment, the specific implementation of the alarm module is as follows:

[0124] Fire alarm unit: By analyzing the fire data provided by the fire identification unit, when a fire is detected, the unit immediately triggers the alarm and activates the sound and light alarms in the area to issue a warning. At the same time, the fire data is uploaded to the GIS+ platform through the network for real-time viewing by remote monitoring personnel.

[0125] Intrusion alarm unit: When the intrusion detection unit detects an illegal intrusion, the intrusion alarm unit issues a warning and links other devices to turn on the lights and cameras to record the intrusion area. At the same time, the intrusion information is uploaded to the GIS+ platform and sent to the control center for further processing by management personnel.

[0126] Alarm response unit: This unit takes different emergency measures according to the level of the alarm event. For high-level fire alarms, the system will automatically dispatch multiple cameras to aim at the fire source and continue recording. For low-level alarms, the system will only give a simple prompt to reduce resource consumption.

[0127] Alarm information upload unit: All alarm information is uploaded to the remote server in real time through the MQTT protocol to ensure that the alarm information can be received by the monitoring center and management personnel in a timely manner. At the same time, the system will store the alarm data for subsequent analysis and processing.

[0128] The alarm module adopts different alarm strategies according to the feedback from the fire and intrusion detection units and the severity of the incident, and interacts with the GIS+ platform to ensure that managers can obtain and process alarm information in a timely manner.

[0129] The integration of the system linkage module and the GIS+ platform enables the system to globally locate fire and intrusion incidents, and dynamically control equipment such as cameras and electronic fences according to the development of the incidents.

[0130] In this embodiment, the specific implementation of the system linkage module is as follows:

[0131] Fire location unit: By combining the real-time data from the camera and the geographic information system of the GIS platform, the system can accurately locate the specific location of the fire. The system calculates the camera's pitch angle, azimuth angle and monitoring range to obtain the latitude and longitude coordinates of the fire source and displays them on the GIS map.

[0132] Camera control unit: After a fire or intrusion occurs, this unit automatically adjusts the camera's viewing angle and focus to track the target area. For example, when an intrusion occurs, the camera automatically focuses on the intrusion area and continues to track the target's movement trajectory to ensure complete recording.

[0133] Electronic fence management unit: This unit monitors the status of the electronic fence in real time. Once the fence is detected to be damaged or illegally entered, the system will immediately trigger an alarm and link other devices such as cameras and lighting to respond.

[0134] Data interaction unit: The system interacts with the GIS+ platform through standardized protocols to ensure the synchronization of monitoring data and map information. Whenever an event occurs, the data interaction unit will upload video data, location information, and alarm information to the GIS+ platform and archive them.

[0135] The seamless integration of the system linkage module and the GIS+ platform enables fire and intrusion incidents to be accurately located on the global map, while the camera control and electronic fence linkage functions ensure dynamic response to emergencies.

[0136] The data storage module is used to store video data, audio data, captured images, and fire, intrusion and other related alarm data to ensure that the system can store a large amount of unstructured data for a long time and can quickly respond to historical data query needs.

[0137] The video data storage unit is used to store continuous video data collected by visible light cameras and infrared thermal imagers. The system requires a storage time of no less than 6 months to ensure sufficient preservation of historical data.

[0138] Storage method: Video data uses timestamp and camera ID as storage index fields. Data is stored in the form of continuous video streams, and the system uses H.264 or H.265 compression encoding format, which saves storage space and ensures video quality.

[0139] Video data is stored in the form of segmented files, with each video lasting 1 hour. When the system detects a fire or intrusion, a key frame video is generated and stored in association with the alarm information. High-capacity hard disks (such as RAID5 arrays) are used to provide redundant backup to ensure that data can be restored in the event of hard disk damage. Video data is compressed in real time during storage, and old data that exceeds the retention period is regularly archived or deleted to ensure sufficient storage space for new data.

[0140] The system uses the AAC format to compress audio data, and the audio data is stored synchronously with the video data to ensure consistent timestamps for subsequent analysis and retrieval.

[0141] When the system detects an intrusion (such as an animal or human intrusion), it will start audio collection. The audio data will be stored together with the video data, and an audio log will be generated when an alarm is sounded to help operation and maintenance personnel trace the on-site situation.

[0142] When the system detects a fire or intrusion, it automatically captures the current key frame image and stores it in association with the alarm timestamp, camera ID and other information. The system uses JPEG format to store captured images with a resolution of 1920x1080 to ensure image clarity and facilitate identification and backtracking. The captured images are indexed with alarm information, camera location and other data for subsequent retrieval. The system also provides a query interface based on time, camera location and other conditions to facilitate users to find images of related events.

[0143] Each alarm data contains the following contents: alarm time, alarm type (fire, intrusion, etc.), alarm level, camera ID, event description, and associated video and audio data addresses. Alarm data is logically associated with related data such as video, audio, and captured images to ensure that relevant multimedia data can be quickly accessed when viewing alarm information. The system provides flexible alarm data retrieval functions, supports querying alarm data by time, event type, event level, etc., and generates alarm logs or reports. The logs are used for post-analysis, auditing, and maintenance.

[0144] To speed up retrieval, the system creates an index table for each type of data. The index includes fields such as timestamp, camera ID, and event type. These indexes support multi-dimensional queries and can quickly find data related to a certain alarm event or time period. The system uses database query optimization technologies, such as partition tables and multi-level caches, to ensure that it can still provide second-level responses when facing large-scale data sets. Users can enter query conditions through a graphical interface, and the system matches the corresponding index records according to the query conditions and quickly returns the results.

