Fire remote monitoring and multistage early warning mechanism integrated system

Abstract

The invention relates to the technical field of fire hazard monitoring, and discloses a fire hazard remote monitoring and multistage early warning mechanism integration system which integrates a data acquisition module, a heat gathering map generation module, an image fusion module, a data calculation module, a risk prediction module, an early warning notification module and the like. Through an infrared temperature measurement unit, a smoke concentration acquisition unit and an image acquisition unit, the system can acquire temperature, smoke and image information of a key area in real time. A heat gathering map generation and image fusion technology is utilized, a fusion image is systematically generated, a fire scene and temperature distribution thereof are visually displayed, and the fire source positioning efficiency is improved. Meanwhile, the system analyzes the gradient change of the temperature, the heat accumulation area and the smoke concentration through a data calculation module, constructs a risk prediction model in combination with a machine learning algorithm, and achieves the refined prediction of the fire risk. And according to a prediction result, the early warning notification module automatically sends a corresponding early warning message to ensure rapid transmission and timely response of the early warning message. According to the invention, the intelligent, precise and efficient level of fire monitoring is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of fire monitoring, and in particular to a fire remote monitoring and multi-level early warning mechanism integrated system. Background Art

[0002] In modern urban construction and industrial production, fire, as a sudden and destructive disaster, poses a serious threat to people's lives and property. In order to effectively prevent and respond to fires in a timely manner, various fire monitoring systems have emerged.

[0003] Existing fire monitoring systems mainly rely on single sensor data output, such as judging fire risks only by numerical changes in temperature or smoke concentration. Although this method can provide certain early warning information, it cannot intuitively display the actual physical space conditions and temperature distribution of the fire scene. Especially in complex environments, such as large warehouses and factory workshops, it is difficult to quickly locate the fire source or high temperature area based on numerical data alone, resulting in low emergency response efficiency.

[0004] Moreover, most existing fire monitoring systems use simple threshold judgments or experience-based rules for early warning, lacking a scientific risk prediction model. This method cannot accurately reflect the dynamic changes in fire risk, nor can it be flexibly adjusted according to different scenarios and conditions. For example, in high-risk areas such as chemical plants, due to differences in raw material properties, production processes and other factors, the fire risk level and early warning standards should be different. However, traditional monitoring systems cannot make refined predictions based on these specific factors, resulting in inaccurate or delayed early warning information, which poses a great challenge to fire prevention and control work. Summary of the invention

[0005] The purpose of the present invention is to provide a fire remote monitoring and multi-level early warning mechanism integrated system to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: a fire remote monitoring and multi-level early warning mechanism integrated system, the system comprising:

[0007] The data acquisition module includes an infrared temperature measurement unit, a smoke concentration acquisition unit and an image acquisition unit, which are used to regularly collect spatial temperature data, smoke concentration and real-time images of key areas;

[0008] The heat map generation module is used to convert the collected spatial temperature data into image form, and use color mapping to display temperature differences and generate temperature distribution images;

[0009] An image fusion module is used to superimpose the temperature distribution image with the real-time image of the key area to generate a fused image, and the fused image is used to simultaneously observe the actual physical space and temperature distribution of the key area;

[0010] The data calculation module includes a spatial temperature gradient change calculation unit, a heat collection area gradient change calculation unit and a smoke concentration gradient change calculation unit, which are responsible for analyzing and calculating the gradient change trends of the spatial temperature, heat collection area and smoke concentration respectively;

[0011] A risk prediction module is used to construct a risk prediction model, which uses the gradient change trend of each indicator output by the data calculation module and combines it with a machine learning algorithm to predict the risk level of a fire;

[0012] The early warning notification module automatically triggers and sends corresponding early warning notification messages according to the fire risk level obtained by the risk prediction module.

[0013] Preferably, the steps of generating a temperature distribution image by the heat map generation module include:

[0014] S101: receiving spatial temperature data transmitted by a data acquisition module, the data being a set of temperature values ​​of each measurement point in a key area, expressed as {T1, T2, ..., Tn}, where n is the number of measurement points;

[0015] S102: preprocessing the received temperature data, including data cleaning to remove abnormal values ​​and noise, and data normalization to map the temperature value into a predetermined value range;

[0016] S103: using a color mapping algorithm to convert the normalized temperature value into a corresponding color value to generate a color mapping table; the color mapping algorithm uses a linear interpolation method to assign different colors according to different temperature value intervals;

[0017] S104: constructing a two-dimensional grid model according to the physical layout of the key area and the location information of the measurement points, and mapping the temperature value and the corresponding color value of each measurement point to a corresponding position of the grid model;

[0018] S105: using an OpenCV-based image rendering algorithm to convert the two-dimensional grid model and the color mapping thereon into an image form, and generate a temperature distribution image; the temperature distribution image represents the high and low temperatures by color depth or hue changes.

