Electric bicycle shed fire extinguishing system and method based on automatic monitoring, alarming and fire extinguishing

Through multi-dimensional data analysis and artificial intelligence technology, combined with fire picture recognition and smoke information, the timeliness and accuracy of traditional electric bicycle shed fire prevention systems has been solved, and automatic monitoring, alarm and fire extinguishing of electric bicycle shed fires has been realized, enhancing the enhancement of the electric bicycle shed fire. The accuracy of fire prediction.

CN119028074APending Publication Date: 2024-11-26JIANGXI VIKEN TECH CO LTD
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202410978801.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-22
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

The fire prevention system of traditional electric bicycle sheds lacks timeliness and accuracy, and is prone to misjudgment due to environmental changes and individual data misjudgment, and cannot effectively predict the occurrence of fires.

Method used

Using an electric bicycle shed system based on automatic monitoring, alarm and fire extinguishing, through multi-dimensional data analysis, combined with fire picture recognition, smoke information, temperature information and current information, use artificial intelligence technology to monitor and predict fires, output fire warning levels and implement corresponding ones Alarm and fire-fighting measures.

Benefits of technology

It improves the accuracy and timeliness of fire monitoring, reduces the misjudgment rate, realizes automatic monitoring, alarm and fire extinguishing of electric bicycle shed fires, and enhances the accuracy and identification capabilities of fire prediction.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119028074A_ABST
    Figure CN119028074A_ABST
Patent Text Reader

Abstract

The invention discloses an electric bicycle shed fire extinguishing system and method based on automatic monitoring, alarming and fire extinguishing, and relates to the technical field of fire monitoring. An electric bicycle shed fire extinguishing system based on automatic monitoring, alarming and fire extinguishing comprises a data acquisition module, a fire monitoring module, a fire prediction module, a fire analysis module and an alarming and fire extinguishing module. According to the invention, through comprehensive analysis of the multi-dimensional data, fire monitoring is carried out in combination with the fire picture recognition result, the smoke information and the temperature information, misjudgment caused by errors of single data is avoided, and the accuracy of fire monitoring is improved; the time sequence data of the smoke information and the temperature information, the weather information and the season information are combined to carry out fire prediction, and the data before the fire is generated are analyzed and predicted, so that the timeliness of fire monitoring is improved; the accuracy of fire judgment is improved according to joint analysis of fire monitoring and fire prediction, and automatic monitoring, alarming and fire extinguishing of the electric bicycle shed fire are achieved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of fire monitoring, and in particular to an electric bicycle shed fire extinguishing system and method based on automatic monitoring, alarming and fire extinguishing. Background Art

[0002] In recent years, with the improvement of environmental awareness and the continuous advancement of transportation technology, the use of electric bicycles in urban transportation has become increasingly popular. Electric bicycles have the advantages of environmental protection and convenience, and have been favored by more and more people. The number of electric bicycles in my country has reached 400 million. The charging process of electric bicycles is very prone to fire, and electric bicycles are parked densely. If a fire occurs, it is very easy to spread and cause secondary disasters. With the frequent occurrence of electric bicycle fire accidents, fire prevention in electric bicycle sheds is imminent.

[0003] Fire prevention in traditional electric bicycle sheds is mainly based on simple detection of smoke sensors and temperature sensors. This method can only detect fires after they occur and lacks timeliness. The use of sensors in outdoor environments is easily distorted due to environmental changes and lacks accuracy. Data from different sensors cannot be analyzed in a linked manner, which is prone to misjudgment.

[0004] With the improvement of computer basic computing power, artificial intelligence technology is increasingly used in our lives. Artificial intelligence technology can imitate human thinking and improve accuracy by analyzing multi-source data when making judgments. In fire monitoring, the use of existing sensor data combined with artificial intelligence technology analysis can greatly improve the accuracy of fire monitoring and achieve the prediction of fire occurrence.

[0005] Therefore, we urgently need to provide an electric bicycle shed fire extinguishing system and method based on automatic monitoring, alarm and fire extinguishing. Summary of the invention

[0006] The present invention proposes an electric bicycle shed fire extinguishing system and method based on automatic monitoring, alarm and fire extinguishing, aiming to improve the accuracy and reliability of automatic fire monitoring through in-depth analysis of data in multiple dimensions.

