Fire alarm video decision control method based on multi-objective optimization algorithm

Abstract

The invention discloses a fire alarm video decision control method based on a multi-objective optimization algorithm. The fire alarm video decision control method comprises the following steps: step 1, arranging a dual-mode camera and a plurality of smoke concentration measuring instruments in a jth monitoring area; 2, normalizing the fire load of the jth monitoring area to obtain the normalized fire load of the jth monitoring area; step 3, performing normalized accumulation correction on the smoke detection risk of the jth monitoring area to obtain a normalized smoke detection risk of the jth monitoring area; 4, performing flame identification judgment on the image of the jth monitoring area to obtain a flame comprehensive score of the jth monitoring area; and step 5, multi-objective optimization decision making is carried out to control fire alarm. The method is simple in steps and reasonable in design, realizes accurate early warning and alarm by combining the fire load, the smoke sensing risk and the flame comprehensive score acquired in real time, and reduces the false alarm rate through video review.

Description

Technical Field

[0001] This invention belongs to the field of fire alarm technology, specifically relating to a fire alarm video decision control method based on a multi-objective optimization algorithm. Background Technology

[0002] Existing fire alarm systems mainly consist of smoke concentration meters, heat detectors, controllers, and alarms. In practical applications, they suffer from several drawbacks: detectors are prone to aging and dust accumulation, leading to frequent false alarms or missed alarms, severely reducing system reliability; they cannot visually present the actual situation of the alarm area, requiring manual on-site confirmation, which delays fire suppression efforts and wastes human resources; and they cannot directly view real-time video footage and perform multi-target analysis of alarm points on the fire alarm system, resulting in poor accuracy in fire assessment and difficulty in quickly implementing appropriate firefighting measures.

[0003] While some patents have attempted to address certain issues, significant shortcomings remain. For instance, the marine fire alarm system disclosed in patent CN111613002A, although capable of transmitting video to the cloud via the fire alarm system, cannot provide timely warnings. While patent CN116664359A enhances recognition capabilities through multi-sensor fusion, it relies on a cloud platform for data processing and decision-making, increasing network dependence and deployment costs. Furthermore, it lacks a multi-objective optimization mechanism—neither incorporating key indicators such as fire load, smoke risk, and comprehensive flame score into a collaborative optimization framework to generate a "safety baseline." Consequently, it cannot provide early warnings at the initial fire stage, nor can it quickly verify the alarm through local video. The system exhibits limitations in both response timeliness and judgment accuracy.

[0004] Therefore, a well-designed fire alarm video decision control method based on a multi-objective optimization algorithm is needed. This method integrates key indicators such as fire load, smoke risk, and flame comprehensive score to construct a multi-objective optimization model. The multi-objective dung beetle optimization algorithm generates attention baseline, early warning baseline, and alarm baseline. Subsequently, it combines real-time acquired fire load, smoke risk, and flame comprehensive score to achieve accurate early warning and alarm. Video verification reduces the false alarm rate, solving the problems of lack of early warning capability and delayed response in existing systems, and achieving more accurate fire alarms. Summary of the Invention

[0005] The technical problem to be solved by this invention is to address the shortcomings of the existing technology by providing a fire alarm video decision control method based on a multi-objective optimization algorithm. The method is simple in steps and reasonable in design. It constructs a multi-objective optimization model based on the fusion of key indicators such as fire load, smoke risk, and flame comprehensive score. The multi-objective dung beetle optimization algorithm generates attention baseline, early warning baseline, and alarm baseline. Subsequently, it combines real-time acquired fire load, smoke risk, and flame comprehensive score to achieve accurate early warning and alarm. Video verification reduces the false alarm rate, solving the problems of lack of early warning capability and delayed response in existing systems, and realizing the accuracy of fire alarm.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a fire alarm video decision control method based on a multi-objective optimization algorithm, which divides the building to be monitored into multiple rooms to be monitored, each room to be monitored being recorded as a monitoring area, and a fire alarm decision control method for any monitoring area, including the following steps: Step 1: Set up a dual-mode camera and multiple smoke concentration meters in the j-th monitoring area; wherein, the shooting range of a dual-mode camera covers the monitoring area it is located in; j is a positive integer; Step 2: Normalize the fire load of the j-th monitoring area to obtain the normalized fire load of the j-th monitoring area; Step 3: Normalize and cumulatively correct the smoke detection risk of the j-th monitoring area to obtain the normalized smoke detection risk of the j-th monitoring area; Step 4: Perform flame identification and judgment on the image of the j-th monitoring area to obtain the comprehensive flame score of the j-th monitoring area; Step 5: Multi-objective optimization decision-making to control fire alarms: Step 501: Based on steps two to four, construct a multi-objective optimization model, and use the NSGA-II algorithm and the dung beetle optimization algorithm to process the multi-objective optimization model, and output the final optimal Pareto solution set; Step 502: Under normal and simulated fire conditions in the monitored area, the final optimal Pareto solution set is mapped to a multi-objective optimization model to obtain the attention baseline, early warning baseline, and alarm baseline. Step 503: In the subsequent real-time monitoring process, based on the early warning baseline and the alarm baseline, the computer controls the flashing lights to provide fire early warning, controls the alarm to provide fire alarm, and displays the real-time video stream for verification on the computer.

