A real-time identification and prediction method, device and medium for fire source parameters and air temperature

Abstract

The present application relates to a kind of fire source parameter and air temperature real-time identification and prediction method, equipment, medium, comprising: the numerical model of single-storey factory building is established;Based on numerical simulation, obtain the single-storey factory building fire heat response data for the training of deep learning agent model;Determine single-storey factory building air temperature measurement position;Deep learning agent model based on recurrent neural network and fully connected neural network is constructed;Measured historical air temperature and fire single-storey factory building geometric parameter are as input, current fire source parameter and future air temperature are as output, and deep learning agent model is trained;The deep learning agent model obtained by training is used as final agent model to identify the fire source parameter of single-storey factory building in actual fire scene and predict future air temperature.Compared with prior art, the present application can identify the fire source state parameter of the building in fire in real time, and further predict the future air temperature development trend, can quickly and accurately obtain fire state and development trend.

Description

Technical Field

[0001] This invention relates to the fields of public safety technology and neural network deep learning, and in particular to a method, device, and medium for real-time identification and prediction of fire source parameters and air temperature. Background Technology

[0002] Building fires pose a serious threat to the lives of trapped personnel and rescue workers. The state and development trend of a fire significantly impact firefighting and rescue decisions. However, in actual fires, fire source parameters constantly change due to the combustion of combustibles and variations in ventilation conditions, making it difficult for firefighters to quickly and accurately assess the fire's state and development trend, thus affecting the efficiency and safety of firefighting and rescue efforts. For a fire-affected factory building, changes in the fire source state can be reflected by on-site measured air temperature. However, the mapping relationship between fire source state parameters and air temperature is highly non-linear, and variations in the geometric dimensions of different factory buildings also affect this mapping relationship, posing a significant challenge to building fire identification and prediction.

[0003] Patent application CN115761409A discloses a fire detection method based on deep learning. This method involves labeling fire sample images with categories and fire regions to obtain a fire training set. A pre-constructed YOLOX network is trained using this training set to obtain predicted fire region bounding boxes and category confidence scores for the fire sample images. The network parameters of the YOLOX network are updated using category loss and detection box position loss until the YOLOX network converges, resulting in a fire detection model. The method extracts images to be detected from surveillance video data and uses the fire detection model to perform fire detection on these images. While this method offers fast fire detection speed, it cannot quickly and accurately identify and predict fire source parameters and air temperature in real time. Summary of the Invention

[0004] The purpose of this invention is to overcome the defects of the prior art by providing a method, device, and medium for real-time identification and prediction of fire source parameters and air temperature. This invention can identify the fire source status parameters of a fire-affected building in real time and further predict the future air temperature development trend. It can quickly and accurately obtain the fire status and development trend, and is applicable to single-story factory buildings with different geometric dimensions and different fire scenarios.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] This invention provides a method for real-time identification and prediction of fire source parameters and air temperature, comprising a training phase and an application phase. Based on extensive and accurate numerical simulations and a pre-trained deep learning model, it can calculate fire source state parameters and future air temperature in real time based on the geometric parameters of the affected factory building and historical air temperature data measured at the rescue site during a fire, providing guidance for fire rescue operations in single-story factory buildings. In this invention, "air temperature" refers to the ambient air temperature, and will be used throughout the following descriptions.

[0007] Specifically, the following steps are included:

[0008] S1: Determine the probability distribution of the geometric parameters and fire source state parameters of a single-story factory building during a fire, perform random sampling based on the determined probability distribution, and establish a numerical model of the single-story factory building.

[0009] S2: Perform fire numerical simulation on the established numerical model, and obtain fire thermal response data of a single-story factory building for training deep learning agent model based on the numerical simulation. The fire thermal response data of the single-story factory building includes the time history curves of air temperature at key temperature measuring points throughout the fire process and the time history curves of fire source state parameters throughout the fire process.

[0010] S3: Determine the location for measuring air temperature in a single-story factory building;

[0011] S4: Construct a deep learning proxy model based on recurrent neural networks and fully connected neural networks. Further, the deep learning proxy model includes a fire source parameter identification module and an air temperature prediction module. The input of the fire source parameter identification module is the historical air temperature measured time history curve (historical air temperature-time curve) and the geometric parameters of the single-story factory building. The output is the real fire model of the single-story factory building and the fire source state parameters. The input of the air temperature prediction module is the real fire model of the single-story factory building. The output is the predicted time history curve of the air temperature of the single-story factory building with a certain confidence τ. Specifically, the geometric parameters of the single-story factory building and the historical air temperature are used as the input layer, the fire source state parameters and the future air temperature are used as the output layer, and the real fire model is used as the internal output input layer. The fire source parameter identification module identifies the real fire model of the fire-affected factory building based on the input factory building geometric parameters and historical air temperature. The real fire model outputs the current fire source state parameters and uses them as intermediate states to input into the air temperature prediction module. The air temperature prediction module obtains the future air temperature based on the identified real fire model through fire dynamics analysis and combustion analysis.