[0007] The fire extinguishing method of the electric bicycle shed based on automatic monitoring, alarm and fire extinguishing includes:

[0008] S1. Collect smoke information, temperature information, current information and monitoring pictures of the electric bicycle shed at a set period T;

[0009] S2. Send smoke information, temperature information and monitoring pictures to the fire monitoring module;

[0010] Send the monitoring picture to the fire picture recognition model in the fire monitoring module for fire picture recognition, and output the monitoring picture recognition results, which include normal, smoke, flame, smoke and flame, and fire hazard;

[0011] The fire monitoring module conducts a comprehensive analysis of smoke information, temperature information and monitoring image recognition results, and outputs fire monitoring results, which include fire occurrence, no fire occurrence and signs of fire;

[0012] S3. Send all smoke information and temperature information in the current cycle and the previous fixed time period to the fire prediction model for fire prediction, and output the fire prediction results, which include fire prediction and no fire prediction;

[0013] S4. Perform fire hazard detection based on the current information of the current cycle, the current information of the previous cycle and the monitoring image recognition results. The fire hazard detection results include whether there is a fire hazard or not;

[0014] S5. Send the fire monitoring results, fire prediction results and fire hazard detection results to the fire analysis module, and output the fire warning level, which includes red warning, yellow warning and blue warning;

[0015] S6. Send the fire warning level to the alarm and fire extinguishing module, which implements alarm and fire extinguishing measures according to the fire warning level. A red warning activates the sound alarm and light alarm, and starts the fire extinguishing device; a yellow warning activates the sound alarm and light alarm; a blue warning only activates the light alarm.

[0016] As a preferred technical solution of the present invention, the fire picture recognition model includes a mode selection layer, a picture preprocessing layer and a picture recognition layer. The mode selection layer is used to select a day mode or a night mode according to the current time. The day mode sends the current monitoring picture to the day channel of the fire picture recognition model for recognition, and the night mode sends the current monitoring picture to the night channel of the fire picture recognition model for recognition;

[0017] The image preprocessing layer is used to preprocess the monitoring image information to obtain the monitoring image data to be identified;

[0018] The image recognition layer is used to classify the monitoring image data to be identified and output the monitoring image recognition results. The monitoring image recognition results include three categories: normal, fire and fire hazard. Fire includes smoke, flame, smoke and flame.

[0019] As a preferred technical solution of the present invention, it also includes establishing a fire image recognition model, the specific steps are:

[0020] A1. Collect several electric bicycle shed images and label them with category labels; pre-process the labeled electric bicycle shed images to obtain a fire image classification dataset; divide the fire image classification dataset into two subsets according to time: a daytime fire image classification dataset and a nighttime fire image classification dataset; split the daytime fire image classification dataset into a daytime fire image classification training set and a daytime fire image test set, and split the nighttime fire image classification dataset into a nighttime fire image classification training set and a nighttime fire image test set;

[0021] A2. Send the daytime fire image classification training set and the nighttime fire image classification training set to the two-stream deep residual network model in batches for alternating training, with the category label as the target. After the training is completed, the initial fire image recognition model is obtained;

[0022] A3. Send the daytime fire image test set and the nighttime fire image test set to the initial fire image recognition model for accuracy evaluation respectively, and obtain the test accuracy with the category label as the target; if the test accuracy is within the error standard range, the initial fire image recognition model is the fire image recognition model; if the test accuracy exceeds the error standard range, adjust the hyperparameters and repeat steps A2 and A3.

[0023] As a preferred technical solution of the present invention, it also includes a fire monitoring module outputting fire monitoring results, specifically the following steps:

[0024] B1. Obtain smoke information, temperature information and monitoring image recognition results;

[0025] B2. Check whether the smoke information and temperature information exceed the alarm threshold. If both exceed the output, a fire occurs.

[0026] If the smoke information exceeds the alarm threshold and the temperature information does not exceed the alarm threshold, check the monitoring image recognition result. If it is flame or flame and smoke, output that a fire has occurred;

[0027] If the temperature information exceeds the alarm threshold and the smoke information does not exceed the alarm threshold, check the monitoring image recognition result. If it is smoke or flame and smoke, output that a fire has occurred;

[0028] If both the smoke information and the temperature information exceed the alarm threshold, check the monitoring image recognition results. If it is flame, smoke, or flame and smoke, output that there is a sign of fire;

[0029] Otherwise the output does not fire.

[0030] As a preferred technical solution of the present invention, the fire prediction model includes a data processing layer and a fire prediction layer;

[0031] The data processing layer is used to collect weather information and seasonal information, and pre-process the weather information, seasonal information, all smoke information and temperature information in the current cycle and the previous fixed time period to obtain the information to be predicted;

[0032] The fire prediction layer is used to classify the information to be predicted and output the fire prediction results, which include those with fire prediction and those without fire prediction.

