A method and device for predicting a fire temperature field of a wooden gallery bridge and an electronic device

Abstract

This application discloses a method, device, and electronic device for predicting the temperature field of a wooden covered bridge fire, belonging to the field of fire temperature field prediction technology. The method includes: acquiring the temperature field sequence data of the ceiling of the wooden covered bridge in the current moment; inputting the temperature field sequence data into a pre-trained spatiotemporal prediction model of the temperature field of a wooden covered bridge fire to obtain predicted temperature field sequence data representing a certain time period after the current moment; and generating early warning information for a wooden covered bridge fire based on the predicted temperature field sequence data. This method solves the technical bottleneck of traditional methods in balancing speed and accuracy, and fills a gap in existing research in this field by specifically modeling fire scenarios in open-space wooden covered bridges.

Description

Technical Field

[0001] This application relates to the field of fire temperature field prediction technology, and more specifically, to a method, apparatus and electronic equipment for predicting the temperature field of a fire on a wooden covered bridge. Background Technology

[0002] With increasing awareness of cultural heritage protection, the fire safety of wooden covered bridges, as an architectural form carrying profound historical and cultural significance, is receiving growing attention. Wooden covered bridges use mortise and tenon joints to construct their wooden frames, providing excellent resistance to earthquakes, bending, and wind. However, their structure, primarily composed of combustible wood, presents them with a severe challenge in the event of a fire. Once a fire breaks out, it can easily spread and ignite the entire bridge. Furthermore, since wooden covered bridges are often located in rural villages and other suburban areas, firefighting is difficult and can easily cause irreversible damage to their historical value.

[0003] Currently, fire evolution prediction methods mainly include empirical formulas and fluid dynamics simulation software. Empirical formulas have shorter calculation times, but the models are relatively simple, suitable for buildings with open spaces, and the accuracy of the results is limited. While fluid dynamics simulation software can improve calculation accuracy, its computational performance is low, making it difficult to meet the needs of rapid prediction. In recent years, neural network methods have been introduced into temperature field prediction, significantly improving computational efficiency and partially achieving real-time prediction capabilities. However, due to the complexity of the fire scene environment and the limited acquisition of additional information, this method suffers from error accumulation in long-term iterative predictions, and its application in actual fire rescue and emergency management still requires further optimization. In addition, existing research focuses more on enclosed space scenarios such as tunnels, rooms, and steel structures, with relatively few studies on fire scenarios involving wooden covered bridges in open spaces. Summary of the Invention

[0004] This application provides a method, apparatus, and electronic device for predicting the temperature field of a fire on a wooden covered bridge, in order to at least solve the technical problems of insufficient prediction accuracy, low computational efficiency, and long-term accumulation of iteration errors in related technologies for fire scenarios on wooden covered bridges in open spaces.

[0005] According to a first aspect of the embodiments of this application, a method for predicting the temperature field of a fire on a wooden covered bridge is provided, comprising:

[0006] Obtain the temperature field sequence data of the ceiling of the wooden covered bridge in the current moment;

[0007] The temperature field sequence data is input into a pre-trained spatiotemporal prediction model for the temperature field of a fire on a wooden covered bridge, and the predicted temperature field sequence data representing a certain time after the current moment is obtained.

[0008] Fire early warning information for wooden covered bridges is generated based on predicted temperature field sequence data.

[0009] This approach utilizes a pre-trained spatiotemporal prediction model for the temperature field in wooden covered bridge fires to analyze real-time collected temperature field sequence data, enabling rapid generation of temperature field change trends over a future period. Combining the prediction results with fire early warning information provides a scientific basis for fire prevention and control in wooden covered bridges. This method significantly improves computational efficiency while maintaining prediction accuracy, overcoming the technical bottleneck of traditional methods that struggle to balance speed and accuracy. Furthermore, targeted modeling of fire scenarios in open-space wooden covered bridges fills a gap in existing research in this area.