[0007] The aforementioned fire alarm video decision control method based on a multi-objective optimization algorithm further includes step two, which is as follows: Step 201: Construct a combustible material attribute database based on the type of combustible material; wherein, the combustible material attribute database includes the mass of the combustible material, the calorific value of the combustible material, and the basic reactivity coefficient of the combustible material; the types of combustible materials include wood, cloth, cables, and oil drums; Step 202: Use a dual-mode camera to capture images of the j-th monitoring area, obtain visible light images and infrared images of the monitoring area, and send them to the computer; Step 203: The computer processes the visible light image of the monitored area through the trained first YOLOv8m network to identify the i-th type of combustible material and its bounding box; where i is a positive integer; 1≤i≤n; and n is the total number of combustible material types. The area enclosed by the boundary box of the i-th type of combustible material is mapped onto the infrared image of the monitoring area, and the highest temperature value of the area enclosed by the boundary box of the i-th type of combustible material is obtained and recorded as the temperature value of the environment where the i-th type of combustible material is located. ; Step 204: Select the mass of the i-th combustible material from the combustible material attribute database, denoted as . The calorific value of the i-th combustible material is denoted as The basic reactivity coefficient of the i-th combustible material is denoted as ; Step 205, according to The reactivity coefficient of the i-th combustible material based on temperature correction is obtained. ;in, Indicates the temperature setting threshold; The wind speed sensor measures the wind speed in the monitored area. If the measured wind speed value is greater than the wind speed threshold, the environment is determined to be a ventilation opening environment; and based on... The corrected reactivity coefficient of the i-th combustible material is obtained. ;in, For correction factors, and The value range is 1.1 to 1.2; Step 206, according to The fire load of the j-th monitoring area is obtained. ; Step 207: With the reactivity coefficient of all combustibles taking a value of 1.2, obtain the maximum fire load of the j-th monitoring area according to the method in step 206. ; Step 208, according to The normalized fire load of the j-th monitoring area is obtained. .

[0008] The aforementioned fire alarm video decision control method based on a multi-objective optimization algorithm further includes step three, which is as follows: Step 301: The smoke concentration meter in the j-th monitoring area measures the smoke concentration in real time, and obtains the smoke concentration measurement value of the e-th smoke concentration meter. The smoke concentration measurement value of the e-th smoke concentration meter and the lower limit setting value of smoke concentration Comparison, if the smoke concentration measurement value of the e-th smoke concentration meter If the smoke concentration is less than the lower limit set value, the normalized operating index of the e-th smoke concentration meter is normal. The value is zero; where e is a positive integer; The smoke concentration measurement value of the e-th smoke concentration meter and smoke concentration upper limit setting value Comparison, if the smoke concentration measurement value of the e-th smoke concentration meter If the smoke concentration is greater than or equal to the upper limit setting value, and the normalized operating index of the e-th smoke concentration meter is abnormal, then the normalized operating index of the e-th smoke concentration meter... =1; If the smoke concentration measurement value of the e-th smoke concentration meter In According to The normalized operating parameters of the e-th smoke concentration meter are obtained. ;in, This represents the maximum normalized operating index of the e-th smoke concentration meter. ; This represents the minimum normalized operating index of the e-th smoke concentration meter, and ; Step 302, according to The normalized smoke risk of the j-th monitoring area is obtained. ;in, This represents the overall correction factor. Let e ​​represent the weight of the e-th smoke concentration meter, and E represent the total number of smoke concentration meters, where 1 ≤ e ≤ E.

[0009] The above-mentioned fire alarm video decision control method based on multi-objective optimization algorithm further includes step four, the specific process of which is as follows: Step 401: Use a dual-mode camera to acquire visible light video of the j-th monitoring area. Detect flame targets by passing each frame of the visible light video through the trained second YOLOv8m network. Output the detection box and confidence score, which is the frame where flames are detected. Step 402: Obtain the flame frame rate St according to St=Nt / T; where T represents the total number of video frames within the set sampling time, and Nt represents the number of frames in which the flame is detected; Step 403, according to The average confidence level of the flame frame count was obtained. ;in, This represents the average confidence level in the p-th frame where a flame is detected; p is a positive integer, 1≤p≤Nt; Step 404, according to The overall flame score for the j-th monitoring area is obtained. ;in, This represents the first coefficient for flame recognition. This represents the second coefficient for flame identification.

[0010] The aforementioned fire alarm video decision control method based on a multi-objective optimization algorithm further includes step 501, which is specifically described below: Step 5011: For the j-th monitoring area, set... The negative number is used as the first objective function of the j-th monitoring area. ,and ;make The negative number is used as the second objective function for the j-th monitoring area. ,and ;make The negative number is used as the third objective function for the j-th monitoring area. ,and ; Step 5012: Construct a multi-objective optimization model, as follows: ;in, Let represent the b-th objective function for the j-th monitoring area, where b represents the number of objective functions and takes a value from 1 to 3. Decision variables include the basic reactivity coefficients of the first to nth combustibles, the weights of the first to Eth smoke concentration meters, and the overall correction coefficient. Flame recognition first coefficient And flame recognition second coefficient ; Step 5013: Set the boundaries of the decision variables; where the lower and upper limits of the basic reactivity coefficients of the n combustibles are 0 to 1, the lower and upper limits of the weights of the E smoke concentration meters are 0 to 1, and the overall correction coefficient is... The upper and lower limits are 0 to 1, and the first coefficient for flame recognition is... The upper and lower limits are 0 to 1, and the second coefficient for flame recognition is... The lower limit is 0 to 1, and ; Step 5014: Use the NSGA-II algorithm and the dung beetle optimization algorithm to process the multi-objective optimization model until the maximum number of iterations H is reached, and output the final optimal Pareto solution set.