[0033] As a preferred technical solution of the present invention, it also includes establishing a fire prediction model, the specific steps are:

[0034] C1. Collect carport smoke information and carport temperature information of a certain fixed time, as well as the weather information and season information at that time as prediction information, add fire labels to the prediction information, pre-process the prediction information with the fire labels, and obtain a prediction data set; divide the prediction data set into a prediction training set and a prediction test set;

[0035] C2. Send the prediction training set to the LSTM model integrated with the self-attention mechanism for training, with the fire label as the target. After the training is completed, the initial fire prediction model is obtained;

[0036] C3. Send the predicted test set to the initial fire prediction model for accuracy evaluation, and take the fire label as the target to obtain the test accuracy; if the test accuracy is within the error standard range, the initial fire prediction model is the fire prediction model. If the test accuracy exceeds the error standard range, adjust the hyperparameters and repeat steps C2 and C3.

[0037] As a preferred technical solution of the present invention, it also includes detecting fire hazards, specifically the following steps:

[0038] D1. Obtain the current information of the current cycle, the current information of the previous cycle and the image recognition result;

[0039] D2. Calculate the change C of the current information and determine whether the change C of the current information exceeds the dangerous threshold RTH;

[0040] D3. If the change C of the current information exceeds the danger threshold RTH or the image recognition result is a fire hazard, output that there is a fire hazard, otherwise output that there is no fire hazard.

[0041] As a preferred technical solution of the present invention, it also includes a fire analysis module, and the specific steps are:

[0042] E1. Obtain fire monitoring results, fire prediction results and fire hazard detection results;

[0043] E2. Check the fire monitoring results for three consecutive times. If all of them indicate a fire, a red warning will be output;

[0044] If there is a fire or signs of fire in three consecutive fire monitoring results, and the fire prediction result shows that there is a fire sign, a red warning will be output;

[0045] If there are signs of fire in three consecutive fire monitoring results and the fire prediction results show that there are signs of fire, a yellow warning will be output;

[0046] If the fire hazard detection is that there is a fire hazard, a blue warning is output.

[0047] The electric bicycle shed fire extinguishing system based on automatic monitoring, alarm and fire extinguishing includes:

[0048] A data acquisition module is used to collect smoke information, temperature information, current information and monitoring pictures of the electric bicycle shed at a set period T;

[0049] The fire monitoring module is used to deploy a fire image recognition model. The fire image recognition model contains two channels. It switches between the daytime channel and the nighttime channel according to the time to perform fire image recognition on the monitoring images and outputs the monitoring image recognition results. The monitoring image recognition results are used to conduct fire assessment in the direction of the monitoring images. The fire monitoring model conducts a comprehensive analysis of the smoke information, temperature information and monitoring image recognition results and outputs the fire monitoring results. The fire monitoring results are used to assess the real-time fire situation.

[0050] The fire prediction module is used to deploy a fire prediction model. The fire prediction model predicts fires based on smoke information, temperature information, weather information, and season information over a period of time, and outputs fire prediction results. The fire prediction results are used to evaluate the signs of fire occurrence.

[0051] The fire analysis module is used to detect fire hazards based on the current information of the current cycle, the current information of the previous cycle and the monitoring image recognition results, and output the fire hazard detection results; the fire analysis module comprehensively analyzes the fire monitoring results, fire prediction results and fire hazard detection results to obtain the fire warning level, which includes red warning, yellow warning and blue warning;

[0052] The alarm and fire extinguishing module is used to implement alarm and fire extinguishing measures according to the fire warning level. The red warning activates the sound alarm and light alarm, and starts the fire extinguishing device; the yellow warning activates the sound alarm and light alarm; the blue warning only activates the light alarm.

[0053] The present invention has the following advantages:

[0054] 1. The present invention conducts fire monitoring by comprehensively analyzing multidimensional data and combining fire image recognition results, smoke information and temperature information, thereby avoiding misjudgment caused by errors in single data and improving the accuracy of fire monitoring; fire prediction is performed by combining time series data of smoke information and temperature information, weather information and seasonal information, and data before the fire occurs is analyzed and predicted, thereby improving the timeliness of fire monitoring; fire hazard detection is performed by combining current information and image recognition information, and fire hazards are discovered in time; the present invention comprehensively judges whether to alarm and extinguish the fire based on the fire monitoring results, fire prediction results and fire hazard results, and monitoring and prediction are coordinated to improve the accuracy of fire judgment, thereby realizing automatic monitoring, alarm and fire extinguishing of electric bicycle shed fires.