[0010] In conjunction with the first aspect, in one optional implementation of the embodiments of this application, temperature field sequence data is input into a pre-trained spatiotemporal prediction model for the temperature field of a wooden covered bridge fire to obtain predicted temperature field sequence data representing a certain time period after the current moment, including:

[0011] Determine the highest temperature distribution value on the ceiling of the wooden corridor bridge based on temperature field sequence data;

[0012] When the highest distribution value is greater than the preset temperature threshold, the temperature field sequence data is input into the pre-trained spatiotemporal prediction model of the temperature field of the wooden covered bridge fire to obtain the predicted temperature field sequence data representing a certain time after the current moment.

[0013] This solution first extracts and judges the highest temperature distribution value in the temperature field sequence data. When it exceeds a preset threshold, the prediction process is triggered. This design avoids redundant calculations for low-risk scenarios, improving the system's operating efficiency. Simultaneously, by focusing on the changing trends of high-temperature areas, the prediction results are enhanced in terms of relevance and practicality, providing more accurate data support for subsequent fire early warning systems.

[0014] In conjunction with the first aspect, in an optional implementation of this application embodiment, the spatiotemporal prediction model for the temperature field of a wooden covered bridge fire is trained using the following method:

[0015] Through fire simulation, the target temperature field dataset of the ceiling of the wooden bridge fire model under different fire source locations is obtained. The target temperature field dataset contains the target temperature field sequence data of the ceiling of the wooden bridge fire model within a certain time period. The time corresponding to the target temperature field sequence data is the annotation information of the target temperature field dataset.

[0016] Input the target temperature field sequence data into the initialized spatiotemporal prediction model of the temperature field of the wooden bridge fire, and compare the annotation information with the target time output by the initialized spatiotemporal prediction model of the temperature field of the wooden bridge fire.

[0017] The spatiotemporal prediction model of the temperature field of a wooden covered bridge fire is trained and initialized based on the difference between the target time and the labeled information, thus obtaining the spatiotemporal prediction model of the temperature field of a wooden covered bridge fire.

[0018] This approach utilizes fire simulation to construct a target temperature field dataset, ensuring the realism and diversity of the training data. By inputting the target temperature field sequence data into the initial model and comparing the labeled information with the target time output by the model, the model parameters are progressively optimized, ultimately achieving high-precision prediction of temperature field changes in a wooden covered bridge fire. This training process fully considers the unique characteristics of wooden covered bridge fire scenarios, enabling the model to adapt to complex and ever-changing fire environments.

[0019] In conjunction with the first aspect, in one optional implementation of the embodiments of this application, the spatiotemporal prediction model of the temperature field of a wooden covered bridge fire is a long-term discrete prediction model. After obtaining the target temperature field dataset of the ceiling of the wooden covered bridge fire model at different fire source locations through fire simulation, the method includes:

[0020] The target temperature field dataset is sampled to obtain discrete temperature field sequence data and corresponding annotation information.

[0021] Discrete temperature field sequence data is used as target temperature field sequence data.

[0022] This approach extracts discrete temperature field sequence data and its annotation information by sampling the target temperature field dataset, providing structured input data for training a long-term discrete prediction model. This design effectively reduces the complexity of data processing while preserving key time-series features, ensuring the model's predictive ability over long time spans.

[0023] In conjunction with the first aspect, in an optional implementation of this application embodiment, the spatiotemporal prediction model for the temperature field of a wooden covered bridge fire includes a long-term discrete prediction sub-model and an instantaneous continuous prediction sub-model. After obtaining the target temperature field dataset of the ceiling of the wooden covered bridge fire model at different fire source locations through fire simulation, the method includes:

[0024] The target temperature field dataset is sampled to obtain discrete temperature field sequence data, the first annotation information corresponding to the discrete temperature field sequence data, continuous temperature field sequence data, and the second annotation information corresponding to the continuous temperature field sequence data.

