Transformer substation outdoor environment intelligent fire early warning video identification method and system

Abstract

The invention relates to the technical field of transformer substations, in particular to a transformer substation outdoor environment intelligent fire early warning video identification method and system. The method comprises the following steps: acquiring monitoring video data in an outdoor area of a transformer substation, and preprocessing the monitoring video data to obtain preprocessed video data; presetting a fire smoke detection model, and inputting the preprocessed video data into the fire smoke detection model; and performing fire early warning according to the output of the fire smoke detection model. According to the invention, the network camera is arranged in the key area of the transformer substation, the network video transmission mode is adopted to replace the traditional serial port transmission mode, and the YOLOv5-based end-to-end target detection network is combined, so that large-range dead-corner-free monitoring and quick response are realized. The system adopts a composite loss function and a transfer learning strategy to carry out model training, the problem of insufficient samples is effectively solved, the system has good environmental adaptability, 91.8% of accuracy can still be kept even in foggy days, and reliable guarantee is provided for fire safety of the transformer substation.

Description

Technical Field

[0001] The present invention relates to the technical field of transformer substations, and in particular to a method and system for intelligent fire warning video recognition in an outdoor environment of a transformer substation. Background Art

[0002] Traditional fire detection technologies mainly include ionization and photoelectric smoke detectors, linear beam smoke detectors, and aspirating smoke detectors. These detectors are widely used in indoor environments, but they have some limitations in open spaces such as outdoor substations. Ionization and photoelectric smoke detectors rely on direct contact between smoke particles and sensors, and are both contact detectors. Therefore, they are easily disturbed by dust and other particulate matter in the environment, resulting in false alarms and poor environmental adaptability. In addition, they have limitations in detecting the location of fire in an environment with smooth airflow, have poor real-time performance, and are not suitable for large-space detection environments.

[0003] Traditional video smoke detection technology mainly refers to designing feature extractors and classifiers to identify smoke, and the core of its research is the extraction of smoke features. However, this method relies too much on manual feature extraction. This method requires experts to design feature extractors and classifiers, which is not only time-consuming and labor-intensive, but also the quality and comprehensiveness of the extracted features directly affect the effect of smoke detection. In addition, this method has poor adaptability to complex environments. Early texture feature extraction methods, such as edge detection, are difficult to adapt to complex backgrounds and lighting changes, resulting in poor detection results in complex environments.

[0004] Video smoke detection technology based on deep learning refers to combining the basic smoke features extracted by the shallow network through a deep neural network, abstracting the more advanced features of the smoke, mining the deep relationship of the feature data, reducing data redundancy, and avoiding complex manual extraction work. However, as the network structure becomes deeper and wider, this method helps to reduce missed detections and false detections, but it also increases the time and space complexity and the scale of network parameters, which may reduce the real-time detection and affect the deployment on resource-constrained devices. In addition, due to the relatively low probability of fire, the accumulation of smoke video image data is relatively weak, and there is currently no authoritative public data set. This limits the training and verification of deep learning models and may affect the accuracy and generalization ability of the models. Summary of the invention

[0005] In view of the problems existing in the prior art, the present invention is proposed.

[0006] Therefore, the problem to be solved by the present invention is how to solve the problems that traditional fire detection technology is prone to false alarms, is not suitable for large-space detection environments, traditional serial port transmission methods have poor real-time performance, traditional smoke detection modules based on artificial feature extraction have poor adaptability to complex environments, and the two-stage target detection network is difficult to lightweight.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] In a first aspect, an embodiment of the present invention provides a method for intelligent fire warning video recognition in an outdoor environment of a substation, which includes acquiring monitoring video data in an outdoor area of ​​the substation, and preprocessing the monitoring video data to obtain preprocessed video data;

[0009] Presetting a fire smoke detection model, and inputting the preprocessed video data into the fire smoke detection model;

[0010] A fire warning is performed according to the output of the fire smoke detection model.

[0011] As a preferred solution of the intelligent fire warning video recognition method for the outdoor environment of the substation described in the present invention, the monitoring video data is obtained by collecting through a network camera, and the network camera is installed at a commanding point or key area of ​​the substation, including near the transformer, cable area and fuel storage point.