[0012] S5: Train the deep learning proxy model by taking the thermal response data of a single-story factory building fire and the geometric parameters of the fire-affected factory building as inputs, and the current fire source state parameters and future air temperature as outputs. Train the model until it meets the prediction accuracy requirements. The specific training process is as follows: Divide the air temperature-time curve of the single-story factory building into the measured historical air temperature-time curve and the predicted future air temperature-time curve according to the measurement time period and the prediction time period. Input the historical air temperature-time curve and the geometric parameters of the factory building into the fire source parameter identification module of the deep learning model, and output the fire source state parameters of the fire-affected single-story factory building at the current moment. Input the fire source state parameters into the air temperature prediction module of the deep learning model, and use the future air temperature-time curve as the output of the deep learning model. The training phase should be completed after the construction of the single-story factory building and before the fire occurs. The aim is to determine the relevant model parameters of the fire source parameter identification module and the air temperature prediction module in the deep learning model in advance through a large amount of reliable numerical analysis data.

[0013] S6: The trained deep learning proxy model is used as the final proxy model for identifying fire source state parameters and predicting future air temperature in a single-story factory building during an actual fire. Based on the measured air temperature data and factory geometric parameters, the data is input into the pre-trained deep learning proxy model. The deep learning proxy model then identifies the fire state parameters in real time based on the input data and outputs a future air temperature prediction value with confidence. Preferably, the air temperature data includes the air temperature at 3 / 4 height of the factory columns, 1 / 4 and 3 / 4 span of the beams. The deep learning proxy model is used after a fire occurs in the factory building, when firefighters arrive at the scene for rescue. It aims to accurately identify fire state parameters in real time based on the trained deep learning model, combined with measured air temperature data and factory geometric feature data, and provide a future air temperature prediction value at a set confidence level τ to guide fire rescue efforts.

[0014] Furthermore, the geometric parameters include span, column spacing, number of stilts, and eaves height. The range of values ​​for the geometric parameters is determined according to the design specifications for single-story factory buildings. The fire source status parameters include ignition location, fire-affected area, and heat release rate per unit area. The probability distribution of the fire source status parameters is determined according to the design specifications for single-story factory buildings. Furthermore, the fire source status parameters include the relative position of the fire source center in the span direction and column spacing direction at the current moment, and the total heat release power of the fire source.

[0015] Furthermore, numerical simulations were performed using FDS and Fluent fire analysis software.

[0016] Furthermore, the training of the deep learning surrogate model uses the root mean square error and quantile error. The loss function of the fire source parameter identification module is set as the root mean square error, and the loss function of the air temperature prediction module is set as the quantile error. The total error is the weighted sum of the root mean square error and the quantile error. The confidence level τ in the calculation of the quantile error is consistent with the confidence level τ of the future air temperature to be predicted. The error backpropagation algorithm is used to update the parameters to be learned in the deep learning surrogate model. The error backpropagation algorithm calculates the difference between the output and the expected output of the deep learning surrogate model, and then backpropagates this difference to each layer of the deep learning surrogate model to update the weights and biases of the deep learning surrogate model, thereby minimizing the error of the deep learning surrogate model. Through the backpropagation algorithm, the deep learning surrogate model can continuously adjust its weights and biases according to the training data, thereby improving the prediction accuracy of the input data.

[0017] Furthermore, the relative error r, interval coverage probability ICP, and interval width IW are used to evaluate the performance of the trained deep learning model on the test set. r is the difference between the fire source parameters identified by the deep learning model on the test set and the true fire source parameters. ICP is the probability that the air temperature values predicted by the deep learning model on the test set in the future Δt time are higher than the true air temperature values. IW statistically calculates the error between the air temperature values predicted by the deep learning model on the test set in the future Δt time and the true air temperature values. The judgment criterion for high-precision samples is defined as r < 0.1 and -5% < ICP - confidence level τ < 5% and IW < 50°C, and the deep learning model with the highest proportion of high-precision test samples is used as the final surrogate model.