[0019] Preferably, the steps of generating a fused image by the image fusion module include:

[0020] S201: Receive a temperature distribution map image transmitted by a heat map generation module, where the image is a two-dimensional image showing the temperature distribution of a key area in color mapping, represented as T_map, where the color value of each pixel corresponds to the temperature value of the point;

[0021] S202: simultaneously receiving a real-time image of the key area transmitted by an image acquisition unit in the data acquisition module, denoted as R_image, which is an actual physical space image of the key area;

[0022] S203: performing image preprocessing on the temperature distribution map image T_map and the real-time image R_image, including adjusting the image sizes to make them match, and performing image enhancement processing to improve image quality and contrast;

[0023] S204: Using an image fusion algorithm, the preprocessed temperature distribution map image T_map is superimposed on the real-time image R_image to generate a fused image F_image; the image fusion algorithm uses a transparency blending method, and its formula is:

[0024] F_image(x,y)=α*T_map(x,y)+(1-α)*R_image(x,y)

[0025] Among them, F_image(x,y) represents the pixel value of the fused image at the coordinate (x,y), T_map(x,y) represents the pixel value of the temperature distribution map image at the coordinate (x,y), R_image(x,y) represents the pixel value of the real-time image at the coordinate (x,y), and α is the weight coefficient, ranging from 0 to 1, which is used to adjust the proportion of the temperature distribution map image and the real-time image in the fused image;

[0026] S205: Post-processing the generated fused image F_image, including image smoothing, to eliminate artifacts and noise generated during the fusion process.

[0027] Preferably, the spatial temperature gradient change calculation unit obtains the temperature data of each measurement point in the key area from the data acquisition module, and calculates the spatial temperature gradient change using the gradient calculation formula. The specific formula is: Where T1 and T2 are the temperature values ​​of adjacent measurement points, and d is the distance between the two points.

[0028] Preferably, the calculation method of the heat collection area gradient change calculation unit is:

[0029] Obtain the fused image from the image fusion module, use the image recognition algorithm to identify the high temperature or flame area from the fused image, and calculate its area, expressed as A(t), where t is the timestamp;

[0030] The heat collection area data at consecutive time points are processed to calculate the gradient change of the heat collection area. The specific formula is: dA / dt=(A(t2)-A(t1)) / (t2-t1), where A(t1) and A(t2) are the heat collection areas at adjacent time points, respectively.

[0031] Preferably, the smoke concentration gradient change calculation unit obtains the smoke concentration data of each measuring point in the key area from the data acquisition module, expressed as {C1, C2, ..., Cm}, where m is the number of measuring points; and calculates the smoke concentration gradient change using the gradient calculation formula, the specific formula is: Where C1 and C2 are the smoke concentration values ​​of adjacent measurement points, and d is the distance between the two points.

[0032] Preferably, the risk prediction module uses a linear regression algorithm to construct a risk prediction model, and the specific steps include:

[0033] S301: Obtaining calculation results of spatial temperature gradient change, heat accumulation area gradient change and smoke concentration gradient change from a data calculation module as input features of a risk prediction model;

[0034] S302: Preprocessing the input features, including data cleaning and standardization;

[0035] S303: Construct a linear regression model to establish a linear relationship between the input features and the fire risk level; assuming that the fire risk level is Y, the spatial temperature gradient change is X1, the heat accumulation area gradient change is X2, and the smoke concentration gradient change is X3, the linear regression model is expressed as:

[0036] Y=β0+β1*X1+β2*X2+β3*X3+ε

[0037] Among them, β0 is the intercept term, β1, β2, β3 are the regression coefficients of each input feature, and ε is the error term;

[0038] S304: using historical fire data to train the linear regression model, solving the regression coefficients β0, β1, β2 and β3 by minimizing the loss function, and using the gradient descent method to update the values ​​of β0, β1, β2 and β3 during the training process to obtain the optimal solution of the loss function;

[0039] S305: Verify and evaluate the trained linear regression model, including calculating the accuracy, recall and F1 score indicators of the model;

[0040] S306: Deploy the verified linear regression model to the risk prediction module for real-time prediction of the fire risk level.

[0041] Preferably, in step S302, the data is standardized using the Z-score method, where each numerical data is subtracted from its mean and divided by its standard deviation, so that the processed data conforms to the standard normal distribution. The formula for the standardized processing is: Among them, X is the original data, μ is the mean of the original data, σ is the standard deviation of the original data, and Z is the standardized data.