[0012] As a preferred solution of the intelligent fire warning video recognition method for the outdoor environment of a substation described in the present invention, the preprocessing includes frame sampling and data enhancement of the monitoring video data; the data enhancement includes random flipping, rotation and brightness adjustment.

[0013] As a preferred solution of the intelligent fire warning video recognition method for the outdoor environment of the substation described in the present invention, the fire smoke detection model is an end-to-end target detection network based on YOLOv5; the loss function of the fire smoke detection model includes classification loss, positioning loss and confidence loss.

[0014] As a preferred solution of the intelligent fire warning video recognition method for the outdoor environment of the substation described in the present invention, the training process of the fire smoke detection model includes: pre-training using a general data set; transfer learning and model fine-tuning using a specific smoke detection data set; and optimizing model parameters through control variable research.

[0015] As a preferred solution of the intelligent fire warning video recognition method for the outdoor environment of the substation described in the present invention, when smoke is detected, the key frames and related detection information of the smoke detection are recorded, a warning signal is generated and the warning information push is triggered, and the smoke detection results are uploaded to the cloud data center.

[0016] As a preferred solution of the intelligent fire warning video recognition method for the outdoor environment of the substation described in the present invention, the warning information push includes sending a notification to emergency response personnel via email, text message or mobile application; the warning information includes the time information and location information of the smoke occurrence.

[0017] In a second aspect, an embodiment of the present invention provides an intelligent fire warning video recognition system for an outdoor environment of a substation, which includes.

[0018] The video acquisition module is used to obtain the monitoring video data in the outdoor area of ​​the substation through a network camera and perform pre-processing;

[0019] Smoke detection module, used to preset the fire smoke detection model and analyze the preprocessed video data;

[0020] The early warning module is used to generate early warning signals and push early warning information based on smoke detection results.

[0021] In a third aspect, an embodiment of the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, the steps of the intelligent fire warning video recognition method for the outdoor environment of the substation as described in the first aspect of the present invention are implemented.

[0022] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, the steps of the method for intelligent fire warning video recognition for the outdoor environment of a substation as described in the first aspect of the present invention are implemented.

[0023] The beneficial effects of the present invention are as follows: by adopting a target recognition algorithm based on an end-to-end target detection network and an optimized YOLOv5 model, the present invention can more accurately identify smoke from a video stream and maintain a high accuracy rate even in a complex and changeable outdoor environment;

[0024] The system design takes into account the need for real-time monitoring, and can quickly respond and detect smoke, thereby issuing early warning signals in a timely manner, buying valuable time for early detection and treatment of fires;

[0025] Through meticulous model training and optimization, as well as control variable research, the present invention significantly reduces the risk of false positives and false negatives.

[0026] The modular design allows the system to be easily integrated into existing monitoring networks without the need for large-scale modifications to existing facilities, facilitating rapid deployment.

[0027] Specially designed webcams and algorithm optimization enable the system to work stably under various environmental conditions, including different lighting, weather, and background noise.

[0028] Compared with the prior art, the substation outdoor environment intelligent fire warning video recognition system proposed in the present invention has good real-time performance, adaptability, scalability and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0030] Figure 1 It is a flow chart of the intelligent fire warning video recognition method for the outdoor environment of the substation;

[0031] Figure 2 A computer equipment diagram for an intelligent fire warning video recognition method for an outdoor environment of a substation;

[0032] Figure 3 A schematic diagram of a fire warning video recognition system for an intelligent fire warning video recognition method for an outdoor environment of a substation;

[0033] Figure 4 A schematic diagram of the hardware connection of a fire warning video recognition system for an intelligent fire warning video recognition method for an outdoor environment of a substation;

[0034] Figure 5 This is a line graph showing the change in loss value during the training process of the YOLOv5 network model with different depths and widths for the intelligent fire warning video recognition method for the outdoor environment of the substation;

[0035] Figure 6 The line graph of the loss value change before and after model fine-tuning for the intelligent fire warning video recognition method for the outdoor environment of the substation;

[0036] Figure 7 It is a line chart of the change in loss value between the optimal model and the basic model in the control variable research of the intelligent fire warning video recognition method for the outdoor environment of the substation;

[0037] Figure 8 This is a visualization diagram of the output of the model on the test set for the intelligent fire warning video recognition method for the outdoor environment of the substation;

[0038] Fig. 9 This is a visualization diagram of the output of the model of the public smoke video for the intelligent fire warning video recognition method for the outdoor environment of the substation. DETAILED DESCRIPTION

[0039] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0040] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0041] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or selective embodiment that is mutually exclusive with other embodiments.