[0025] This approach decomposes the target temperature field dataset into discrete and continuous temperature field sequences, and labels each sequence with corresponding information, providing a foundation for the joint training of long-term discrete prediction sub-models and instantaneous continuous prediction sub-models. This design fully leverages the advantages of both models, capturing both long-term temperature change trends and accurately describing short-term temperature fluctuations, thus achieving comprehensive prediction of the temperature field in wooden covered bridge fires.

[0026] In conjunction with the first aspect, in one optional implementation of the embodiments of this application, the target temperature field sequence data is input into the initialized spatiotemporal prediction model of the temperature field of a wooden covered bridge fire, and the annotation information is compared with the target time corresponding to the output of the initialized spatiotemporal prediction model of the temperature field of a wooden covered bridge fire, including:

[0027] Input the discrete temperature field sequence data into the initialized long-term discrete prediction sub-model, and compare the first annotation information with the first target time output by the initialized long-term discrete prediction sub-model.

[0028] The continuous temperature field sequence data is input into the initialized instantaneous continuous prediction sub-model, and the second annotation information is compared with the second target time output by the initialized instantaneous continuous prediction sub-model.

[0029] This approach trains the long-term discrete prediction sub-model and the instantaneous continuous prediction sub-model independently, ensuring optimized performance for each model on their respective tasks. By comparing the labeled information with the target time output by the model, the model parameters are gradually adjusted, ultimately achieving accurate prediction of the temperature field in a wooden covered bridge fire. This design significantly improves the model's prediction accuracy and robustness.

[0030] In conjunction with the first aspect, in an optional implementation of the embodiments of this application, the method further includes:

[0031] Through fire simulation, the verification temperature field dataset of the ceiling of the wooden bridge fire model under different fire source locations is obtained. The verification temperature field dataset contains the verification temperature field sequence data of the ceiling of the wooden bridge fire model over a certain period of time, as well as the verification time corresponding to the verification temperature field sequence data.

[0032] The accuracy of the spatiotemporal prediction model for temperature field in wooden covered bridge fires was verified using a validation temperature field dataset.

[0033] This approach validates the model's predictions by constructing a verification temperature field dataset, ensuring the model's effectiveness in practical applications. This design not only assesses the model's prediction accuracy but also identifies potential sources of error, providing a basis for further model optimization.

[0034] In conjunction with the first aspect, in one optional implementation of the embodiments of this application, generating fire early warning information for wooden covered bridges based on predicted temperature field sequence data includes:

[0035] Based on the predicted temperature field sequence data, the warning time when the temperature at any location on the ceiling of the wooden covered bridge scene exceeds the temperature threshold is determined.

[0036] Fire warning information for wooden covered bridges is generated based on the warning time.

[0037] This solution analyzes predicted temperature field sequence data to determine the time point at which any location on the ceiling of a wooden covered walkway reaches the high-temperature threshold, and generates fire early warning information accordingly. This design can identify potential fire risks in advance, buying valuable time for emergency response and significantly improving the timeliness and effectiveness of fire prevention and control in wooden covered walkways.

[0038] According to a second aspect of the embodiments of this application, a device for predicting the temperature field of a fire on a wooden covered bridge is provided, comprising:

[0039] The acquisition unit is used to acquire the temperature field sequence data of the ceiling of the wooden covered bridge in the current moment.

[0040] The processing unit is used to input the temperature field sequence data into the pre-trained spatiotemporal prediction model of the temperature field of the wooden covered bridge fire, and obtain the predicted temperature field sequence data representing a certain time after the current moment.

[0041] The early warning unit is used to generate fire early warning information for wooden covered bridges based on predicted temperature field sequence data.