[0011] The aforementioned fire alarm video decision control method based on a multi-objective optimization algorithm further includes step 5014, which is as follows: Step 50141: Set the dung beetle population size N to 80-120, randomly initialize the population, set the maximum number of iterations H to 500, and initialize the optimal Pareto solution set A0 as an empty set; Step 50142: Using the initial population as the parent population, based on the fitness values ​​calculated by the objective function, divide the parent population into rolling dung beetles, reproductive dung beetles, small dung beetles, and stealing dung beetles. Then, use the dung beetle optimization algorithm to perform rolling, reproductive, foraging, and stealing operations on the rolling dung beetles, reproductive dung beetles, small dung beetles, and stealing dung beetles respectively, generating the offspring population QA; where the number of offspring population QA is N; Step 50143: Using the initial population as the parent population, the NSGA-II algorithm is used to perform selection, crossover, and mutation to obtain the offspring population QB; where the number of offspring population QB is N. Step 50144: Merge the parent population, offspring population QA, and offspring population QB to form a merged population R; Step 50145: Using non-dominated sorting and elite retention strategies, select from the merged population R to obtain the new parent population Q1 for the first iteration; where the number of the new parent population Q1 for the first iteration is N; The new parent population Q1 in the first iteration is sorted based on non-dominance level and crowding distance. The first front solution is added to the initialized optimal Pareto solution set A0 to obtain the optimal Pareto solution set A1 in the first iteration. Step 50146: Take the new parent population Q1 from the first iteration as the parent population, and perform the next iteration according to the method of steps 50142 to 50145 to obtain the new parent population Q2 for the second iteration and the optimal Pareto solution set of the current generation in the second iteration. Merge the optimal Pareto solution set of the current generation in the second iteration and the optimal Pareto solution set A1 of the first iteration and perform non-dominated sorting. Then the first frontier solution is the optimal Pareto solution set A2 of the second iteration. Step 50147: Repeat step 50146 multiple times to iterate the new parent population in the H-1th iteration, and obtain the new parent population QH in the Hth iteration and the optimal Pareto solution set AH in the Hth iteration. Step 50148: Denote the optimal Pareto solution set AH of the Hth iteration as the final optimal Pareto solution set.

[0012] The aforementioned fire alarm video decision control method based on a multi-objective optimization algorithm further includes step 502, which is as follows: Step 5021: Substitute the k-th solution in the final optimal Pareto solution set into the first objective function, the second objective function, and the third objective function, and take the absolute value to obtain the normalized fire load Qk, normalized smoke risk Fk, and flame comprehensive score Rk corresponding to the k-th solution; where k is a positive integer; Step 5022: According to Uk=λ1×Qk+λ2×Fk+λ3×Rk, the alarm score Uk of the kth solution is obtained; where λ1, λ2 and λ3 represent the first weight, the second weight and the third weight respectively; the values ​​of λ1, λ2 and λ3 are in the range of 0 to 1, and λ1+λ2+λ3=1; Step 5023: Under normal circumstances in the monitored area, sort the alarm scores corresponding to all solutions in the final optimal Pareto solution set from largest to smallest, and take the absolute value of the objective function corresponding to the minimum value as the baseline of concern. Step 5024: In the case of a simulated fire in the monitored area, the alarm scores are sorted according to the methods in steps 5021 to 5023, and the absolute value of the objective function corresponding to the minimum value is taken as the alarm baseline. Step 5025: Use the average value between the alarm baseline and the attention baseline as the early warning baseline.

[0013] The aforementioned fire alarm video decision control method based on a multi-objective optimization algorithm further includes step 503, which is as follows: Step 5031: In the subsequent real-time monitoring process, at each set sampling time, according to the methods in steps 2 to 502, obtain the normalized fire load Qc, normalized smoke risk Fc and flame comprehensive score Rc corresponding to the minimum alarm score; Step 5032: If the normalized fire load Qc, normalized smoke risk Fc, and flame comprehensive score Rc at any set sampling time are greater than or equal to the attention baseline and less than the warning baseline, then a fire attention signal is triggered to remind the fire department to increase its attention. If any of the normalized fire load Qc, normalized smoke risk Fc, and flame comprehensive score Rc at a set sampling time are greater than the warning baseline but less than the alarm baseline, the computer controls the flashing lights to issue a fire warning and remind the fire department to conduct inspections. If the normalized fire load Qc, normalized smoke risk Fc, and flame comprehensive score Rc are greater than or equal to the alarm baseline for three consecutive sampling times, the computer will control the alarm to sound a fire alarm. Step 5033: When the computer-generated fire warning and fire alarm are triggered, the computer synchronously displays and verifies the real-time video stream.

[0014] Compared with the prior art, the present invention has the following advantages: 1. The method of the present invention has simple steps and reasonable design, which solves the problem of current fire alarms based on single factors and improves the accuracy of alarms.

[0015] 2. This invention normalizes the fire load of the j-th monitoring area to obtain the normalized fire load of the j-th monitoring area, normalizes and accumulates the smoke detection risk of the j-th monitoring area to obtain the normalized smoke detection risk of the j-th monitoring area, and performs flame recognition judgment on the image of the j-th monitoring area to obtain the comprehensive flame score of the j-th monitoring area, thereby realizing the fusion of key indicators such as fire load, smoke detection risk, and comprehensive flame score.

[0016] 3. Based on the normalized fire load, normalized smoke risk and flame comprehensive score of the j-th monitoring area, this invention constructs a multi-objective optimization model, and uses the NSGA-II algorithm and the dung beetle optimization algorithm to process the multi-objective optimization model and output the final optimal Pareto solution set; based on the final optimal Pareto solution set, it is mapped to the multi-objective optimization model to obtain the attention baseline, early warning baseline and alarm baseline.

[0017] 4. In the subsequent real-time monitoring process, this invention provides early warning and alarm based on the attention baseline, early warning baseline and alarm baseline. Through the design of early warning, alarm and video frame verification, it solves the problems of lagging response chain and single judgment mechanism in traditional fire early warning mechanism, and realizes the accuracy of fire alarm.