[0042] Example 1

[0043] Reference Figures 1 to 9 , which is the first embodiment of the present invention, and provides a method for intelligent fire warning video recognition in outdoor environment of a substation, comprising:

[0044] In the embodiment of the present application, the video acquisition module uses a network camera for real-time monitoring and transmits data via network video transmission;

[0045] In an optional embodiment, the network camera is installed at the commanding heights and key areas of the substation, including but not limited to key fire prevention areas such as near transformers, cable areas, fuel storage points, etc. The camera has waterproof, dustproof, and corrosion-resistant properties to ensure stable operation in harsh environments.

[0046] In an embodiment of the present application, preprocessing includes frame sampling and data enhancement of surveillance video data; the data enhancement includes operations such as random flipping, rotation, and brightness adjustment.

[0047] In an optional embodiment, data enhancement may also include Mosaic data enhancement, which increases the adaptability of the model to different scenes by splicing multiple images together. In addition, the diversity of the data may be further enhanced by adjusting the contrast, adding noise, etc.

[0048] It should be noted that the preprocessing step can improve the image quality, enhance the generalization ability of the model, and reduce the impact of environmental factors on the detection results. At the same time, through a reasonable preprocessing strategy, the computational load can be reduced and the real-time performance of the system can be improved.

[0049] For example, in the embodiment of the present application, smoke video data from multiple substations are collected, totaling about 800 videos, to ensure that smoke scenes under different weather conditions and lighting conditions are covered. After preprocessing, a data set containing 2000 training samples and a test set of 500 verification samples are generated.

[0050] S100: Acquire monitoring video data in an outdoor area of ​​a substation, and preprocess the monitoring video data to obtain preprocessed video data;

[0051] S101: The monitoring video data is acquired through a network camera, which is installed at a commanding point or a key area of ​​the substation, including near the transformer, the cable area, and the fuel storage point.

[0052] S102: The preprocessing includes frame sampling and data enhancement of the surveillance video data; the data enhancement includes random flipping, rotation and brightness adjustment.

[0053] S200: Preset a fire smoke detection model, and input the preprocessed video data into the fire smoke detection model;

[0054] Preferably, a fire smoke detection model is preset, and the preprocessed video data is input into the fire smoke detection model;

[0055] In an embodiment of the present application, the fire smoke detection model adopts an end-to-end target detection network based on YOLOv5, and its loss function includes classification loss, positioning loss and confidence loss.

[0056] In an optional embodiment, the model training adopts a transfer learning strategy, first pre-training using a large-scale general dataset, and then fine-tuning on a specific smoke detection dataset. In this way, the feature extraction capability of the pre-trained model can be fully utilized to improve the performance of the model.

[0057] It should be noted that the model training process requires careful adjustment of hyperparameters, including learning rate, batch size, etc. At the same time, the model structure is optimized by controlling variables to achieve lightweight deployment while ensuring detection accuracy.

[0058] In the embodiment of the present application, YOLOv5s is selected as the basic model, and the following key parameters are set:

[0059] Learning rate: The initial value is set to 0.01, and the cosine annealing strategy is adopted; batch size: 32; number of training rounds: 100 rounds; optimizer: Adam optimizer; IOU threshold: 0.5;

[0060] The loss function of the model is defined as follows:

[0061] L=λ1Lcls+λ2Lbox+λ3Lobj

[0062] Among them: Lcls is the classification loss, using cross entropy loss; Lbox is the bounding box regression loss, using CIoU loss; Lobj is the target confidence loss; λ1, λ2, λ3 are weight coefficients, set to 1.0, 0.05, and 0.5 respectively;

[0063] S201: The fire smoke detection model is an end-to-end target detection network based on YOLOv5; the loss function of the fire smoke detection model includes classification loss, positioning loss and confidence loss.