[0055] 2. The present invention uses a dual-stream fire image recognition model to adjust the recognition model to day mode and night mode according to time, so as to better recognize images under different lighting conditions and improve the accuracy of fire image recognition. The present invention introduces a self-attention mechanism into the fire prediction model, so that the model can pay more attention to features that are strongly related to fire prediction, thereby improving the accuracy of fire prediction.

[0056] 3. The present invention can avoid misjudgment of fire monitoring and reduce the misjudgment rate of fire monitoring by analyzing continuous fire monitoring results. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 This is a schematic diagram of the structure of an electric bicycle shed fire extinguishing system based on automatic monitoring, alarm and fire extinguishing adopted in an embodiment of the present invention. DETAILED DESCRIPTION

[0058] In order to enable persons skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0059] Embodiment 1, a method for extinguishing a fire in an electric bicycle shed based on automatic monitoring, alarming and extinguishing, comprising:

[0060] S1. Collect smoke information, temperature information, current information and monitoring pictures of the electric bicycle shed at a set period T;

[0061] The smoke information is obtained by the smoke sensor, the temperature information is obtained by the temperature sensor, the current information is obtained by the current sensor, and the monitoring picture is obtained by the monitoring camera;

[0062] S2. Send smoke information, temperature information and monitoring pictures to the fire monitoring module;

[0063] Send the monitoring picture to the fire picture recognition model in the fire monitoring module for fire picture recognition, and output the monitoring picture recognition results, which include normal, smoke, flame, smoke and flame, and fire hazard;

[0064] The fire image recognition model includes a mode selection layer, an image preprocessing layer and an image recognition layer. The mode selection layer is used to select a day mode or a night mode according to the current time. The day mode sends the current monitoring image to the day channel of the fire image recognition model for recognition, and the night mode sends the current monitoring image to the night channel of the fire image recognition model for recognition.

[0065] The image preprocessing layer is used to preprocess the monitoring image information to obtain the monitoring image data to be identified;

[0066] The image recognition layer is used to classify the surveillance image data to be identified and output the surveillance image recognition results. The surveillance image recognition results include three categories: normal, fire and fire hazard. Fire includes smoke, flame, smoke and flame.

[0067] Establish a fire image recognition model. Specific steps:

[0068] PyTorch was chosen as the deep learning framework. PyTorch uses a dynamic computational graph (real-time execution), and the graph structure is built at runtime. This dynamic nature greatly facilitates the debugging and optimization of the fire image recognition model, and the model can be easily adjusted and expanded.

[0069] A1. Collect several electric bicycle shed images and label them with category labels; pre-process the labeled electric bicycle shed images to obtain a fire image classification dataset; divide the fire image classification dataset into two subsets according to time: a daytime fire image classification dataset and a nighttime fire image classification dataset; split the daytime fire image classification dataset into a daytime fire image classification training set and a daytime fire image test set, and split the nighttime fire image classification dataset into a nighttime fire image classification training set and a nighttime fire image test set;

[0070] When collecting pictures, it is necessary to collect diverse pictures of fire hazards at different times, with and without fire, to improve the generalization of the model;

[0071] Different preprocessing operations are used to simulate the image differences under day and night conditions, and data augmentation technology is used to simulate images under different lighting conditions. For daytime images, the brightness and contrast are randomly adjusted to enhance the robustness of the model to different lighting conditions. For nighttime images, the exposure of the image is increased to improve the image quality in low-light environments.

[0072] A2. Send the daytime fire image classification training set and the nighttime fire image classification training set to the two-stream deep residual network model in batches for alternating training, with the category label as the target. After the training is completed, the initial fire image recognition model is obtained;

[0073] The model architecture is a two-stream deep residual network. The two-stream network processes the image features of daytime and nighttime respectively, and each stream uses a different network structure to adapt to different lighting conditions. In the two-stream network, there is a shared base layer to extract common features, which then branch to their own top layers to process daytime and nighttime-specific features. The two-stream network structure can achieve more accurate classification of images under different time backgrounds. The deep residual network solves the problem of gradient disappearance in the image recognition process by introducing residual blocks, thereby improving the model performance.

[0074] Use alternating training methods to train with daytime and nighttime images to ensure that the model can perform well in both modes.

[0075] The optimizer uses the gradient descent method to minimize the loss function by continuously updating the parameters.

[0076] The loss function is cross entropy loss, which is often used for multi-classification problems.