[0018] In summary, the method of this invention is simple in steps and reasonable in design. It constructs a multi-objective optimization model based on the fusion of key indicators such as fire load, smoke risk, and flame comprehensive score. It generates attention baseline, early warning baseline, and alarm baseline through a multi-objective dung beetle optimization algorithm. Subsequently, it combines real-time acquired fire load, smoke risk, and flame comprehensive score to achieve accurate early warning and alarm. Video verification reduces the false alarm rate, solving the problems of lack of early warning capability and delayed response in existing systems, and realizing the accuracy of fire alarm.

[0019] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0020] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0021] like Figure 1As shown, this invention discloses a fire alarm video decision control method based on a multi-objective optimization algorithm. The method divides the building to be monitored into multiple rooms to be monitored, with each room designated as a monitoring area. The fire alarm decision control method for any given monitoring area includes the following steps: Step 1: Set up a dual-mode camera and multiple smoke concentration meters in the j-th monitoring area; wherein, the shooting range of a dual-mode camera covers the monitoring area it is located in; j is a positive integer; Step 2: Normalize the fire load of the j-th monitoring area to obtain the normalized fire load of the j-th monitoring area; Step 3: Normalize and cumulatively correct the smoke detection risk of the j-th monitoring area to obtain the normalized smoke detection risk of the j-th monitoring area; Step 4: Perform flame identification and judgment on the image of the j-th monitoring area to obtain the comprehensive flame score of the j-th monitoring area; Step 5: Multi-objective optimization decision-making to control fire alarms: Step 501: Based on steps two to four, construct a multi-objective optimization model, and use the NSGA-II algorithm and the dung beetle optimization algorithm to process the multi-objective optimization model, and output the final optimal Pareto solution set; Step 502: Under normal and simulated fire conditions in the monitored area, the final optimal Pareto solution set is mapped to a multi-objective optimization model to obtain the attention baseline, early warning baseline, and alarm baseline. Step 503: In the subsequent real-time monitoring process, based on the early warning baseline and the alarm baseline, the computer controls the flashing lights to provide fire early warning, controls the alarm to provide fire alarm, and displays the real-time video stream for verification on the computer.

[0022] In this embodiment, step two is as follows: Step 201: Construct a combustible material attribute database based on the type of combustible material; wherein, the combustible material attribute database includes the mass of the combustible material, the calorific value of the combustible material, and the basic reactivity coefficient of the combustible material; the types of combustible materials include wood, cloth, cables, and oil drums; Step 202: Use a dual-mode camera to capture images of the j-th monitoring area, obtain visible light images and infrared images of the monitoring area, and send them to the computer; Step 203: The computer processes the visible light image of the monitored area through the trained first YOLOv8m network to identify the i-th type of combustible material and its bounding box; where i is a positive integer; 1≤i≤n; and n is the total number of combustible material types. The area enclosed by the boundary box of the i-th type of combustible material is mapped onto the infrared image of the monitoring area, and the highest temperature value of the area enclosed by the boundary box of the i-th type of combustible material is obtained and recorded as the temperature value of the environment where the i-th type of combustible material is located. ; Step 204: Select the mass of the i-th combustible material from the combustible material attribute database, denoted as . The calorific value of the i-th combustible material is denoted as The basic reactivity coefficient of the i-th combustible material is denoted as ; Step 205, according to The reactivity coefficient of the i-th combustible material based on temperature correction is obtained. ;in, Indicates the temperature setting threshold; The wind speed sensor measures the wind speed in the monitored area. If the measured wind speed value is greater than the wind speed threshold, the environment is determined to be a ventilation opening environment; and based on... The corrected reactivity coefficient of the i-th combustible material is obtained. ;in, For correction factors, and The value range is 1.1 to 1.2; Step 206, according to The fire load of the j-th monitoring area is obtained. ; Step 207: With the reactivity coefficient of all combustibles taking a value of 1.2, obtain the maximum fire load of the j-th monitoring area according to the method in step 206. ; Step 208, according to The normalized fire load of the j-th monitoring area is obtained. .

[0023] In this embodiment, step three is as follows: Step 301: The smoke concentration meter in the j-th monitoring area measures the smoke concentration in real time, and obtains the smoke concentration measurement value of the e-th smoke concentration meter. The smoke concentration measurement value of the e-th smoke concentration meter and the lower limit setting value of smoke concentration Comparison, if the smoke concentration measurement value of the e-th smoke concentration meter If the smoke concentration is less than the lower limit set value, the normalized operating index of the e-th smoke concentration meter is normal. The value is zero; where e is a positive integer; The smoke concentration measurement value of the e-th smoke concentration meter and smoke concentration upper limit setting value Comparison, if the smoke concentration measurement value of the e-th smoke concentration meter If the smoke concentration is greater than or equal to the upper limit setting value, and the normalized operating index of the e-th smoke concentration meter is abnormal, then the normalized operating index of the e-th smoke concentration meter... =1; If the smoke concentration measurement value of the e-th smoke concentration meter In According to The normalized operating parameters of the e-th smoke concentration meter are obtained. ;in, This represents the maximum normalized operating index of the e-th smoke concentration meter. ; This represents the minimum normalized operating index of the e-th smoke concentration meter, and ; Step 302, according to The normalized smoke risk of the j-th monitoring area is obtained. ;in, This represents the overall correction factor. Let e ​​represent the weight of the e-th smoke concentration meter, and E represent the total number of smoke concentration meters, where 1 ≤ e ≤ E.

[0024] In this embodiment, step four is as follows: Step 401: Use a dual-mode camera to acquire visible light video of the j-th monitoring area. Detect flame targets by passing each frame of the visible light video through the trained second YOLOv8m network. Output the detection box and confidence score, which is the frame where flames are detected. Step 402: Obtain the flame frame rate St according to St=Nt / T; where T represents the total number of video frames within the set sampling time, and Nt represents the number of frames in which the flame is detected; Step 403, according to The average confidence level of the flame frame count was obtained. ;in, This represents the average confidence level in the p-th frame where a flame is detected; p is a positive integer, 1≤p≤Nt; Step 404, according to The overall flame score for the j-th monitoring area is obtained. ;in, This represents the first coefficient for flame recognition. This represents the second coefficient for flame identification.