[0018] Furthermore, the air temperature is measured by embedding thermocouples during the construction of the factory building.

[0019] The present invention also provides an electronic device, including a memory and a processor. The processor is used to execute the program in the memory to implement the real-time identification and prediction method of fire source parameters and air temperature as described above.

[0020] The present invention also provides a storage medium containing computer-executable instructions. When the storage medium of the computer-executable instructions is executed by a computer processor, it is used to execute the real-time identification and prediction method of fire source parameters and air temperature as described above.

[0021] Using advanced machine learning methods to perform real-time identification of the fire state of a fire-affected building structure based on measured air temperature data and reliable prediction of the development trend, this method is applicable to single-story factory buildings with different geometric sizes and different fire scenarios, and is of great significance to building fire rescue, so it has great public safety value.

[0022] Compared with the prior art, the present invention has the following advantages:

[0023] (1) Based on deep learning, this invention proposes a method, equipment and medium for real-time identification and prediction of fire source parameters and air temperature of single-story factory buildings. Using measured historical air temperature and geometric parameters of the fire-affected single-story factory buildings as data sources, it can identify the fire source status parameters of the fire-affected buildings in real time and further predict the future air temperature development trend. It can quickly and accurately obtain the fire status and development trend.

[0024] (2) It has many application scenarios and is suitable for single-story factory buildings with different geometric dimensions and different fire scenarios. Attached Figure Description

[0025] Figure 1 This is a flowchart of a method for real-time identification and prediction of fire source parameters and air temperature.

[0026] Figure 2 This is a structural diagram of a single-story factory building.

[0027] Figure 3 This is a structural diagram of a deep learning agent model.

[0028] Figure 4 This is a schematic diagram of a deep learning agent model training method.

[0029] Figure 5 This is a schematic diagram illustrating the application methods of deep learning agent models.

[0030] Figure 6 A method for real-time acquisition of air temperature when applying deep learning agent models. Detailed Implementation

[0031] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. Unless otherwise specified in this technical solution, features such as component models, material names, connection structures, control methods, and algorithms are considered common technical features disclosed in the prior art. It should be particularly noted that the key technology of this invention lies in using deep learning theory to identify the fire source state parameters of a fire-affected building and predict future air temperature. Any modifications or refinements made to the main design concept and spirit of this invention that are not substantially meaningful, but whose technical methods are still consistent with this invention, should be included within the scope of protection of this invention.

[0032] Example 1

[0033] This embodiment provides a method for real-time identification and prediction of ignition source parameters and ambient temperature, such as... Figure 1 As shown, it includes the following steps:

[0034] S1: Determine the probability distribution of geometric parameters and fire source state parameters of a single-story factory building during a fire. Perform random sampling based on the determined probability distribution and establish a numerical model of a single-story factory building fire in the fire simulation software FDS.

[0035] S2: Perform fire numerical simulation on the established numerical model, and obtain single-story factory building fire thermal response data for training deep learning agent model based on the numerical simulation. The single-story factory building fire thermal response data includes the time history curve of air temperature at key temperature measurement points during the entire fire process (air temperature-time data) and the time history curve of fire source state parameters during the entire fire process (fire source parameters-time data). The fire source parameters-time data and air temperature-time data are used as training samples for deep learning agent model.

[0036] S3: Determine the location for measuring air temperature in a single-story factory building;

[0037] S4: Construct a deep learning agent model based on recurrent neural networks and fully connected neural networks, such as... Figure 3 As shown, the deep learning proxy model includes a fire source parameter identification module and an air temperature prediction module. The geometric parameters of the single-story factory building and the historical air temperature are used as the input layer, the fire source state parameters and the future air temperature are used as the output layer, and the real fire model is used as the internal output and input layer. The fire source parameter identification module identifies the real fire model of the fire-affected factory building based on the input factory building geometric parameters and historical air temperature. The real fire model outputs the current fire source state parameters and uses them as intermediate states to input into the air temperature prediction module. The air temperature prediction module obtains the future air temperature based on the identified real fire model through fire dynamics analysis and combustion analysis.