[0077] Sample weight adjustment: dynamically adjust the weights of samples in different modes based on the performance of the model in daytime mode and nighttime mode to strengthen the training of the model in the mode with poor performance;

[0078] A3. Send the daytime fire image test set and the nighttime fire image test set to the initial fire image recognition model for accuracy evaluation, and use the category label as the target to obtain the test accuracy. If the test accuracy is within the error standard range, the initial fire image recognition model is the fire image recognition model. If the test accuracy exceeds the error standard range, adjust the hyperparameters and repeat steps A2 and A3.

[0079] Independent evaluation: evaluate the model performance on the daytime fire image test set and the nighttime fire image test set to ensure that the model can achieve satisfactory accuracy in both modes;

[0080] The fire monitoring module conducts a comprehensive analysis of smoke information, temperature information and monitoring image recognition results, and outputs fire monitoring results, which include fire occurrence, no fire occurrence and signs of fire;

[0081] The fire monitoring module outputs the fire monitoring results. The specific steps are:

[0082] B1. Obtain smoke information, temperature information and monitoring image recognition results;

[0083] B2. Check whether the smoke information and temperature information exceed the alarm threshold. If both exceed the output, a fire occurs. The alarm threshold is given by professionals.

[0084] If the smoke information exceeds the alarm threshold and the temperature information does not exceed the alarm threshold, check the monitoring image recognition result. If it is flame or flame and smoke, output that a fire has occurred. This step can avoid misjudgment of a fire caused by smoking.

[0085] If the temperature information exceeds the alarm threshold and the smoke information does not exceed the alarm threshold, check the monitoring image recognition result. If it is smoke or flame and smoke, output that a fire has occurred. This step can avoid misjudgment of a fire caused by a local heat source.

[0086] If both the smoke information and the temperature information exceed the alarm threshold, check the monitoring image recognition results. If it is flame, smoke, or flame and smoke, output that there are signs of fire. If the sensor does not alarm, but there are signs of fire, it may be that the image recognition is wrong, but there is still a possibility of fire. It is necessary to further analyze it in combination with the fire prediction results to determine whether there is a fire.

[0087] Otherwise the output is no fire occurred;

[0088] S3. Send all smoke information and temperature information in the current cycle and the previous fixed time period to the fire prediction model for fire prediction, and output the fire prediction results, which include fire prediction and no fire prediction;

[0089] The fire prediction model includes a data processing layer and a fire prediction layer;

[0090] The data processing layer is used to collect weather information and seasonal information, and pre-process the weather information, seasonal information, all smoke information and temperature information in the current cycle and the previous fixed time period to obtain the information to be predicted;

[0091] The fire prediction layer is used to classify the information to be predicted and output the fire prediction results, which include those with fire prediction signs and those without fire prediction signs.

[0092] Establish a fire prediction model, specific steps:

[0093] C1. Collect carport smoke information and carport temperature information of a certain fixed time, as well as the weather information and season information at that time as prediction information, add fire labels to the prediction information, pre-process the prediction information with the fire labels, and obtain a prediction data set; divide the prediction data set into a prediction training set and a prediction test set;

[0094] When collecting prediction information, it is necessary to collect prediction information before the fire occurs and prediction information under normal conditions;

[0095] Data preprocessing: conversion, converting character data (weather information, seasonal information) into numerical data for model processing; removing outliers and filling missing values ​​from forecast information; feature engineering, creating new features, smoke time series data and temperature time series data;

[0096] C2. Send the prediction training set to the LSTM model integrated with the self-attention mechanism for training, with the fire label as the target. After the training is completed, the initial fire prediction model is obtained;

[0097] LSTM (Long Short-Term Memory Network) is selected as the basic model architecture. The LSTM model performs well when processing time series data. LSTM consists of an input gate, a forget gate, and an output gate. The self-attention layer is integrated on the LSTM infrastructure, allowing the model to focus more on features that are strongly related to fire prediction. The self-attention layer is placed in the forget gate of the LSTM to help the model screen out important historical fire features.

[0098] The self-attention mechanism can help the model capture key fire features in the data, thereby improving the accuracy and timeliness of the prediction. In the fire prediction model, the specific steps are to map each element of the prediction information into a query vector Q, a key vector K and a value vector V through a linear transformation. The vectors are transformed into a query vector Q, a key vector K and a value vector V through a weight matrix W. Q , W K , W V Calculated, where W Q , W K , W V is a learnable parameter matrix; calculates the dot product between the query vector Q and the key vector K of the prediction information; the size of the dot product represents the strength of the association between different elements in the sequence; by calculating the dot product, the model can determine which parts of the sequence should be given more attention; the result of the dot product calculation is normalized using the Softmax function to convert it into a probability distribution; this probability distribution represents the attention weight of each element in the sequence to the current element; Softmax ensures that the values ​​of all weights are between 0 and 1, and the sum of all weights is 1; this allows the model to more reasonably allocate the degree of attention to different parts of the sequence; through the self-attention mechanism, the model can learn the weight matrix that best suits the current task, thereby effectively capturing the dependencies in the sequence;