[0042] Preferably, in step 304, the specific method of updating the regression coefficient is:

[0043] Step 1: Set the predicted value of the linear regression model as Y_pred, the actual fire risk level as Y_true, and the calculation formula of the mean square error MSE is: MSE = (1 / n) * Σ (Y_pred-Y_true)^2, where n is the number of samples and Σ represents the sum of all samples;

[0044] Step 2: In order to minimize the MSE, the regression coefficient β of the linear regression model is iteratively updated using the gradient descent method, and the initial regression coefficients β0, β1, β2, β3, and the learning rate η are set;

[0045] Step 3: Calculate the partial derivative of MSE with respect to each regression coefficient β. For each regression coefficient βi, the gradient calculation formula is:

[0046]

[0047] in, It is the partial derivative of the predicted value Y_pred with respect to the regression coefficient βi, which is obtained according to the formula of the linear regression model;

[0048] Step 4: Update each regression coefficient βi according to the gradient and learning rate η. The update formula is:

[0049]

[0050] Among them, βi_old is the old regression coefficient, and βi_new is the updated regression coefficient;

[0051] Step 5: Repeat steps 3 and 4 until the MSE converges to the preset threshold.

[0052] Preferably, the risk prediction module divides the fire risk level into multiple levels according to the output results of the risk prediction model, including low risk, medium risk, high risk and extremely high risk, and each level corresponds to different early warning notification strategies and emergency response measures.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] Through the image fusion module, the present invention can accurately superimpose the temperature distribution image with the real-time image of the key area to generate a fused image, so that monitoring personnel can simultaneously observe the actual physical space and its temperature distribution conditions at the fire scene, intuitively display the fire situation, greatly improve the efficiency of fire source positioning and high-temperature area identification, and help to quickly take emergency response measures.

[0055] The present invention introduces a risk prediction module, which utilizes the spatial temperature gradient change, heat collection area gradient change and smoke concentration gradient change trends output by the data calculation module, combined with a machine learning algorithm, to construct an accurate risk prediction model. The risk prediction model can dynamically reflect the real-time changes in fire risk, and can be flexibly adjusted according to different scenarios and conditions, thereby achieving refined prediction of fire risk and improving the accuracy and timeliness of early warning information.

[0056] Through the early warning notification module, the present invention can automatically trigger and send corresponding early warning notification messages according to the fire risk level obtained by the risk prediction module, ensuring the rapid transmission of early warning information, so that relevant departments and personnel can take response measures in time, effectively control the fire, and reduce the losses caused by the fire. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 It is the overall structure diagram of the present invention;

[0058] Figure 2 Diagram of the steps for fusion image generation;

[0059] Figure 3 This is the training flow chart of the risk prediction model;

[0060] Figure 4 This is a diagram of the working principle of the MSE loss function. DETAILED DESCRIPTION

[0061] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0062] See also Figure 1-4 The present invention provides a technical solution: a fire remote monitoring and multi-level early warning mechanism integrated system, the system comprising:

[0063] Data acquisition module: The data acquisition module consists of an infrared temperature measurement unit, a smoke concentration acquisition unit, and an image acquisition unit. The infrared temperature measurement unit uses a high-precision infrared thermal imager to scan key areas at regular intervals to obtain spatial temperature data; the smoke concentration acquisition unit uses smoke sensors arranged in key areas to collect real-time information on smoke concentration in the air; the image acquisition unit uses a high-definition camera to capture real-time images of key areas. Each acquisition unit is connected to the central processor, and transmits the collected data to the central processor for subsequent processing at preset time intervals.

[0064] Heat map generation module: The heat map generation module receives the temperature data transmitted by the infrared temperature measurement unit and converts the temperature data into an image using a professional image processing algorithm. Through color mapping technology, different temperature values ​​are mapped to different colors to generate an intuitive temperature distribution image. This image can clearly show the temperature differences and distribution in key areas, facilitating subsequent analysis.

[0065] Image fusion module: The image fusion module receives the temperature distribution image and the real-time image, and superimposes the two through the image fusion algorithm. During the fusion process, the spatial coordinates of the temperature distribution image and the real-time image are ensured to be consistent, so that the generated fusion image can simultaneously display the actual physical space of the key area and its corresponding temperature distribution.

[0066] Data calculation module: The data calculation module includes a spatial temperature gradient change calculation unit, a heat collection area gradient change calculation unit, and a smoke concentration gradient change calculation unit. Each calculation unit receives the data transmitted by the heat collection map generation module and the smoke concentration acquisition unit, and uses mathematical analysis methods to calculate the change trend of spatial temperature, heat collection area, and smoke concentration over time (i.e., gradient change).

[0067] Risk prediction module: The risk prediction module builds a risk prediction model based on machine learning algorithms. The model receives the gradient change trend of each indicator output by the data calculation module, combines historical fire data and expert knowledge, and predicts the fire risk level through training and optimization. The prediction results are output in the form of numerical values ​​or levels to provide decision-making basis for the early warning notification module.

[0068] Early warning notification module: The early warning notification module automatically triggers the corresponding early warning notification mechanism according to the fire risk level output by the risk prediction module. When the risk level reaches the preset threshold, the early warning notification module promptly sends early warning notification messages to monitoring personnel and relevant personnel through SMS, email, sound and light alarms, etc. The early warning notification message contains information such as the fire risk level, the location of key areas, and recommended measures, so that relevant personnel can respond quickly and take necessary fire prevention measures.