[0038] S5: As Figure 4 As shown, the thermal response data of a single-story factory building fire and the geometric parameters of the fire-affected factory building are used as inputs, and the current fire source state parameters and future air temperature are used as outputs to train the deep learning proxy model until the prediction accuracy requirements are met. The specific training process is as follows: the air temperature-time curve of the single-story factory building is divided into the measured historical air temperature-time curve and the predicted future air temperature-time curve according to the measurement time period and the prediction time period. The historical air temperature-time curve and the geometric parameters of the factory building are input into the fire source parameter identification module of the deep learning model, and the fire source state parameters of the fire-affected single-story factory building at the current moment are output. The fire source state parameters are input into the air temperature prediction module of the deep learning model, and the predicted future air temperature-time curve is used as the output of the deep learning model. The training phase should be completed after the construction of the single-story factory building and before the fire occurs. The aim is to determine the relevant model parameters of the fire source parameter identification module and the air temperature prediction module in the deep learning model in advance through a large amount of reliable numerical analysis data.

[0039] S6: As Figure 5 As shown, the trained deep learning proxy model is used as the final proxy model for identifying fire source state parameters and predicting future air temperature in a single-story factory building during an actual fire. Based on the measured air temperature data and the factory's geometric parameters, the data is input into the pre-trained deep learning proxy model. The deep learning proxy model then identifies the fire state parameters in real time based on the input data and outputs a predicted future air temperature with confidence level τ. In a specific implementation, such as... Figure 2 , Figure 6 As shown, the air temperature is measured by pre-embedded thermocouples during the construction of the factory building. The air temperature data on any column of the factory building includes the air temperature at 3 / 4 height, 1 / 4 and 3 / 4 span of the beam. The fire source status parameters include the relative position of the fire source center in the span direction and column spacing direction at the current moment, and the total heat release power of the fire source.

[0040] In specific implementations, the geometric parameters include span, column spacing, number of purlins, and eaves height, and the range of values ​​for the geometric parameters is determined according to the design specifications for single-story factory buildings.

[0041] The deep learning proxy model is trained using root mean square error (RMSE) and quantile error. The loss function for the fire source parameter identification module is set to RMSE, while the loss function for the air temperature prediction module is set to quantile error. The total error is the weighted sum of RMSE and quantile error. The confidence level τ in the quantile error calculation is consistent with the confidence level τ of the desired predicted future air temperature. An error backpropagation algorithm is used to update the learnable parameters in the deep learning proxy model. The error backpropagation algorithm calculates the difference between the deep learning proxy model's output and the expected output, then propagates this difference back to each layer of the deep learning proxy model to update its weights and biases, thereby minimizing the model's error. Through backpropagation, the deep learning proxy model can continuously adjust its weights and biases based on the training data, thus improving the prediction accuracy of the input data.

[0042] The performance of the trained deep learning model on the test set is evaluated using the relative error r, the interval coverage probability ICP, and the interval width IW. r is the difference between the fire source parameters identified by the deep learning model on the test set and the true fire source parameters. ICP is the probability that the predicted air temperature value within the future Δt time by the deep learning model on the test set is higher than the true air temperature value. IW statistically calculates the error between the predicted air temperature value within the future Δt time by the deep learning model on the test set and the true air temperature value. The training samples will be divided into a training set, a validation set, and a test set in a ratio of 6:2:2. The training set and the validation set participate in model training. When the model on the test set satisfies r < 0.1 and -5% < ICP - confidence level τ < 5% and IW < 50°C, it is considered to meet the prediction accuracy requirements, and the training is terminated. The deep learning model with the highest proportion of high-precision test samples is used as the final surrogate model.

[0043] This embodiment also provides an electronic device, including a memory and a processor. The processor is used to execute the program in the memory to implement the real-time identification and prediction method of fire source parameters and air temperature as described above. The processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components; the memory may include a random access memory (RAM), and may also include a non-volatile memory, such as at least one disk memory. The memory can be an internal memory of the random access memory (RAM) type. The processor and the memory can be integrated into one or more independent circuits or hardware, such as: an application-specific integrated circuit (ASIC). It should be noted that when the computer program in the above memory is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to enable a computer device (which can be a personal computer, an electronic device, or a network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present invention.

[0044] This embodiment also provides a storage medium containing computer-executable instructions. When executed by a computer processor, the storage medium with computer-executable instructions is used to perform the real-time identification and prediction method for fire source parameters and air temperature as described above. The storage medium can be an electronic medium, magnetic medium, optical medium, electromagnetic medium, infrared medium, or semiconductor system or propagation medium. The storage medium can also include semiconductor or solid-state memory, magnetic tape, removable computer disk, random access memory (RAM), read-only memory (ROM), hard disk, and optical disk. Optical disks can include optical disc-read-only memory (CD-ROM), optical disc-read / write (CD-RW), and DVD.