[0099] The optimizer selects the Adam algorithm, which is an optimization algorithm with adaptive learning rate. It combines the characteristics of momentum gradient descent and adaptive learning rate, and can effectively accelerate the convergence speed of the fire prediction model and improve the convergence effect. The fire prediction model is trained through multiple training cycles, and the model parameters are optimized using the batch gradient descent method. In each cycle, the model learns different features in the data and strengthens the learning of key features through the self-attention mechanism. The loss function selects the logarithmic loss function.

[0100] C3. Send the predicted test set to the initial fire prediction model for accuracy evaluation, and take the fire label as the target to obtain the test accuracy; if the test accuracy is within the error standard range, the initial fire prediction model is the fire prediction model; if the test accuracy exceeds the error standard range, adjust the hyperparameters and repeat steps C2 and C3;

[0101] S4. Perform fire hazard detection based on the current information of the current cycle, the current information of the previous cycle and the monitoring image recognition results. The fire hazard detection results include whether there is a fire hazard or not. The specific steps are:

[0102] D1. Obtain the current information of the current cycle, the current information of the previous cycle and the image recognition result;

[0103] D2. Calculate the change C of the current information and determine whether the change C of the current information exceeds the dangerous threshold RTH;

[0104] The danger threshold is given by professionals; the charging power of electric vehicles is relatively stable. If the current changes too much, it usually means a short circuit occurs, which will cause a great fire hazard;

[0105] D3. If the change C of the current information exceeds the danger threshold RTH or the image recognition result is a fire hazard, output that there is a fire hazard, otherwise output that there is no fire hazard;

[0106] S5. Send the fire monitoring results, fire prediction results and fire hazard detection results to the fire analysis module, and output the fire warning level. The fire warning level includes red warning, yellow warning and blue warning. The specific steps are:

[0107] E1. Obtain fire monitoring results, fire prediction results and fire hazard detection results;

[0108] E2. Check the fire monitoring results for three consecutive times. If all of them indicate a fire, a red warning will be output;

[0109] Continuous multiple detections can improve the accuracy of fire monitoring and reduce the rate of misjudgment;

[0110] If there is a fire or signs of fire in three consecutive fire monitoring results, and the fire prediction result shows that there is a fire sign, a red warning will be output;

[0111] If there are signs of fire in three consecutive fire monitoring results and the fire prediction results show that there are signs of fire, a yellow warning will be output;

[0112] If the fire hazard detection is that there is a fire hazard, a blue warning is output.

[0113] S6. Send the fire warning level to the alarm and fire extinguishing module, which implements alarm and fire extinguishing measures according to the fire warning level. A red warning activates the sound alarm and light alarm, and starts the fire extinguishing device; a yellow warning activates the sound alarm and light alarm; a blue warning only activates the light alarm.

[0114] Example 2, electric bicycle shed fire extinguishing system based on automatic monitoring, alarm and fire extinguishing, see Figure 1 As shown, including:

[0115] A data acquisition module is used to collect smoke information, temperature information, current information and monitoring pictures of the electric bicycle shed at a set period T;

[0116] The fire monitoring module is used to deploy a fire image recognition model. The fire image recognition model contains two channels. It switches between the daytime channel and the nighttime channel according to the time to perform fire image recognition on the monitoring images and outputs the monitoring image recognition results. The monitoring image recognition results are used to conduct fire assessment in the direction of the monitoring images. The fire monitoring model conducts a comprehensive analysis of the smoke information, temperature information and monitoring image recognition results and outputs the fire monitoring results. The fire monitoring results are used to assess the real-time fire situation.

[0117] The fire prediction module is used to deploy a fire prediction model. The fire prediction model predicts fires based on smoke information, temperature information, weather information, and season information over a period of time, and outputs fire prediction results. The fire prediction results are used to evaluate the signs of fire occurrence.

[0118] The fire analysis module is used to detect fire hazards based on the current information of the current cycle, the current information of the previous cycle and the monitoring image recognition results, and output the fire hazard detection results; the fire analysis module comprehensively analyzes the fire monitoring results, fire prediction results and fire hazard detection results to obtain the fire warning level;

[0119] The alarm and fire extinguishing module is used to implement alarm and fire extinguishing measures according to the fire warning level. The red warning activates the sound alarm and light alarm, and starts the fire extinguishing device; the yellow warning activates the sound alarm and light alarm; the blue warning only activates the light alarm.