[0069] The present invention will be further described below in conjunction with Examples 1 to 4:

[0070] Embodiment 1:

[0071] The implementation of the heat map generation module mainly includes the following steps:

[0072] S101: The module receives spatial temperature data from the data acquisition module. These data are obtained by regularly collecting data from key areas, including the temperature values ​​of each measurement point. Taking a warehouse as an example, assuming that 100 temperature sensors are arranged in the warehouse, the heat map generation module will receive the temperature data provided by these 100 sensors, expressed as {T1, T2, ..., T100}.

[0073] S102: Preprocess the received temperature data, including data cleaning and data normalization. Data cleaning is to remove outliers and noise caused by sensor failure or environmental factors to ensure data accuracy. Data normalization is to map the temperature value to a predetermined range of values, such as 0 to 1, to facilitate subsequent color mapping. For example, if the temperature range in the warehouse is between -20°C and 50°C, then this range can be linearly mapped to between 0 and 1.

[0074] S103: Convert the normalized temperature value to a corresponding color value using a color mapping algorithm. The color mapping algorithm may use a linear interpolation method or a piecewise function method. Taking the linear interpolation method as an example, the low temperature end may correspond to blue, the high temperature end may correspond to red, and the intermediate temperature may be assigned a corresponding color according to the linear interpolation method. Each temperature value will correspond to a unique color value, thereby generating a color mapping table.

[0075] S104: A two-dimensional grid model is constructed based on the physical layout of the warehouse and the location information of the sensor. This model divides the warehouse into multiple small grids, each grid corresponding to a sensor and its temperature value. Then, the temperature value and corresponding color value of each sensor are mapped to the corresponding position of the grid model.

[0076] S105: The two-dimensional grid model and the color map on it are converted into an image form by using an image rendering algorithm to generate a heat map, which can be implemented by using image processing libraries such as OpenCV and Matplotlib. The generated heat map will represent the high and low temperatures by color depth or hue changes, thereby intuitively showing the temperature distribution in the warehouse. For example, a high temperature area may be displayed in red, a low temperature area in blue, and an intermediate temperature in a color between red and blue.

[0077] The following describes in detail the implementation of the image fusion module with specific examples:

[0078] First, the heat map image T_map and the real-time image R_image are preprocessed. Assume that the resolution of the heat map image T_map is 640x480, and the resolution of the real-time image R_image is 1280x720. In order to overlay the two, the resolution of the heat map image T_map needs to be adjusted to the same as the real-time image R_image. This can be achieved by an image interpolation algorithm (such as bilinear interpolation or nearest neighbor interpolation). The interpolated heat map image is recorded as T_map_resized.

[0079] Perform image enhancement on T_map_resized and R_image, including histogram equalization to improve image contrast, or Gaussian filtering to smooth the image and remove noise.

[0080] The transparency blending method is used for image fusion. The weight coefficient α is set to 0.5, indicating that the heat map image and the real-time image each account for 50% of the weight in the fused image. For each pixel point (x, y) in the fused image F_image, its pixel value is calculated by the following formula:

[0081] F_image(x,y)=α*T_map_resized(x,y)+(1-α)*R_image(x,y)

[0082] It should be noted here that since T_map_resized and R_image are both color images, the above formula actually calculates each color channel (red, green, and blue) separately.

[0083] The generated fused image F_image is post-processed. First, the brightness and contrast of the image are adjusted to ensure that the fused image has a suitable visual effect when displayed. Then, an image smoothing algorithm such as median filtering or bilateral filtering is used to eliminate artifacts and noise that may be generated during the fusion process.

[0084] The processed fusion image F_image can be output to a display terminal (such as a display screen in a monitoring center) or a storage device (such as a hard disk or cloud storage). In this way, monitoring personnel can simultaneously observe the actual physical space and its temperature distribution in the key area, thereby more accurately judging the fire risk.

[0085] Embodiment 2:

[0086] The gradient change calculation module of the present invention includes a spatial temperature gradient change calculation unit, a heat collection area gradient change calculation unit and a smoke concentration gradient change calculation unit. The implementation of each unit will be described in detail below with reference to specific examples.

[0087] ① Implementation method of spatial temperature gradient change calculation unit:

[0088] Take a warehouse as an example. Multiple temperature sensors are deployed in the warehouse to monitor the temperature of each point in the warehouse in real time. The specific implementation steps are as follows:

[0089] Data acquisition: The temperature data of each measuring point in the warehouse is obtained from the data acquisition module, such as {T1=25℃, T2=26℃,..., Tn=30℃}.

[0090] Data preprocessing: Clean the acquired temperature data to remove outliers caused by sensor failure or environmental factors to ensure the accuracy and reliability of the data.