[0042] In conjunction with the second aspect, in one optional implementation of the embodiments of this application, the processing unit is specifically used for:

[0043] Determine the highest temperature distribution value on the ceiling of the wooden corridor bridge based on temperature field sequence data;

[0044] When the highest distribution value is greater than the preset temperature threshold, the temperature field sequence data is input into the pre-trained spatiotemporal prediction model of the temperature field of the wooden covered bridge fire to obtain the predicted temperature field sequence data representing a certain time after the current moment.

[0045] In conjunction with the second aspect, in an optional implementation of this application embodiment, the processing unit specifically trains the spatiotemporal prediction model of the temperature field for fires in wooden covered bridges using the following method:

[0046] Through fire simulation, the target temperature field dataset of the ceiling of the wooden bridge fire model under different fire source locations is obtained. The target temperature field dataset contains the target temperature field sequence data of the ceiling of the wooden bridge fire model within a certain time period. The time corresponding to the target temperature field sequence data is the annotation information of the target temperature field dataset.

[0047] Input the target temperature field sequence data into the initialized spatiotemporal prediction model of the temperature field of the wooden bridge fire, and compare the annotation information with the target time output by the initialized spatiotemporal prediction model of the temperature field of the wooden bridge fire.

[0048] The spatiotemporal prediction model of the temperature field of a wooden covered bridge fire is trained and initialized based on the difference between the target time and the labeled information, thus obtaining the spatiotemporal prediction model of the temperature field of a wooden covered bridge fire.

[0049] In conjunction with the second aspect, in one optional implementation of the embodiments of this application, the processing unit is specifically used for:

[0050] The target temperature field dataset is sampled to obtain discrete temperature field sequence data and corresponding annotation information.

[0051] Discrete temperature field sequence data is used as target temperature field sequence data.

[0052] In conjunction with the second aspect, in an optional implementation of this application's embodiments, the spatiotemporal prediction model for the temperature field of a wooden covered bridge fire includes a long-term discrete prediction sub-model and an instantaneous continuous prediction sub-model. The processing unit is specifically used for:

[0053] The target temperature field dataset is sampled to obtain discrete temperature field sequence data, the first annotation information corresponding to the discrete temperature field sequence data, continuous temperature field sequence data, and the second annotation information corresponding to the continuous temperature field sequence data.

[0054] In conjunction with the second aspect, in one optional implementation of the embodiments of this application, the processing unit is specifically used for:

[0055] Input the discrete temperature field sequence data into the initialized long-term discrete prediction sub-model, and compare the first annotation information with the first target time output by the initialized long-term discrete prediction sub-model.

[0056] The continuous temperature field sequence data is input into the initialized instantaneous continuous prediction sub-model, and the second annotation information is compared with the second target time output by the initialized instantaneous continuous prediction sub-model.

[0057] In conjunction with the second aspect, in an optional implementation of the embodiments of this application, the processing unit is further configured to:

[0058] Through fire simulation, the verification temperature field dataset of the ceiling of the wooden bridge fire model under different fire source locations is obtained. The verification temperature field dataset contains the verification temperature field sequence data of the ceiling of the wooden bridge fire model over a certain period of time, as well as the verification time corresponding to the verification temperature field sequence data.

[0059] The accuracy of the spatiotemporal prediction model for temperature field in wooden covered bridge fires was verified using a validation temperature field dataset.

[0060] In conjunction with the second aspect, in one optional implementation of the embodiments of this application, the early warning unit is specifically used for:

[0061] Based on the predicted temperature field sequence data, the warning time when the temperature at any location on the ceiling of the wooden covered bridge scene exceeds the temperature threshold is determined.

[0062] Fire warning information for wooden covered bridges is generated based on the warning time.

[0063] According to a third aspect of the embodiments of this application, the present invention provides an electronic device, including: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to perform the flexible control method of the second aspect or any corresponding embodiment described above.

[0064] According to a fourth aspect of the embodiments of this application, the embodiments of this specification provide a computer-readable storage medium storing computer instructions, which, when executed by a processor, implement the method for predicting the temperature field of a fire on a wooden covered bridge as described in any of the preceding claims.