[0064] S202: The training process of the fire smoke detection model includes: using a general data set for pre-training; using a specific smoke detection data set for transfer learning and model fine-tuning; and optimizing model parameters by controlling variable research.

[0065] S203: When smoke is detected, the key frames and related detection information of the smoke detection are recorded, an early warning signal is generated and a push of early warning information is triggered, and the smoke detection result is uploaded to the cloud data center.

[0066] S300: Providing a fire warning according to the output of the fire smoke detection model.

[0067] S301: The warning information push includes sending a notification to emergency response personnel via email, text message or mobile application; the warning information includes the time information and location information of the smoke occurrence.

[0068] In the embodiment of the present application, when smoke is detected, the system will:

[0069] Record the key frames and related detection information of smoke detection;

[0070] Generate early warning signals and trigger early warning information push;

[0071] Upload smoke detection results to the cloud data center;

[0072] In an optional embodiment, the warning information is pushed through multiple channels, including email, text messages, and mobile application push. The warning information contains key information such as the time and location of the smoke, ensuring that emergency response personnel can respond in time.

[0073] In addition, the present embodiment also includes a specially designed circuit breaker device for automatically cutting off non-fire power supply after fire is confirmed. The circuit breaker device includes a shunt release and a micro switch, and ensures safe and reliable power-off operation through a special control mechanism.

[0074] Furthermore, this embodiment also provides a substation outdoor environment intelligent fire warning video recognition system, including:

[0075] The video acquisition module is used to obtain the monitoring video data in the outdoor area of ​​the substation through a network camera and perform pre-processing;

[0076] Smoke detection module, used to preset the fire smoke detection model and analyze the preprocessed video data;

[0077] The early warning module is used to generate early warning signals and push early warning information based on smoke detection results.

[0078] This embodiment also provides a computer device, which is suitable for the case of an intelligent fire warning video recognition method for an outdoor environment of a substation, and includes a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the intelligent fire warning video recognition method for an outdoor environment of a substation as proposed in the above embodiment.

[0079] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0080] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method for realizing intelligent fire warning video recognition for outdoor environment of a substation as proposed in the above embodiment is implemented.

[0081] In summary, by placing network cameras at commanding heights and key areas of the substation to replace traditional contact detectors, a wide range of monitoring without blind spots is achieved. This non-contact monitoring method avoids the problem of false alarms caused by traditional detectors being easily disturbed by dust, while overcoming the defect of limited detection range of traditional detectors, and improving the reliability and coverage of fire warnings.

[0082] By adopting network video transmission instead of traditional serial port transmission, the data transmission speed is greatly improved. This method is easy to deploy in open and complex substation environments, reducing the need for additional wiring, not only reducing installation and maintenance costs, but more importantly, achieving real-time transmission of video data, reducing the system response time from the traditional 15 seconds to less than 3 seconds.

[0083] The training samples were effectively expanded by introducing frame sampling and data enhancement preprocessing techniques, including random flipping, rotation, and brightness adjustment. This preprocessing method not only improves the model's adaptability to different scenarios, but also solves the problem of scarce fire data, increasing the model's recognition accuracy to 95.8% under various environmental conditions.

[0084] By adopting an end-to-end target detection network based on YOLOv5, replacing the traditional two-stage detection network and manual feature extraction method, the network structure is lightweight. This innovative model design significantly reduces the computational complexity, making the average processing delay of the system on the edge computing device less than 200ms, while keeping the false alarm rate below 2.3%.

[0085] By designing a composite loss function (including classification loss, positioning loss and confidence loss) and adopting a transfer learning strategy, the problem of insufficient model training samples was solved. This method not only improves the generalization ability of the model, but also enables the system to maintain stable detection performance under different weather conditions, and can maintain an accuracy of 91.8% even in the most unfavorable foggy conditions.

[0086] By designing a multi-level warning mechanism, including key frame recording, multi-channel information push and cloud data upload, a complete fire warning response chain has been established. This mechanism not only ensures the timely delivery of warning information, but also provides detailed data support for later analysis, further improving the efficiency of fire handling.

[0087] In summary, the present invention solves the adaptability problem of the traditional fire warning system in the outdoor environment of the substation, and has achieved significant improvements in detection accuracy, response speed and system reliability. In particular, in complex outdoor environments, the system can still maintain stable high performance, and these technical effects are difficult to predict under the existing technical level.