[0091] Constructing a temperature distribution model: Based on the location information of the measurement points, such as (x1, y1), (x2, y2), ..., (xn, yn), a spatial temperature distribution model in the warehouse is constructed. The temperature data is mapped to the corresponding positions in the model to form a temperature distribution graph.

[0092] Calculate the temperature gradient: Use the gradient calculation formula, such as Calculate the temperature gradient between adjacent measurement points. For multi-dimensional space, the partial derivative calculation formula can be used for calculation. For example, if the temperature of two adjacent points is T1 = 25 ° C and T2 = 26 ° C, and the distance is d = 1m, then the temperature gradient is

[0093] Output results: The calculated temperature gradient change results are output to the risk prediction module for subsequent risk assessment and early warning analysis.

[0094] ② Implementation method of heat collection area gradient change calculation unit:

[0095] Taking the images captured by the surveillance cameras in the warehouse as an example, the specific implementation steps are as follows:

[0096] Image acquisition: The fused image is obtained from the image fusion module, which simultaneously shows the actual physical space in the warehouse and its temperature distribution.

[0097] Image recognition: Use image recognition algorithms, such as convolutional neural networks (CNNs), to identify high-temperature or flame areas from the fused image and calculate their area A(t), where t is the timestamp.

[0098] Calculate the heat accumulation area gradient: process the heat accumulation area data at consecutive time points, such as A(t1) = 10m 2 and A(t2)=15m 2 , time is t2-t1=5s, then the heat accumulation area gradient is dA / dt=(15-10) / 5=1m 2 / s.

[0099] Output results: The calculated heat accumulation area gradient change results are output to the risk prediction module to evaluate the fire spread speed and risk level.

[0100] ③ Implementation method of smoke concentration gradient change calculation unit:

[0101] Taking the smoke sensor in the warehouse as an example, the specific implementation steps are as follows:

[0102] Data collection: The smoke concentration data of each measuring point in the warehouse is obtained from the data collection module, such as {C1=0.5mg / m 3 ,C2=0.6mg / m 3 ,...,Cm=1.0mg / m 3}.

[0103] Data preprocessing: Clean the acquired smoke concentration data to remove outliers caused by sensor failure or environmental factors.

[0104] Constructing a smoke concentration distribution model: Based on the location information of the measurement points, a smoke concentration distribution model is constructed in the warehouse. The smoke concentration data is mapped to the corresponding positions in the model to form a smoke concentration distribution map.

[0105] Calculate the smoke concentration gradient: Use the gradient calculation formula, such as Calculate the smoke concentration gradient between adjacent measurement points. For example, if the smoke concentration between two adjacent points is C1 = 0.5 mg / m 3 and C2 = 0.6 mg / m 3 , the distance is d = 1m, then the smoke concentration gradient is

[0106] Output results: The calculated smoke concentration gradient change results are output to the risk prediction module to monitor the diffusion of smoke and assess fire risks.

[0107] Embodiment 3:

[0108] The risk prediction module uses a linear regression algorithm to build a risk prediction model to accurately predict the fire risk level. The implementation steps include:

[0109] S301: Obtain calculation results of spatial temperature gradient change, heat accumulation area gradient change and smoke concentration gradient change from the data calculation module as input features of the risk prediction model.

[0110] S302: Clean the input features, remove outliers and missing values, and ensure the accuracy and completeness of the data. At the same time, standardize the temperature data and smoke concentration data to eliminate the impact of different dimensional features on model training.

[0111] The Z-score method is used to standardize the data. The mean of each numerical data is subtracted and divided by its standard deviation so that the processed data conforms to the standard normal distribution. The standardization formula is: Z = (X-μ) / σ, where X is the original data, μ is the mean of the original data, σ is the standard deviation of the original data, and Z is the standardized data.

[0112] S303: Take the spatial temperature gradient change, the heat collection area gradient change and the smoke concentration gradient change as input features and the fire risk level as the output target to construct a linear regression model. Assume that the fire risk level is Y, the spatial temperature gradient change is X1, the heat collection area gradient change is X2, and the smoke concentration gradient change is X3, then the linear regression model can be expressed as: Y = β0 + β1*X1 + β2*X2 + β3*X3 + ε, where β0 is the intercept term, β1, β2, β3 are the regression coefficients of each input feature, and ε is the error term.

[0113] S304: Use historical fire data to train the linear regression model. Solve the regression coefficients β0, β1, β2, and β3 by minimizing the loss function (such as mean square error). During the training process, optimization algorithms such as gradient descent and least squares method can be used to iteratively update the model parameters until convergence conditions are reached.

[0114] S305: Verify and evaluate the trained linear regression model, including calculating the accuracy, recall and F1 score indicators of the model;

[0115] S306: Deploy the verified linear regression model to the risk prediction module for real-time prediction of the fire risk level.