[0065] According to a fifth aspect of the embodiments of this application, this specification provides a computer program product or computer program, the computer program product including a computer program stored in a computer-readable storage medium; a processor of the computer device reads the computer program from the computer-readable storage medium, and when the processor executes the computer program, it implements the method for predicting the temperature field of a wooden covered bridge fire as described in any of the preceding claims.

[0066] The technical effects achieved by the second to fifth aspects mentioned above are similar to those achieved by the corresponding technical means in the first aspect, and will not be repeated here. Attached Figure Description

[0067] Figure 1 This is a flowchart illustrating the method for predicting the temperature field of a wooden covered bridge fire provided in an embodiment of this application.

[0068] Figure 2 This is a flowchart illustrating a training method for a spatiotemporal prediction model of the temperature field in a wooden covered bridge fire, as provided in an embodiment of this application.

[0069] Figure 3 This is a schematic diagram illustrating the principle of gated spatiotemporal attention provided in an embodiment of this application;

[0070] Figure 4 This is a flowchart illustrating another method for training a spatiotemporal prediction model of the temperature field in a wooden covered bridge fire, as provided in an embodiment of this application.

[0071] Figure 5 This is a schematic diagram of the structure of the temperature field prediction device for a wooden covered bridge fire provided in this application embodiment;

[0072] Figure 6 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0073] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0074] It should be understood that "multiple" as mentioned herein refers to two or more. In the description of the embodiments of this application, unless otherwise stated, " / " means "or," for example, A / B can mean A or B; "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. In addition, to facilitate a clear description of the technical solutions of the embodiments of this application, the terms "first," "second," etc., are used in the embodiments of this application to distinguish identical or similar items with substantially the same function and effect. Those skilled in the art will understand that the terms "first," "second," etc., do not limit the quantity or execution order, and the terms "first," "second," etc., do not necessarily imply that they are different.

[0075] Furthermore, the terms “comprising” and “having”, and any variations thereof, are intended to cover non-exclusive inclusion, such that a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such process, method, product, or apparatus.

[0076] As mentioned in the background section, with increasing awareness of cultural heritage protection, the fire safety of wooden covered bridges, as an architectural form carrying profound historical and cultural significance, has received growing attention. Wooden covered bridges utilize mortise and tenon joints to construct their wooden frames, providing excellent earthquake, bending, and wind resistance. However, their structure, primarily composed of combustible wood, presents them with significant challenges in fires. Once a fire breaks out, it can easily spread and ignite the entire bridge. Furthermore, since wooden covered bridges are often located in rural villages and other suburban areas, firefighting efforts are difficult, and irreversible damage to their historical value is highly likely.

[0077] Currently, fire evolution prediction methods mainly include empirical formulas and fluid dynamics simulation software. Empirical formulas have shorter calculation times, but the models are relatively simple, suitable for buildings with open spaces, and the accuracy of the results is limited. While fluid dynamics simulation software can improve calculation accuracy, its computational performance is low, making it difficult to meet the needs of rapid prediction. In recent years, neural network methods have been introduced into temperature field prediction, significantly improving computational efficiency and partially achieving real-time prediction capabilities. However, due to the complexity of the fire scene environment and the limited acquisition of additional information, this method suffers from error accumulation in long-term iterative predictions, and its application in actual fire rescue and emergency management still requires further optimization. In addition, existing research focuses more on enclosed space scenarios such as tunnels, rooms, and steel structures, with relatively few studies on fire scenarios involving wooden covered bridges in open spaces.

[0078] This invention provides a method for predicting the temperature field in a wooden covered bridge fire. The method involves analyzing real-time collected temperature field sequence data using a pre-trained spatiotemporal prediction model for the temperature field in a wooden covered bridge fire, thereby generating a temperature field change trend over a future period. The prediction results are then combined to generate fire early warning information, such as... Figure 1 As shown, it includes the following steps:

[0079] S101: Obtain the temperature field sequence data of the wooden covered bridge ceiling in the current moment.