[0025] In this embodiment, step 501 is specifically performed as follows: Step 5011: For the j-th monitoring area, set... The negative number is used as the first objective function of the j-th monitoring area. ,and ;make The negative number is used as the second objective function for the j-th monitoring area. ,and ;make The negative number is used as the third objective function for the j-th monitoring area. ,and ; Step 5012: Construct a multi-objective optimization model, as follows: ;in, Let represent the b-th objective function for the j-th monitoring area, where b represents the number of objective functions and takes a value from 1 to 3. Decision variables include the basic reactivity coefficients of the first to nth combustibles, the weights of the first to Eth smoke concentration meters, and the overall correction coefficient. Flame recognition first coefficient And flame recognition second coefficient ; Step 5013: Set the boundaries of the decision variables; where the lower and upper limits of the basic reactivity coefficients of the n combustibles are 0 to 1, the lower and upper limits of the weights of the E smoke concentration meters are 0 to 1, and the overall correction coefficient is... The upper and lower limits are 0 to 1, and the first coefficient for flame recognition is... The upper and lower limits are 0 to 1, and the second coefficient for flame recognition is... The lower limit is 0 to 1, and ; Step 5014: Use the NSGA-II algorithm and the dung beetle optimization algorithm to process the multi-objective optimization model until the maximum number of iterations H is reached, and output the final optimal Pareto solution set.

[0026] In this embodiment, step 5014 is specifically performed as follows: Step 50141: Set the dung beetle population size N to 80-120, randomly initialize the population, set the maximum number of iterations H to 500, and initialize the optimal Pareto solution set A0 as an empty set; Step 50142: Using the initial population as the parent population, based on the fitness values ​​calculated by the objective function, divide the parent population into rolling dung beetles, reproductive dung beetles, small dung beetles, and stealing dung beetles. Then, use the dung beetle optimization algorithm to perform rolling, reproductive, foraging, and stealing operations on the rolling dung beetles, reproductive dung beetles, small dung beetles, and stealing dung beetles respectively, generating the offspring population QA; where the number of offspring population QA is N; Step 50143: Using the initial population as the parent population, the NSGA-II algorithm is used to perform selection, crossover, and mutation to obtain the offspring population QB; where the number of offspring population QB is N. Step 50144: Merge the parent population, offspring population QA, and offspring population QB to form a merged population R; Step 50145: Using non-dominated sorting and elite retention strategies, select from the merged population R to obtain the new parent population Q1 for the first iteration; where the number of the new parent population Q1 for the first iteration is N; The new parent population Q1 in the first iteration is sorted based on non-dominance level and crowding distance. The first front solution is added to the initialized optimal Pareto solution set A0 to obtain the optimal Pareto solution set A1 in the first iteration. Step 50146: Take the new parent population Q1 from the first iteration as the parent population, and perform the next iteration according to the method of steps 50142 to 50145 to obtain the new parent population Q2 for the second iteration and the optimal Pareto solution set of the current generation in the second iteration. Merge the optimal Pareto solution set of the current generation in the second iteration and the optimal Pareto solution set A1 of the first iteration and perform non-dominated sorting. Then the first frontier solution is the optimal Pareto solution set A2 of the second iteration. Step 50147: Repeat step 50146 multiple times to iterate the new parent population in the H-1th iteration, and obtain the new parent population QH in the Hth iteration and the optimal Pareto solution set AH in the Hth iteration. Step 50148: Denote the optimal Pareto solution set AH of the Hth iteration as the final optimal Pareto solution set.

[0027] In this embodiment, step 502 is specifically performed as follows: Step 5021: Substitute the k-th solution in the final optimal Pareto solution set into the first objective function, the second objective function, and the third objective function, and take the absolute value to obtain the normalized fire load Qk, normalized smoke risk Fk, and flame comprehensive score Rk corresponding to the k-th solution; where k is a positive integer; Step 5022: According to Uk=λ1×Qk+λ2×Fk+λ3×Rk, the alarm score Uk of the kth solution is obtained; where λ1, λ2 and λ3 represent the first weight, the second weight and the third weight respectively; the values ​​of λ1, λ2 and λ3 are in the range of 0 to 1, and λ1+λ2+λ3=1; Step 5023: Under normal circumstances in the monitored area, sort the alarm scores corresponding to all solutions in the final optimal Pareto solution set from largest to smallest, and take the absolute value of the objective function corresponding to the minimum value as the baseline of concern. Step 5024: In the case of a simulated fire in the monitored area, the alarm scores are sorted according to the methods in steps 5021 to 5023, and the absolute value of the objective function corresponding to the minimum value is taken as the alarm baseline. Step 5025: Use the average value between the alarm baseline and the attention baseline as the early warning baseline.

[0028] In this embodiment, step 503 is specifically performed as follows: Step 5031: In the subsequent real-time monitoring process, at each set sampling time, according to the methods in steps 2 to 502, obtain the normalized fire load Qc, normalized smoke risk Fc and flame comprehensive score Rc corresponding to the minimum alarm score; Step 5032: If the normalized fire load Qc, normalized smoke risk Fc, and flame comprehensive score Rc at any set sampling time are greater than or equal to the attention baseline and less than the warning baseline, then a fire attention signal is triggered to remind the fire department to increase its attention. If any of the normalized fire load Qc, normalized smoke risk Fc, and flame comprehensive score Rc at a set sampling time are greater than the warning baseline but less than the alarm baseline, the computer controls the flashing lights to issue a fire warning and remind the fire department to conduct inspections. If the normalized fire load Qc, normalized smoke risk Fc, and flame comprehensive score Rc are greater than or equal to the alarm baseline for three consecutive sampling times, the computer will control the alarm to sound a fire alarm. Step 5033: When the computer-generated fire warning and fire alarm are triggered, the computer synchronously displays and verifies the real-time video stream.