[0116] The risk prediction module uses mean squared error (MSE) as the loss function and iteratively updates the regression coefficients of the linear regression model through the gradient descent method to minimize the MSE process.

[0117] First, the predicted value of the linear regression model is set as Y_pred, which is obtained by linearly combining the spatial temperature gradient change X1, the heat accumulation area gradient change X2, and the smoke concentration gradient change X3 as input features with the regression coefficients β0, β1, β2, and β3. The actual fire risk level is Y_true, which is known historical data.

[0118] Calculate the mean square error MSE, the formula is:

[0119] MSE=(1 / n)*Σ(Y_pred-Y_true)^2

[0120] Where n is the number of samples, that is, the number of fire events in the historical data; Σ represents the sum of all samples. MSE measures the mean square error between the model prediction value and the actual value, which is the goal to be minimized.

[0121] In order to minimize the MSE, the gradient descent method is used. First, the initial regression coefficients β0, β1, β2, and β3 are set. These values ​​can be random or based on prior knowledge. At the same time, the learning rate η is set, which determines the step size of updating the regression coefficients in each iteration.

[0122] Calculate the partial derivative of MSE for each regression coefficient β, that is, the gradient. For each regression coefficient βi, the gradient calculation formula is:

[0123]

[0124] in, is the partial derivative of the predicted value Y_pred with respect to the regression coefficient βi. Since Y_pred is a linear combination of βi, is the corresponding input eigenvalue (one of X1, X2, X3).

[0125] Update each regression coefficient βi according to the gradient and learning rate η. The update formula is:

[0126]

[0127] Among them, βi_old is the old regression coefficient, and βi_new is the updated regression coefficient.

[0128] Repeat the above steps of calculating the gradient and updating the regression coefficient until the MSE converges to a preset threshold or reaches a preset number of iterations. The convergence criterion can be that the change in the MSE is less than a certain small value, or reaches a preset maximum number of iterations.

[0129] Finally, the linear regression model obtained after minimizing the MSE is output. This model can predict the risk level of fire based on the new spatial temperature gradient change, heat accumulation area gradient change and smoke concentration gradient change data.

[0130] Assume that there are 100 historical fire event data, each event contains the values ​​of spatial temperature gradient change, heat accumulation area gradient change and smoke concentration gradient change, as well as the corresponding fire risk level. Set the initial regression coefficients to β0 = 0, β1 = 0.1, β2 = 0.1, β3 = 0.1, and the learning rate to η = 0.01.

[0131] Then, follow the above steps to calculate the MSE and gradient, and update the regression coefficients. After multiple iterations, the MSE gradually converges to a smaller value, such as 0.01. At this point, the regression coefficients obtained may be β0 = 0.5, β1 = 0.3, β2 = 0.2, β3 = 0.4. These regression coefficients constitute a linear regression model that can be used to predict the risk level of new fire events.

[0132] When a new fire event occurs, the data of spatial temperature gradient change, heat accumulation area gradient change and smoke concentration gradient change are collected and substituted into the linear regression model to obtain the predicted fire risk level. According to this risk level, the corresponding early warning notification message can be triggered so that timely measures can be taken to prevent the occurrence or expansion of the fire.

[0133] Embodiment 4:

[0134] The following describes in detail the implementation of the fire level design of the present invention in combination with specific examples:

[0135] ① Determine the threshold range of key indicators

[0136] Spatial temperature gradient change: Based on historical fire data and expert experience, set the threshold range of spatial temperature gradient change. For example, when the temperature gradient change is less than 0.5℃ / m, the risk is considered low; when the temperature gradient change is between 0.5℃ / m and 1.5℃ / m, the risk is medium; when the temperature gradient change is greater than 1.5℃ / m, the risk is high.

[0137] Heat collection area gradient change: Based on historical data and expert opinions, set the threshold of heat collection area gradient change. For example, when the heat collection area gradient change is less than 0.1m 2 / s, the risk is low; when the change is 0.1m 2 / s to 0.3m 2 / s, the risk is medium; when the change is greater than 0.3m 2 / s, the risk is higher.

[0138] Smoke density gradient change: For the smoke density gradient change, set the corresponding threshold. For example, when the smoke density gradient change is less than 0.01mg / m 3 / m, the risk is low; when the change is 0.01mg / m 3 / m to 0.05mg / m 3 / m, the risk is moderate; when the change is greater than 0.05mg / m 3 / m, the risk is higher.

[0139] ②Risk level scoring and comprehensive

[0140] Assign risk level score: For each indicator, assign a risk level score based on its threshold range. For example, low risk is 1 point, medium risk is 3 points, and high risk is 5 points.