[0088] Example 2

[0089] Reference Figure 2 - Fig. 9 , which is the second embodiment of the present invention, and this embodiment provides a method for intelligent fire warning video recognition in the outdoor environment of a substation. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0090] S1: Video acquisition: Use a network camera to capture the video image outside the substation in real time, and transmit the image to the smoke detection module through network video transmission;

[0091] S2: Smoke detection: Real-time analysis of the collected video stream, using an end-to-end object detection network to detect whether there is smoke in the video. If smoke is detected, key frames are extracted and time and location information is recorded.

[0092] S3: Warning signal generation: Once smoke is detected, the system will generate a warning signal and trigger the early fire warning module, while temporarily storing the key frames of smoke detection and related detection information;

[0093] S4: Warning information push and data upload: The system pushes warning information to emergency response personnel via SMS and email, and uploads smoke detection results to the cloud data center for fire cause analysis and accident archiving.

[0094] The video acquisition module designed according to step S1 is responsible for acquiring the on-site video stream to be detected, ensuring the continuity and high quality of the video data so that the subsequent smoke detection module can accurately analyze the video content and extract smoke features. The input part of the video acquisition module consists of high-definition network cameras, which are installed in key locations outside the substation to achieve comprehensive monitoring. These cameras support the transmission of high-resolution video streams and usually have night vision capabilities and stable working capabilities under severe weather conditions to ensure that video data can be reliably captured under any environmental conditions.

[0095] The smoke detection module designed according to step S2 is responsible for real-time analysis of the video stream transmitted by the video acquisition module to detect whether there is smoke. This module uses a target recognition algorithm based on an end-to-end target detection network, which can identify the visual characteristics of smoke. The module first samples the video stream at a certain frame rate to obtain a single frame image of the video. Then, these images are analyzed using a well-trained target detection network to determine whether there is smoke. If the model recognizes the smoke feature, the module will mark the current frame, record the time and location of the smoke, and trigger an early warning signal. At the same time, the key frames containing smoke and related detection information are stored to provide a basis for subsequent early warning and analysis. In addition, the module can also adapt to different environmental conditions, reduce false alarms and missed alarms, and ensure the accuracy and timeliness of fire warnings.

[0096] The early fire warning module designed according to step S3 and step S4 is responsible for receiving the warning signal generated by the smoke detection module, and on this basis, immediately sends a warning notification to the emergency response personnel, through SMS, email or other communication methods, to ensure that relevant personnel can quickly obtain fire warning information so as to take timely response measures. After that, the key frame images of the on-site smoke detection, the time and location of the smoke generation and other detailed information are packaged and remotely transmitted to the cloud data center via the wireless network to facilitate fire cause analysis and accident records.

[0097] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0098] The video acquisition module first obtains the on-site source video to be detected, and then transmits the video stream to the smoke detection module. After receiving the video stream to be detected, the smoke detection module begins to analyze the video data. If the analysis result shows that smoke is generated at the scene, the smoke detection result (smoke detection key frames and related detection information) is temporarily stored and a warning signal is generated, triggering the early fire warning module to start the warning task. After receiving the warning signal, the early fire warning module immediately issues a warning to the relevant emergency response personnel, and then uploads the on-site smoke detection results to the cloud data center to assist the cloud data center in conducting fire cause analysis and accident archiving later. The system operation process is as follows: Figure 1 shown.

[0099] The outdoor industrial park has a wide space, and the hardware system deployment needs to take into account the ease of installation, maintainability and scalability. The hardware of the new substation outdoor environment intelligent fire warning video recognition system is as follows: Figure 2 As shown in the figure, it is mainly composed of four parts: main control board, network camera, power supply module and heat dissipation device. The implementation plan of each functional module is discussed in detail below.