[0029] In this embodiment, during specific implementation, a visible light image of the monitored area is captured in advance using a dual-mode camera to obtain an image sample containing various combustibles, and various combustibles are marked on the image sample to obtain an image sample with combustibles marking. Then, the image samples labeled with combustibles are trained using the YOLOv8m network to obtain the first trained YOLOv8m network, which is used for combustible type detection.

[0030] In this embodiment, the specific implementation involves the temperature value of the environment where the i-th type of combustible material is located. Greater than 0.

[0031] In this embodiment, a temperature threshold is set during implementation. The value is 80℃~100℃, which is not specifically limited and can be adjusted according to the actual monitoring area requirements.

[0032] In this embodiment, the wind speed threshold is set to 3 to 5 meters per second, without being specifically limited, and is adjusted adaptively according to the requirements of the actual monitoring area.

[0033] In this embodiment, it should be noted that the lower limit setting value and the upper limit setting value of smoke concentration are set according to actual requirements and are not specifically limited.

[0034] In this embodiment, during specific implementation, visible light images of the monitored area are captured in advance using a dual-mode camera to obtain image samples containing flames. Flames are then labeled on the image samples to obtain flame-labeled image samples, which constitute the flame image dataset.

[0035] The YOLOv8m network was trained using a flame image dataset to obtain a trained second YOLOv8m network, which was then used for flame detection.

[0036] In this embodiment, it should be noted that the YOLOv8m network and its training can be performed in accordance with conventional methods in the art.

[0037] In this embodiment, the NSGA-II algorithm, namely the non-dominated sorting genetic algorithm with an elitist strategy, reduces the number of iterations and significantly improves the convergence speed of the algorithm by using fast non-dominated sorting and an elitist retention strategy.

[0038] In this embodiment, when dividing the parent population into dung beetles, reproductive dung beetles, small dung beetles, and thieving dung beetles based on fitness values ​​in step 50142, the fitness values ​​are first arranged in descending order. Then, individuals corresponding to the first 50% of fitness values ​​are designated as dung beetles, individuals corresponding to the middle 30% of fitness values ​​are designated as reproductive dung beetles, individuals corresponding to the next 15% of fitness values ​​are designated as small dung beetles, and individuals corresponding to the last 5% of fitness values ​​are designated as thieving dung beetles.

[0039] In this embodiment, step 50143 is specifically performed as follows: Step A: Use the initial population as the parent population P1, and use the binary tournament selection method to obtain the mating pool; wherein, the number of individuals in the mating pool is N; Step B: Use the simulated binary crossover method to crossbreed individuals in the mating pool to produce offspring individuals; Step C: Use polynomial mutation to mutate the offspring individuals to obtain mutated individuals; Step D: Based on the boundary of the decision variables, use the box constraint pruning method to process the mutated individuals, obtain the mutated individuals within the boundary range, and denote them as the offspring population QB.

[0040] In this embodiment, it should be noted that the archive size of the optimal Pareto solution set A is N / 2, that is, the optimal Pareto solution set A has at most N / 2 solutions, so as to achieve a good balance between efficiency and quality.

[0041] In this embodiment, it should be noted that the values ​​of λ1, λ2, and λ3 can be adjusted according to actual requirements and are given by the actual fire safety management party in combination with scenario preferences, without specific limitations. For example, in a warehouse storing heavy-duty combustible materials, λ1 can be made larger to emphasize the control of fire load; in an electrical equipment room where the cost of false alarms is extremely high, λ2 and λ3 can be made larger to strengthen the constraint on smoke detectors and video false alarms.

[0042] In this embodiment, it should be noted that, compared with traditional alarm systems, the present invention achieves an upgrade from "passive alarm" to "active early warning" through algorithm empowerment, which significantly improves the accuracy of judgment and the efficiency of emergency response, and provides technical support for building safety management and smart fire protection in cities.

[0043] In this embodiment, it should be noted that if the normalized fire load Qc, normalized smoke risk Fc, and flame comprehensive score Rc at any set sampling time are all less than the baseline of concern, then they are not normally considered, and the system only performs data recording and background analysis.

[0044] In this embodiment, it should be noted that the sampling time is set to 5 seconds. This can be set according to actual requirements and is not specifically limited.

[0045] In summary, the method of this invention is simple in steps and reasonable in design. It constructs a multi-objective optimization model based on the fusion of key indicators such as fire load, smoke risk, and flame comprehensive score. It generates attention baseline, early warning baseline, and alarm baseline through a multi-objective dung beetle optimization algorithm. Subsequently, it combines real-time acquired fire load, smoke risk, and flame comprehensive score to achieve accurate early warning and alarm. Video verification reduces the false alarm rate, solving the problems of lack of early warning capability and delayed response in existing systems, and realizing the accuracy of fire alarm.