[0141] Comprehensive score: The risk level scores of the three indicators are combined using the weighted average method. Assuming that the weights of the spatial temperature gradient change, the heat accumulation area gradient change, and the smoke concentration gradient change are 0.4, 0.3, and 0.3 respectively, the comprehensive risk score is:

[0142] Comprehensive risk score = 0.4*spatial temperature gradient change score + 0.3*heat accumulation area gradient change score + 0.3*smoke concentration gradient change score.

[0143] ③ Classification of fire risk levels

[0144] Set level range: Based on the comprehensive risk score, the fire risk level is divided into four levels: low risk (1-2 points), medium risk (3-4 points), high risk (5-6 points) and extremely high risk (7-9 points).

[0145] Early warning notification and emergency response: For different risk levels, set corresponding early warning notification strategies and emergency response measures. For example, only send reminder information when the risk is low; send early warning information and initiate preliminary emergency response when the risk is medium; send emergency early warning information and initiate comprehensive emergency response when the risk is high; and immediately trigger the alarm system and initiate the highest level of emergency response when the risk is extremely high.

[0146] ④Dynamic adjustment

[0147] The fire risk rating design method has the ability to be dynamically adjusted. When the environmental factors, time period or fire history data in the monitored area changes, the threshold range, weight and comprehensive score calculation method of the key indicators are adjusted accordingly to ensure the accuracy and effectiveness of the fire risk rating.

[0148] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0149] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. Fire remote monitoring and multi-level early warning mechanism integrated system, characterized by: The system comprises: The data acquisition module includes an infrared temperature measurement unit, a smoke concentration acquisition unit and an image acquisition unit, which are used to regularly collect spatial temperature data, smoke concentration and real-time images of key areas; The heat map generation module is used to convert the collected spatial temperature data into image form, and use color mapping to display temperature differences and generate temperature distribution images; An image fusion module is used to superimpose the temperature distribution image with the real-time image of the key area to generate a fused image, and the fused image is used to simultaneously observe the actual physical space and temperature distribution of the key area; The data calculation module includes a spatial temperature gradient change calculation unit, a heat collection area gradient change calculation unit and a smoke concentration gradient change calculation unit, which are responsible for analyzing and calculating the gradient change trends of the spatial temperature, heat collection area and smoke concentration respectively; The risk prediction module is used to construct a risk prediction model, which uses the gradient change trend of each indicator output by the data calculation module and combines the machine learning algorithm to predict the risk level of the fire; The early warning notification module automatically triggers and sends corresponding early warning notification messages according to the fire risk level obtained by the risk prediction module.

2. The fire remote monitoring and multi-level early warning mechanism integrated system according to claim 1 is characterized in that: The steps of generating a temperature distribution image by the heat map generation module include: S101: receiving spatial temperature data transmitted by a data acquisition module, the data being a set of temperature values ​​of each measurement point in a key area, expressed as {T1, T2, ..., Tn}, where n is the number of measurement points; S102: preprocessing the received temperature data, including data cleaning to remove abnormal values ​​and noise, and data normalization to map the temperature value into a predetermined value range; S103: using a color mapping algorithm to convert the normalized temperature value into a corresponding color value to generate a color mapping table; the color mapping algorithm uses a linear interpolation method to assign different colors according to different temperature value intervals; S104: constructing a two-dimensional grid model according to the physical layout of the key area and the location information of the measurement points, and mapping the temperature value and the corresponding color value of each measurement point to a corresponding position of the grid model; S105: using an OpenCV-based image rendering algorithm to convert the two-dimensional grid model and the color mapping thereon into an image form, and generate a temperature distribution image; the temperature distribution image represents the high and low temperatures by color depth or hue changes.

3. The fire remote monitoring and multi-level early warning mechanism integrated system according to claim 2 is characterized in that: The steps of generating a fused image by the image fusion module include: S201: receiving a temperature distribution map image transmitted by a heat map generation module, the image being a two-dimensional image showing the temperature distribution of a key area in color mapping, denoted as T_map, where the color value of each pixel corresponds to the temperature value of the point; S202: simultaneously receiving a real-time image of the key area transmitted by an image acquisition unit in the data acquisition module, denoted as R_image, which is an actual physical space image of the key area; S203: performing image preprocessing on the temperature distribution map image T_map and the real-time image R_image, including adjusting the image sizes to make them match, and performing image enhancement processing to improve image quality and contrast; S204: Using an image fusion algorithm, the preprocessed temperature distribution map image T_map is superimposed on the real-time image R_image to generate a fused image F_image; the image fusion algorithm uses a transparency blending method, and its formula is: F_image(x,y)=α*T_map(x,y)+(1-α)*R_image(x,y) Wherein, F_image(x,y) represents the pixel value of the fused image at the coordinate (x,y), T_map(x,y) represents the pixel value of the temperature distribution image at the coordinate (x,y), R_image(x,y) represents the pixel value of the real-time image at the coordinate (x,y), and α is the weight coefficient, ranging from 0 to 1, which is used to adjust the proportion of the temperature distribution image and the real-time image in the fused image; S205: Post-processing the generated fused image F_image, including image smoothing, to eliminate artifacts and noise generated during the fusion process.