[0120] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all such improvements and changes should fall within the scope of protection of the appended claims of the present invention. Parts not described in detail in this specification belong to the prior art known to those skilled in the art.

Claims

1. A fire extinguishing method for an electric bicycle shed based on automatic monitoring, alarming and fire extinguishing, characterized in that: include: S1. Collect smoke information, temperature information, current information and monitoring pictures of the electric bicycle shed at a set period T; S2. Send smoke information, temperature information and monitoring pictures to the fire monitoring module; Send the monitoring picture to the fire picture recognition model in the fire monitoring module for fire picture recognition, and output the monitoring picture recognition results, which include normal, smoke, flame, smoke and flame, and fire hazard; The fire monitoring module conducts a comprehensive analysis of smoke information, temperature information and monitoring image recognition results, and outputs fire monitoring results, which include fire occurrence, no fire occurrence and signs of fire; S3. Send all smoke information and temperature information in the current cycle and the previous fixed time period to the fire prediction model for fire prediction, and output the fire prediction results, which include fire prediction and no fire prediction; S4. Perform fire hazard detection based on the current information of the current cycle, the current information of the previous cycle and the monitoring image recognition results. The fire hazard detection results include whether there is a fire hazard or not; S5. Send the fire monitoring results, fire prediction results and fire hazard detection results to the fire analysis module, and output the fire warning level, which includes red warning, yellow warning and blue warning; S6. Send the fire warning level to the alarm and fire extinguishing module, which implements alarm and fire extinguishing measures according to the fire warning level. A red warning activates the sound alarm and light alarm, and starts the fire extinguishing device; a yellow warning activates the sound alarm and light alarm; a blue warning only activates the light alarm.

2. The electric bicycle shed fire extinguishing method based on automatic monitoring, alarm and fire extinguishing according to claim 1 is characterized in that: The fire image recognition model includes a mode selection layer, an image preprocessing layer and an image recognition layer. The mode selection layer is used to select a day mode or a night mode according to the current time. The day mode sends the current monitoring image to the day channel of the fire image recognition model for recognition, and the night mode sends the current monitoring image to the night channel of the fire image recognition model for recognition. The image preprocessing layer is used to preprocess the monitoring image information to obtain the monitoring image data to be identified; The image recognition layer is used to classify the monitoring image data to be identified and output the monitoring image recognition results. The monitoring image recognition results include three categories: normal, fire and fire hazard. Fire includes smoke, flame, smoke and flame.

3. The electric bicycle shed fire extinguishing method based on automatic monitoring, alarm and fire extinguishing according to claim 2 is characterized in that: Establish a fire image recognition model. Specific steps: A1. Collect several electric bicycle shed images and label them with category labels; pre-process the labeled electric bicycle shed images to obtain a fire image classification dataset; divide the fire image classification dataset into two subsets according to time: a daytime fire image classification dataset and a nighttime fire image classification dataset; split the daytime fire image classification dataset into a daytime fire image classification training set and a daytime fire image test set, and split the nighttime fire image classification dataset into a nighttime fire image classification training set and a nighttime fire image test set; A2. Send the daytime fire image classification training set and the nighttime fire image classification training set to the two-stream deep residual network model in batches for alternating training, with the category label as the target. After the training is completed, the initial fire image recognition model is obtained; A3. Send the daytime fire image test set and the nighttime fire image test set to the initial fire image recognition model for accuracy evaluation, taking the category label as the target, and obtain the test accuracy; If the test accuracy is within the error standard range, the initial fire image recognition model is the fire image recognition model. If the test accuracy exceeds the error standard range, the hyperparameters are adjusted and steps A2 and A3 are repeated.

4. The electric bicycle shed fire extinguishing method based on automatic monitoring, alarm and fire extinguishing according to claim 1 is characterized in that: The fire monitoring module outputs the fire monitoring results. The specific steps are: B1. Obtain smoke information, temperature information and monitoring image recognition results; B2. Check whether the smoke information and temperature information exceed the alarm threshold. If both exceed the output, a fire occurs. If the smoke information exceeds the alarm threshold and the temperature information does not exceed the alarm threshold, check the monitoring image recognition result. If it is flame or flame and smoke, output that a fire has occurred; If the temperature information exceeds the alarm threshold and the smoke information does not exceed the alarm threshold, check the monitoring image recognition result. If it is smoke or flame and smoke, output that a fire has occurred; If both the smoke information and the temperature information exceed the alarm threshold, check the monitoring image recognition results. If it is flame, smoke, or flame and smoke, output that there is a sign of fire; Otherwise the output does not fire.