[0046] The above description is merely a preferred embodiment of the present invention and does not constitute any limitation on the present invention. Any simple modifications, alterations, or equivalent structural changes made to the above embodiments based on the technical essence of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A fire alarm video decision control method based on a multi-objective optimization algorithm, characterized in that, The building to be monitored is divided into multiple rooms to be monitored, and each room is designated as a monitoring area. A decision-making and control method for fire alarms in any given monitoring area includes the following steps: Step 1: Set up a dual-mode camera and multiple smoke concentration meters in the j-th monitoring area; wherein, the shooting range of a dual-mode camera covers the monitoring area it is located in; j is a positive integer; Step 2: Normalize the fire load of the j-th monitoring area to obtain the normalized fire load of the j-th monitoring area; Step 3: Normalize and cumulatively correct the smoke detection risk of the j-th monitoring area to obtain the normalized smoke detection risk of the j-th monitoring area; Step 4: Perform flame identification and judgment on the image of the j-th monitoring area to obtain the comprehensive flame score of the j-th monitoring area; Step 5: Multi-objective optimization decision-making to control fire alarms: Step 501: Based on steps two to four, construct a multi-objective optimization model, and use the NSGA-II algorithm and the dung beetle optimization algorithm to process the multi-objective optimization model, and output the final optimal Pareto solution set; Step 502: Under normal and simulated fire conditions in the monitored area, the final optimal Pareto solution set is mapped to a multi-objective optimization model to obtain the attention baseline, early warning baseline, and alarm baseline. Step 503: In the subsequent real-time monitoring process, based on the early warning baseline and the alarm baseline, the computer controls the flashing lights to provide fire early warning, controls the alarm to provide fire alarm, and displays the real-time video stream for verification on the computer.

2. The fire alarm video decision control method based on a multi-objective optimization algorithm according to claim 1, characterized in that: Step two, the specific process is as follows: Step 201: Construct a combustible material attribute database based on the type of combustible material; wherein, the combustible material attribute database includes the mass of the combustible material, the calorific value of the combustible material, and the basic reactivity coefficient of the combustible material; the types of combustible materials include wood, cloth, cables, and oil drums; Step 202: Use a dual-mode camera to capture images of the j-th monitoring area, obtain visible light images and infrared images of the monitoring area, and send them to the computer; Step 203: The computer processes the visible light image of the monitored area through the trained first YOLOv8m network to identify the i-th type of combustible material and its bounding box; where i is a positive integer; 1≤i≤n; and n is the total number of combustible material types. The area enclosed by the boundary box of the i-th type of combustible material is mapped onto the infrared image of the monitoring area, and the highest temperature value of the area enclosed by the boundary box of the i-th type of combustible material is obtained and recorded as the temperature value of the environment where the i-th type of combustible material is located. ; Step 204: Select the mass of the i-th combustible material from the combustible material attribute database, denoted as . The calorific value of the i-th combustible material is denoted as The basic reactivity coefficient of the i-th combustible material is denoted as ; Step 205, according to The reactivity coefficient of the i-th combustible material based on temperature correction is obtained. ;in, Indicates the temperature setting threshold; The wind speed sensor measures the wind speed in the monitored area. If the measured wind speed value is greater than the wind speed threshold, the environment is determined to be a ventilation opening environment; and based on... The corrected reactivity coefficient of the i-th combustible material is obtained. ;in, For correction factors, and The value range is 1.1 to 1.2; Step 206, according to The fire load of the j-th monitoring area is obtained. ; Step 207: With the reactivity coefficient of all combustibles set to 1.2, obtain the maximum fire load of the j-th monitoring area using the method described in Step 206. ; Step 208, according to The normalized fire load of the j-th monitoring area is obtained. .

3. The fire alarm video decision control method based on a multi-objective optimization algorithm according to claim 1, characterized in that: Step three, the specific process is as follows: Step 301: The smoke concentration meter in the j-th monitoring area measures the smoke concentration in real time, and obtains the smoke concentration measurement value of the e-th smoke concentration meter. The smoke concentration measurement value of the e-th smoke concentration meter and the lower limit setting value of smoke concentration Comparison, if the smoke concentration measurement value of the e-th smoke concentration meter If the smoke concentration is less than the lower limit set value, the normalized operating index of the e-th smoke concentration meter is normal. The value is zero; where e is a positive integer; The smoke concentration measurement value of the e-th smoke concentration meter and smoke concentration upper limit setting value Comparison, if the smoke concentration measurement value of the e-th smoke concentration meter If the smoke concentration is greater than or equal to the upper limit setting value, and the normalized operating index of the e-th smoke concentration meter is abnormal, then the normalized operating index of the e-th smoke concentration meter... =1; If the smoke concentration measurement value of the e-th smoke concentration meter In According to The normalized operating parameters of the e-th smoke concentration meter are obtained. ;in, This represents the maximum normalized operating index of the e-th smoke concentration meter. ; This represents the minimum normalized operating index of the e-th smoke concentration meter, and ; Step 302, according to The normalized smoke risk of the j-th monitoring area is obtained. ;in, This represents the overall correction factor. Let e ​​represent the weight of the e-th smoke concentration meter, and E represent the total number of smoke concentration meters, where 1 ≤ e ≤ E.

4. A fire alarm video decision control method based on a multi-objective optimization algorithm as described in claim 1, characterized in that: Step four, the specific process is as follows: Step 401: Use a dual-mode camera to acquire visible light video of the j-th monitoring area. Detect flame targets by passing each frame of the visible light video through the trained second YOLOv8m network. Output the detection box and confidence score, which is the frame where flames are detected. Step 402: Obtain the flame frame rate St according to St=Nt / T; where T represents the total number of video frames within the set sampling time, and Nt represents the number of frames in which the flame is detected; Step 403, according to The average confidence level of the flame frame count was obtained. ;in, This represents the average confidence level in the p-th frame where a flame is detected; p is a positive integer, 1≤p≤Nt; Step 404, according to The overall flame score for the j-th monitoring area is obtained. ;in, This represents the first coefficient for flame recognition. This represents the second coefficient for flame identification.