4. The fire remote monitoring and multi-level early warning mechanism integrated system according to claim 1 is characterized in that: The spatial temperature gradient change calculation unit obtains the temperature data of each measurement point in the key area from the data acquisition module, and calculates the spatial temperature gradient change using the gradient calculation formula. The specific formula is: Where T1 and T2 are the temperature values ​​of adjacent measurement points, and d is the distance between the two points.

5. The fire remote monitoring and multi-level early warning mechanism integrated system according to claim 3 is characterized in that: The calculation method of the heat collection area gradient change calculation unit is: Obtain the fused image from the image fusion module, use the image recognition algorithm to identify the high temperature or flame area from the fused image, and calculate its area, expressed as A(t), where t is the timestamp; The heat collection area data at consecutive time points are processed to calculate the gradient change of the heat collection area. The specific formula is: dA / dt=(A(t2)-A(t1)) / (t2-t1), where A(t1) and A(t2) are the heat collection areas at adjacent time points, respectively.

6. The fire remote monitoring and multi-level early warning mechanism integrated system according to claim 1 is characterized in that: The smoke concentration gradient change calculation unit obtains the smoke concentration data of each measurement point in the key area from the data acquisition module, expressed as {C1, C2, ..., Cm}, where m is the number of measurement points; and calculates the smoke concentration gradient change using the gradient calculation formula, the specific formula is: / d, where C1 and C2 are the smoke concentration values ​​of adjacent measuring points, and d is the distance between the two points.

7. The fire remote monitoring and multi-level early warning mechanism integrated system according to claim 1 is characterized in that: The risk prediction module uses a linear regression algorithm to construct a risk prediction model, and the specific steps include: S301: Obtaining calculation results of spatial temperature gradient change, heat accumulation area gradient change and smoke concentration gradient change from a data calculation module as input features of a risk prediction model; S302: Preprocessing the input features, including data cleaning and standardization; S303: Construct a linear regression model to establish a linear relationship between the input features and the fire risk level; assuming that the fire risk level is Y, the spatial temperature gradient change is X1, the heat accumulation area gradient change is X2, and the smoke concentration gradient change is X3, the linear regression model is expressed as: Y=β0+β1*X1+β2*X2+β3*X3+ε Among them, β0 is the intercept term, β1, β2, β3 are the regression coefficients of each input feature, and ε is the error term; S304: using historical fire data to train the linear regression model, solving the regression coefficients β0, β1, β2 and β3 by minimizing the loss function, and using the gradient descent method to update the values ​​of β0, β1, β2 and β3 during the training process to obtain the optimal solution of the loss function; S305: Verify and evaluate the trained linear regression model, including calculating the accuracy, recall and F1 score indicators of the model; S306: Deploy the verified linear regression model to the risk prediction module for real-time prediction of the fire risk level.

8. The fire remote monitoring and multi-level early warning mechanism integrated system according to claim 7 is characterized by: In step S302, the Z-score method is used to standardize the data. The mean of each numerical data is subtracted and divided by its standard deviation so that the processed data conforms to the standard normal distribution. The formula for standardization is: Z = (X-μ) / σ, where X is the original data, μ is the mean of the original data, σ is the standard deviation of the original data, and Z is the standardized data.

9. The fire remote monitoring and multi-level early warning mechanism integrated system according to claim 7 is characterized in that: In step 304, the specific method of updating the regression coefficient is: Step 1: Set the predicted value of the linear regression model as Y_pred, the actual fire risk level as Y_true, and the calculation formula of the mean square error MSE is: MSE = (1 / n) * Σ (Y_pred-Y_true)^2, where n is the number of samples and Σ represents the sum of all samples; Step 2: In order to minimize the MSE, the regression coefficient β of the linear regression model is iteratively updated using the gradient descent method, and the initial regression coefficients β0, β1, β2, β3, and the learning rate η are set; Step 3: Calculate the partial derivative of MSE with respect to each regression coefficient β. For each regression coefficient βi, the gradient calculation formula is: in, It is the partial derivative of the predicted value Y_pred with respect to the regression coefficient βi, which is obtained according to the formula of the linear regression model; Step 4: Update each regression coefficient βi according to the gradient and learning rate η. The update formula is: Among them, βi_old is the old regression coefficient, and βi_new is the updated regression coefficient; Step 5: Repeat steps 3 and 4 until the MSE converges to the preset threshold.

10. The fire remote monitoring and multi-level early warning mechanism integrated system according to claim 9 is characterized in that: The risk prediction module divides the fire risk level into multiple levels according to the output results of the risk prediction model, including low risk, medium risk, high risk and extremely high risk, and each level corresponds to different early warning notification strategies and emergency response measures.

Images