[0100] In the video acquisition module constructed by the present invention, the transmission process adopts the network video transmission mode, which is convenient for deployment in an open and possibly complex outdoor environment such as a substation. In this embodiment, a gun-type network camera dedicated to park video surveillance is used. The network camera can be directly connected to the existing network system of the substation, reducing the need for additional wiring, making the deployment process more convenient and economical, with high imaging quality, flexible installation, and fast transmission rate. The selection of the camera installation position takes into account the breadth and clarity of the monitoring field of view, as well as the coverage of potential fire sources, to ensure the comprehensiveness and effectiveness of video acquisition. In actual deployment, the camera is usually installed at the commanding heights or key areas of the substation, such as near the transformer, the cable area or the fuel storage point, to achieve real-time monitoring of areas with high incidence of fire. At the same time, considering outdoor environmental factors such as weather changes, temperature fluctuations, etc., the selected network camera must have waterproof, dustproof, corrosion-resistant and other characteristics to ensure stable operation in harsh environments. In addition, in order to ensure the stability and security of network video transmission, the video data will be encrypted during the transmission process to prevent the data from being intercepted or tampered with. The choice of network bandwidth and transmission rate will also be adjusted based on actual monitoring needs to balance video quality and transmission efficiency.

[0101] In the smoke detection module constructed by the present invention, a target recognition algorithm based on an end-to-end target detection network is adopted, and the module recognizes the visual features of smoke by training the model. The specific implementation of the module depends on a high-performance computing platform, such as Raspberry Pi 4 Model B, which is equipped with a powerful SOC main control chip and can handle complex video analysis tasks. In this embodiment, YOLOv5 is selected as the core algorithm. YOLOv5 is widely recognized in the field of object detection for its fast and accurate detection capabilities. Its single-stage detection mechanism effectively reduces the calculation steps and improves the processing speed. In order to train this model, a comprehensive smoke image dataset is constructed, which contains smoke samples captured under various environmental conditions to ensure the adaptability of the model to different scenarios.

[0102] During the model training process, this embodiment applies data enhancement technology to improve the generalization ability of the model. Through random flipping, rotation, and brightness adjustment, overfitting is effectively avoided and the robustness of the model is enhanced. In addition, through transfer learning technology, a large-scale general data set is first used for pre-training, and then fine-tuned on a specific smoke detection data set, further improving the model's ability to recognize smoke.

[0103] In terms of model optimization, this embodiment has carried out detailed hyperparameter adjustment and network structure improvement to seek the best balance between detection speed and accuracy. After a series of experiments, the experimental results are as follows: Figure 5As shown in Figure 1, the YOLOv5s model is selected, which not only meets the real-time requirements but also ensures high detection accuracy. In addition, a composite loss function is used, including classification loss, localization loss, and confidence loss, to comprehensively guide model training.

[0104] In the model fine-tuning stage, this implementation adopts heterogeneous transfer learning technology and adopts a detailed strategy to adjust the model parameters, including the arrangement of learning rate, the selection of batch size, and the optimization of loss function. The implementation of these strategies further ensures the accuracy and robustness of the model in smoke detection and further improves the detection performance of the model. Finally, through testing on multiple public smoke video datasets, the test results are as follows Figure 6 As shown, the robustness and practicability of the model are verified, ensuring its reliability in the outdoor environment of substations.

[0105] In the control variable study, this embodiment systematically adjusted the key factors in the training process, including data enhancement technology, the number of predicted anchor boxes, and the batch size, and conducted multiple groups of comparative experiments. The experimental implementation process is shown in the following table:

[0106]

[0107] By introducing Mosaic data enhancement, the model can handle more complex backgrounds and improve the model's ability to recognize smoke in different backgrounds. In addition, by increasing the number of predicted anchor boxes, the model can more flexibly detect smoke targets of different sizes, improving the accuracy of detection. The adjustment of the batch size affects the direction and stability of the gradient descent during model training. By optimizing the batch size, the performance of the model is further improved. The effect of the optimized model is as follows: Figure 7 shown.

[0108] Finally, this example uses the test set to perform a visual test on the optimized model, and the results are as follows: Figure 8 Next, the smoke detection module of this embodiment is used to test 12 public smoke videos. Among them, 3 videos have high similarity with the background of the rest of the videos, so only the detection screenshots of 9 videos with large background differences are shown, as shown in Figure 1. Fig. 9 shown.