[0045] The above description of the embodiments is provided to enable those skilled in the art to understand and use the invention. It will be apparent to those skilled in the art that various modifications can be made to these embodiments, and the general principles described herein can be applied to other embodiments without inventive effort. Therefore, the present invention is not limited to the above embodiments, and any improvements and modifications made by those skilled in the art based on the disclosure of the present invention without departing from the scope of the invention should be within the protection scope of the present invention.

Claims

1. A method for real-time identification and prediction of ignition source parameters and ambient temperature, characterized in that, Includes the following steps: S1: Determine the probability distribution of the geometric parameters and fire source state parameters of a single-story factory building during a fire, perform random sampling based on the determined probability distribution, and establish a numerical model of the single-story factory building. S2: Perform fire numerical simulation on the established numerical model, and obtain single-story factory fire thermal response data for training deep learning agent models based on the numerical simulation. S3: Determine the measurement location for air temperature in a single-story factory building; S4: Construct a deep learning agent model based on recurrent neural networks and fully connected neural networks; S5: Train the deep learning proxy model by taking the thermal response data of a single-story factory building fire and the geometric parameters of the single-story factory building during the fire as inputs, and the current fire source state parameters and future air temperature as outputs, until the prediction accuracy requirements are met. The deep learning proxy model is trained using root mean square error (RMSE) and quantile error. The loss function for the fire source parameter identification module is set to RMSE, and the loss function for the air temperature prediction module is set to quantile error. The total error is the weighted sum of RMSE and quantile error. The confidence level is used in the quantile error calculation. t Confidence level with the required predicted future air temperature t Consistent, the backpropagation algorithm is used to update the learnable parameters in the deep learning agent model; Using relative error r Interval coverage ICP and interval width IW Evaluate the performance of the trained deep learning model on the test set, and define... r < 0.1 and -5% < ICP -Confidence t < 5% and IW < 50 ℃ is the criterion for judging high-precision samples, and the deep learning model with the highest proportion of high-precision test samples is used as the final surrogate model; S6: The trained deep learning proxy model is used as the final proxy model for the identification of fire source state parameters and prediction of future air temperature in a single-story factory building in an actual fire scene. Based on the measured air temperature data and geometric parameters of the single-story factory building, the data is input into the pre-trained deep learning proxy model. Then, the deep learning proxy model identifies the fire state parameters in real time based on the input data and outputs a future air temperature prediction value with confidence. In S2 and S5, the single-story factory building fire thermal response data includes the time history curves of air temperature at key temperature measuring points throughout the fire process and the time history curves of fire source state parameters throughout the fire process. In S4, S5, and S6, the deep learning proxy model includes a fire source parameter identification module and an air temperature prediction module. The inputs to the fire source parameter identification module are the historical measured air temperature time history curve and the geometric parameters of the single-story factory building, and the outputs are the actual fire model of the single-story factory building and the fire source status parameters. The input to the air temperature prediction module is a realistic fire model of a single-story factory building, and the output is a future fire prediction model of a single-story factory building with a certain confidence level. t Air temperature prediction time history curve.

2. The method for real-time identification and prediction of fire source parameters and air temperature according to claim 1, characterized in that, In S1, the geometric parameters of the single-story factory building include span, column spacing, number of frames, and eaves height; The probability distribution of the fire source state parameters is determined according to the design specifications of a single-story factory building; The fire source status parameters include the ignition location, the area affected by the fire, and the heat release rate per unit area.

3. The method for real-time identification and prediction of fire source parameters and air temperature according to claim 1, characterized in that, In S2, the numerical simulation is performed using FDS or Fluent fire analysis software.

4. The method for real-time identification and prediction of fire source parameters and air temperature according to claim 1, characterized in that, In S6, the air temperature data includes the air temperature at 3 / 4 height of the factory column, 1 / 4 and 3 / 4 span of the beam; The fire source status parameters include the relative position of the fire source center in the span direction and column spacing direction at the current moment, and the total heat release power of the fire source; The air temperature is measured by pre-embedding thermocouples during the construction of the factory.

5. An electronic device, comprising a memory and a processor, characterized in that, The processor is used to execute the program in the memory to realize the real-time identification and prediction method of fire source parameters and air temperature as described in any one of claims 1 to 4.

6. A storage medium containing computer-executable instructions, characterized in that, When the storage medium containing the computer-executable instructions is executed by a computer processor, it is used to perform the real-time identification and prediction method of fire source parameters and air temperature as described in any one of claims 1 to 4.

Images