5. A fire alarm video decision control method based on a multi-objective optimization algorithm according to claim 4, characterized in that: Step 501, the specific process is as follows: Step 5011: For the j-th monitoring area, set... The negative number is used as the first objective function of the j-th monitoring area. ,and ;make The negative number is used as the second objective function for the j-th monitoring area. ,and ;make The negative number is used as the third objective function for the j-th monitoring area. ,and ; Step 5012: Construct a multi-objective optimization model, as follows: ;in, Let represent the b-th objective function for the j-th monitoring area, where b represents the number of objective functions and takes a value from 1 to 3. Decision variables include the basic reactivity coefficients of the first to nth combustibles, the weights of the first to Eth smoke concentration meters, and the overall correction coefficient. Flame recognition first coefficient And flame recognition second coefficient ; Step 5013: Set the boundaries of the decision variables; where the lower and upper limits of the basic reactivity coefficients of the n combustibles are 0 to 1, the lower and upper limits of the weights of the E smoke concentration meters are 0 to 1, and the overall correction coefficient is... The upper and lower limits are 0 to 1, and the first coefficient for flame recognition is... The upper and lower limits are 0 to 1, and the second coefficient for flame recognition is... The lower limit is 0 to 1, and ; Step 5014: Use the NSGA-II algorithm and the dung beetle optimization algorithm to process the multi-objective optimization model until the maximum number of iterations H is reached, and output the final optimal Pareto solution set.

6. A fire alarm video decision control method based on a multi-objective optimization algorithm as described in claim 5, characterized in that: Step 5014, the specific process is as follows: Step 50141: Set the dung beetle population size N to 80-120, randomly initialize the population, set the maximum number of iterations H to 500, and initialize the optimal Pareto solution set A0 as an empty set; Step 50142: Using the initial population as the parent population, based on the fitness values ​​calculated by the objective function, divide the parent population into rolling dung beetles, reproductive dung beetles, small dung beetles, and stealing dung beetles. Then, use the dung beetle optimization algorithm to perform rolling, reproductive, foraging, and stealing operations on the rolling dung beetles, reproductive dung beetles, small dung beetles, and stealing dung beetles respectively, generating the offspring population QA; where the number of offspring population QA is N; Step 50143: Using the initial population as the parent population, the NSGA-II algorithm is used to perform selection, crossover, and mutation to obtain the offspring population QB; where the number of offspring population QB is N. Step 50144: Merge the parent population, offspring population QA, and offspring population QB to form a merged population R; Step 50145: Using non-dominated sorting and elite retention strategies, select from the merged population R to obtain the new parent population Q1 for the first iteration; where the number of the new parent population Q1 for the first iteration is N; The new parent population Q1 in the first iteration is sorted based on non-dominance level and crowding distance. The first front solution is added to the initialized optimal Pareto solution set A0 to obtain the optimal Pareto solution set A1 in the first iteration. Step 50146: Take the new parent population Q1 from the first iteration as the parent population, and perform the next iteration according to the method of steps 50142 to 50145 to obtain the new parent population Q2 for the second iteration and the optimal Pareto solution set of the current generation in the second iteration. Merge the optimal Pareto solution set of the current generation in the second iteration and the optimal Pareto solution set A1 of the first iteration and perform non-dominated sorting. Then the first frontier solution is the optimal Pareto solution set A2 of the second iteration. Step 50147: Repeat step 50146 multiple times to iterate the new parent population in the H-1th iteration, and obtain the new parent population QH in the Hth iteration and the optimal Pareto solution set AH in the Hth iteration. Step 50148: Denote the optimal Pareto solution set AH of the Hth iteration as the final optimal Pareto solution set.

7. A fire alarm video decision control method based on a multi-objective optimization algorithm according to claim 6, characterized in that: Step 502, the specific process is as follows: Step 5021: Substitute the k-th solution in the final optimal Pareto solution set into the first objective function, the second objective function, and the third objective function, and take the absolute value to obtain the normalized fire load Qk, normalized smoke risk Fk, and flame comprehensive score Rk corresponding to the k-th solution; where k is a positive integer; Step 5022: According to Uk=λ1×Qk+λ2×Fk+λ3×Rk, the alarm score Uk of the kth solution is obtained; where λ1, λ2 and λ3 represent the first weight, the second weight and the third weight respectively; the values ​​of λ1, λ2 and λ3 are in the range of 0 to 1, and λ1+λ2+λ3=1; Step 5023: Under normal circumstances in the monitored area, sort the alarm scores corresponding to all solutions in the final optimal Pareto solution set from largest to smallest, and take the absolute value of the objective function corresponding to the minimum value as the baseline of concern. Step 5024: In the case of a simulated fire in the monitored area, the alarm scores are sorted according to the methods in steps 5021 to 5023, and the absolute value of the objective function corresponding to the minimum value is taken as the alarm baseline. Step 5025: Use the average value between the alarm baseline and the attention baseline as the early warning baseline.

8. A fire alarm video decision control method based on a multi-objective optimization algorithm according to claim 7, characterized in that: Step 503, the specific process is as follows: Step 5031: In the subsequent real-time monitoring process, at each set sampling time, according to the methods in steps 2 to 502, obtain the normalized fire load Qc, normalized smoke risk Fc and flame comprehensive score Rc corresponding to the minimum alarm score; Step 5032: If the normalized fire load Qc, normalized smoke risk Fc, and flame comprehensive score Rc at any set sampling time are greater than or equal to the attention baseline and less than the warning baseline, then a fire attention signal is triggered to remind the fire department to increase its attention. If any of the normalized fire load Qc, normalized smoke risk Fc, and flame comprehensive score Rc at a set sampling time are greater than the warning baseline but less than the alarm baseline, the computer controls the flashing lights to issue a fire warning and remind the fire department to conduct inspections. If the normalized fire load Qc, normalized smoke risk Fc, and flame comprehensive score Rc are greater than or equal to the alarm baseline for three consecutive sampling times, the computer will control the alarm to sound a fire alarm. Step 5033: When the computer-generated fire warning and fire alarm are triggered, the computer synchronously displays and verifies the real-time video stream.

Images