[0109] In the early fire warning module constructed by the present invention, this embodiment implements a set of efficient notification and data upload mechanisms for rapid response when smoke is detected. The early fire warning module immediately sends warning notifications to emergency response personnel through preset communication channels, such as email, text messages or mobile application push. These notifications contain key information of smoke detection, such as the time and location of occurrence. At the same time, the early fire warning module uploads the smoke detection results, including key frame images and detection information, to the cloud data center. This step ensures the backup and traceability of accident data, and provides important information for subsequent fire cause analysis and accident archiving. In this embodiment, after the fire is confirmed, the expansion module issues a linkage disconnection command to the circuit of the non-fire power supply that needs to be disconnected in the dynamic lighting box or low-voltage cabinet through the control module. For the non-fire power supply circuit that needs to be automatically disconnected in the event of a fire, a specially designed circuit breaker device is used. The circuit breaker device includes a shunt release, which is configured to remotely control the disconnection operation of the circuit breaker. The shunt release is designed for short-time operation, which means that the power-on time of its coil must be strictly controlled within 1 second to prevent the coil from burning due to long-term power-on. In order to avoid the risk of coil burning, a micro switch is integrated in the circuit breaker. The micro switch is connected in series with the coil of the shunt release and is set to a normally closed state. When the shunt release is actuated, the armature is attracted to cause the state of the micro switch to switch from normally closed to normally open, thereby cutting off the power control circuit of the shunt release. In this way, even if the operator continues to press the control button, the shunt coil will not be energized again, effectively avoiding the risk of the coil burning due to long-term power-on. Furthermore, when the circuit breaker returns to the closed position, the micro switch also returns to its initial normally closed state, ensuring the safety and reliability of the system. Through this mechanism, the present invention can ensure that non-fire power is quickly cut off in a fire emergency, thereby improving the efficiency and safety of fire rescue.

[0110] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for intelligent fire warning video recognition in outdoor environment of substation, characterized by: include, Acquire monitoring video data in the outdoor area of ​​the substation, and preprocess the monitoring video data to obtain preprocessed video data; Presetting a fire smoke detection model, and inputting the preprocessed video data into the fire smoke detection model; A fire warning is performed according to the output of the fire smoke detection model.

2. The method for intelligent fire warning video recognition in outdoor environment of substation according to claim 1, characterized in that: The monitoring video data is acquired through network cameras, which are installed at commanding heights or key areas of the substation, including near transformers, cable areas, and fuel storage points.

3. The method for intelligent fire warning video recognition in outdoor environment of substation according to claim 2, characterized in that: The preprocessing includes frame sampling and data enhancement of the surveillance video data; the data enhancement includes random flipping, rotation and brightness adjustment.

4. The method for intelligent fire warning video recognition in outdoor environment of substation according to claim 3, characterized in that: The fire smoke detection model is an end-to-end target detection network based on YOLOv5; the loss function of the fire smoke detection model includes classification loss, positioning loss and confidence loss.

5. The method for intelligent fire warning video recognition in outdoor environment of substation according to claim 4, characterized in that: The training process of the fire smoke detection model includes: using a general data set for pre-training; using a specific smoke detection data set for transfer learning and model fine-tuning; and optimizing model parameters by controlling variable research.

6. The method for intelligent fire warning video recognition in outdoor environment of substation according to claim 5, characterized in that: When smoke is detected, the key frames and related detection information of smoke detection are recorded, an early warning signal is generated and the early warning information push is triggered, and the smoke detection results are uploaded to the cloud data center.

7. The method for intelligent fire warning video recognition in outdoor environment of substation according to claim 6, characterized in that: The push of warning information includes sending a notification to emergency response personnel via email, text message or mobile application; the warning information includes the time information and location information of the smoke occurrence.

8. A substation outdoor environment intelligent fire warning video recognition system, based on the substation outdoor environment intelligent fire warning video recognition method according to any one of claims 1 to 7, characterized in that: Also includes, The video acquisition module is used to obtain the monitoring video data in the outdoor area of ​​the substation through a network camera and perform pre-processing; Smoke detection module, used to preset the fire smoke detection model and analyze the preprocessed video data; The early warning module is used to generate early warning signals and push early warning information based on smoke detection results.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for intelligent fire warning video recognition for outdoor environment of a substation according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for intelligent fire warning video recognition for outdoor environment of a substation according to any one of claims 1 to 7 are implemented.

Images