5. The electric bicycle shed fire extinguishing method based on automatic monitoring, alarming and fire extinguishing according to claim 1 is characterized in that: The fire prediction model includes a data processing layer and a fire prediction layer; The data processing layer is used to collect weather information and seasonal information, and pre-process the weather information, seasonal information, all smoke information and temperature information in the current cycle and the previous fixed time period to obtain the information to be predicted; The fire prediction layer is used to classify the information to be predicted and output the fire prediction results, which include those with fire prediction and those without fire prediction.

6. The electric bicycle shed fire extinguishing method based on automatic monitoring, alarming and fire extinguishing according to claim 5 is characterized in that: Establish a fire prediction model, specific steps: C1. Collect carport smoke information and carport temperature information of a certain fixed time, as well as the weather information and season information at that time as prediction information, add fire labels to the prediction information, pre-process the prediction information with the fire labels, and obtain a prediction data set; divide the prediction data set into a prediction training set and a prediction test set; C2. Send the prediction training set to the LSTM model integrated with the self-attention mechanism for training, with the fire label as the target. After the training is completed, the initial fire prediction model is obtained; C3. Send the predicted test set to the initial fire prediction model for accuracy evaluation, and take the fire label as the target to obtain the test accuracy; if the test accuracy is within the error standard range, the initial fire prediction model is the fire prediction model. If the test accuracy exceeds the error standard range, adjust the hyperparameters and repeat steps C2 and C3.

7. The electric bicycle shed fire extinguishing method based on automatic monitoring, alarming and fire extinguishing according to claim 1 is characterized in that: Detect fire hazards, specific steps: D1. Obtain the current information of the current cycle, the current information of the previous cycle and the image recognition result; D2. Calculate the change C of the current information and determine whether the change C of the current information exceeds the dangerous threshold RTH; D3. If the change C of the current information exceeds the danger threshold RTH or the image recognition result is a fire hazard, output that there is a fire hazard, otherwise output that there is no fire hazard.

8. The electric bicycle shed fire extinguishing method based on automatic monitoring, alarming and fire extinguishing according to claim 1 is characterized in that: Fire analysis module, specific steps: E1. Obtain fire monitoring results, fire prediction results and fire hazard detection results; E2. Check the fire monitoring results for three consecutive times. If all of them indicate a fire, a red warning will be output; If there is a fire or signs of fire in three consecutive fire monitoring results, and the fire prediction result shows that there is a fire sign, a red warning will be output; If there are signs of fire in three consecutive fire monitoring results and the fire prediction results show that there are signs of fire, a yellow warning will be output; If the fire hazard detection is that there is a fire hazard, a blue warning is output.

9. The electric bicycle shed fire extinguishing system based on automatic monitoring, alarm and fire extinguishing is characterized by: The system applies the electric bicycle shed fire extinguishing method based on automatic monitoring, alarm and fire extinguishing as described in any one of claims 1 to 8, including: A data acquisition module is used to collect smoke information, temperature information, current information and monitoring pictures of the electric bicycle shed at a set period T; The fire monitoring module is used to deploy a fire image recognition model. The fire image recognition model contains two channels. It switches between the daytime channel and the nighttime channel according to the time to perform fire image recognition on the monitoring images and outputs the monitoring image recognition results. The monitoring image recognition results are used to conduct fire assessment in the direction of the monitoring images. The fire monitoring model conducts a comprehensive analysis of the smoke information, temperature information and monitoring image recognition results and outputs the fire monitoring results. The fire monitoring results are used to assess the real-time fire situation. The fire prediction module is used to deploy a fire prediction model. The fire prediction model predicts fires based on smoke information, temperature information, weather information, and season information over a period of time, and outputs fire prediction results. The fire prediction results are used to evaluate the signs of fire occurrence. The fire analysis module is used to detect fire hazards based on the current information of the current cycle, the current information of the previous cycle and the monitoring image recognition results, and output the fire hazard detection results; the fire analysis module comprehensively analyzes the fire monitoring results, fire prediction results and fire hazard detection results to obtain the fire warning level, which includes red warning, yellow warning and blue warning; The alarm and fire extinguishing module is used to implement alarm and fire extinguishing measures according to the fire warning level. The red warning activates the sound alarm and light alarm, and starts the fire extinguishing device; the yellow warning activates the sound alarm and light alarm; the blue warning only activates the light alarm.

Citation Information

Cited By

  • Alarm and fire extinguishing system based on charging shed fire automatic monitoring

    CN120437523A