[0145] Query result display: Query results are displayed in table format, and detailed multimedia data (such as video clips, audio recordings, and snapshots) can be clicked. Users can also export query results as log files or reports for subsequent analysis or archiving.

[0146] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An AI video fire and smoke recognition alarm system for photovoltaic fields, characterized in that: include: Video acquisition module, used to collect visible light video and infrared video data of the photovoltaic field; Data transmission module, which transmits video data to the AI ​​integrated management platform through the optical fiber ring network; AI integrated management module, which uses artificial intelligence algorithms to perform fireworks recognition and intrusion detection on transmitted video data; Alarm module: When the AI ​​integrated management module detects fireworks or intrusion, it sends alarm information to the control center and triggers the corresponding alarm response; A data storage module for storing visible light and infrared video data and performing historical data retrieval based on time and event type; The system linkage module realizes the linkage control between the AI ​​integrated management platform and the GIS+ platform to perform fire location, camera priority control and electronic fence management; The simulation expulsion module works in conjunction with the intrusion detection unit to expel intruders based on the detection results.

2. The AI ​​video fire and smoke recognition alarm system for photovoltaic field according to claim 1 is characterized in that: The video acquisition module comprises: Visible light acquisition unit, used to collect visible light video data of photovoltaic fields around the clock; Infrared acquisition unit, used for thermal imaging video data acquisition at night and in bad weather; Motion compensation unit, used to correct image deviation caused by camera movement or vibration during video acquisition; Electronic image stabilization unit, used to stabilize video output and ensure the clarity and stability of captured images; The video ring network unit is used to connect various monitoring camera nodes in the photovoltaic field through optical fiber Ethernet to transmit information.

3. The AI ​​video fire and smoke recognition alarm system for photovoltaic field according to claim 1 is characterized in that: The data transmission module comprises: Optical fiber transmission unit, which transmits video data to the AI ​​integrated management platform at high speed through the optical fiber ring network; A wireless transmission unit, used for data transmission with a wireless bridge using solar power supply; A signal conversion unit converts the collected analog signal into a digital signal for transmission; The bandwidth management unit dynamically adjusts the network bandwidth to ensure the real-time and stability of video data transmission.

4. The AI ​​video fire and smoke recognition alarm system for photovoltaic fields according to claim 1 is characterized in that: The AI ​​comprehensive management module includes: The fire identification unit can intelligently identify the fire situation in the field by analyzing the fire characteristics in the video; Intrusion detection unit, which detects intrusion targets appearing in the video in real time; Self-learning unit, which continuously optimizes the accuracy of fire identification and intrusion detection through deep learning algorithms; Self-diagnosis unit monitors system status in real time and performs fault diagnosis.

5. The AI ​​video fire and smoke recognition alarm system for photovoltaic fields according to claim 4 is characterized in that: The fire identification unit comprises: Smoke detection unit, used to identify fire precursors by analyzing smoke shapes and color changes in the video; The flame detection unit determines fire by detecting the flame outline and dynamic characteristics in the video; Thermal imaging analysis unit, combining infrared video data to analyze and determine the temperature of the fire source; The false alarm filtering unit is used to filter the false alarm information caused by environmental factors.

6. The AI ​​video fire and smoke recognition alarm system for photovoltaic fields according to claim 4 is characterized in that: The intrusion detection unit comprises: Human body recognition unit, used to identify people entering the site and determine whether it is an illegal intrusion; Animal identification unit, used to identify and distinguish animal intrusions in the field to reduce false alarms; The behavior analysis unit analyzes the behavior characteristics of the intruder target to determine whether it poses a threat to the security of the site; The multi-target tracking unit tracks and locates multiple intrusion targets in real time, improving the response speed of intrusion detection.

7. The AI ​​video fire and smoke recognition alarm system for photovoltaic fields according to claim 1 is characterized in that: The alarm module comprises: Fire alarm unit: When the fire identification unit detects a fire, it triggers an alarm signal and uploads it to the control center; The intrusion alarm unit will immediately sound an alarm and link other security devices when an illegal intrusion is detected; The alarm response unit initiates corresponding emergency measures according to the alarm level; The alarm information upload unit uploads the fire and intrusion alarm information to the GIS+ platform for further processing and display.

8. The AI ​​video fire and smoke recognition alarm system for photovoltaic fields according to claim 1 is characterized in that: The data storage module comprises: A video data storage unit for storing visible light and infrared video data for a long time; An audio data storage unit, used to store the collected audio data and store it synchronously with the video data; Image storage unit, which stores the captured images and supports retrieval based on event type; Data retrieval unit, quickly retrieve historical data based on time, event type or abnormal situation conditions.

9. The AI ​​video fire and smoke recognition alarm system for photovoltaic fields according to claim 1 is characterized in that: The system linkage module includes: The fire location unit locates the latitude and longitude of the fire location by linking with the GIS+ platform; The camera control unit supports priority control of the camera to achieve automatic tracking of fire sources or intrusion targets; The electronic fence management unit detects intrusion according to the set fence area and automatically triggers an alarm when an intrusion occurs; The data interaction unit realizes data interaction and linkage control between the AI ​​comprehensive management platform and the GIS+ platform.

10. The AI ​​video fire and smoke recognition alarm system for photovoltaic fields according to claim 1, characterized in that: The simulation driving module comprises: A language prompt unit, used to play language prompts for warning and reminder when human intrusion is detected; An animal repellent unit that repels detected animals by playing the sounds of their natural enemies; Noise jammers are used to drive away animals by emitting high-frequency or low-frequency noise when animal intrusion is detected; The environmental adaptation unit automatically adjusts the type and intensity of the repelling measures according to the type of invader and environmental changes.

Images