[0080] Specifically, in practice, the first step is to acquire the temperature field sequence data of the wooden walkway ceiling at the current moment. This process is accomplished by temperature sensors, which are typically placed in several key locations on the wooden walkway ceiling, such as areas where fire sources might occur, near ventilation openings, and structural weak points. These sensors are connected to the data acquisition module via wired or wireless means to ensure real-time data transmission. The data acquisition module preprocesses the received raw data, including filtering and noise reduction, time synchronization, and formatting, ultimately generating a temperature field sequence data that can be used for subsequent analysis. The temporal and spatial resolutions of the temperature field sequence data need to be set according to actual requirements, for example, a temporal resolution of 1 second and a spatial resolution of 0.5 meters × 0.5 meters, to ensure the accuracy and usability of the data.

[0081] S102: Input the temperature field sequence data into the pre-trained spatiotemporal prediction model of the temperature field of the wooden covered bridge fire to obtain the predicted temperature field sequence data representing a certain time after the current moment.

[0082] In practice, temperature field sequence data is input into a pre-trained spatiotemporal prediction model for the temperature field of a wooden covered bridge fire. The training process of this model is one of the core components of the entire method. To ensure the model's prediction accuracy, it is first necessary to obtain target temperature field datasets for the ceiling of the wooden covered bridge fire model under different fire source locations through fire simulation. The fire simulation uses professional fire dynamics software, such as FDS (Fire Dynamics Simulator), to generate diverse target temperature field datasets by setting different fire source locations, fire intensities, and environmental conditions. The target temperature field dataset contains target temperature field sequence data for the ceiling of the wooden covered bridge fire model over a certain period of time, along with corresponding annotation information. The annotation information includes the timestamp corresponding to each target temperature field sequence data, used for comparison during subsequent model training.

[0083] After acquiring the target temperature field dataset, it needs to be sampled to extract discrete and continuous temperature field sequence data. The discrete temperature field sequence data is primarily used for training the long-term discrete prediction sub-model, while the continuous temperature field sequence data is used for training the instantaneous continuous prediction sub-model. The sampling process must follow certain rules, such as extracting a discrete data point at fixed time intervals while retaining high-resolution data within the continuous time period. The sampled data are then labeled with first and second annotation information, where the first annotation information corresponds to the timestamp of the discrete temperature field sequence data, and the second annotation information corresponds to the timestamp of the continuous temperature field sequence data.

[0084] Subsequently, discrete temperature field sequence data is input into the initialized long-term discrete prediction sub-model, and the first annotation information is compared with the first target time output by the model. Similarly, continuous temperature field sequence data is input into the initialized instantaneous continuous prediction sub-model, and the second annotation information is compared with the second target time output by the model. By calculating the difference between the target time and the annotation information, the model parameters are gradually adjusted until the model's prediction error reaches a predetermined threshold. This training process fully considers the special characteristics of the wooden covered bridge fire scenario, enabling the model to adapt to complex and ever-changing fire environments.

[0085] After model training, it is applied to the actual prediction process. When temperature field sequence data is input into the model, the highest temperature distribution value on the ceiling of the wooden walkway needs to be determined first. After triggering the prediction process, the model outputs predicted temperature field sequence data representing a certain time period after the current moment. This process is achieved by scanning the temperature field sequence data point by point to find the highest temperature value on the ceiling at the current moment and its corresponding location. If the highest distribution value is greater than a preset temperature threshold, the prediction process is triggered; otherwise, monitoring of temperature field data continues. The preset temperature threshold can be set according to needs, such as 30 degrees or 50 degrees. This design avoids redundant calculations for low-risk scenarios and improves the system's operating efficiency.

[0086] S103: Generate fire early warning information for wooden covered bridges based on predicted temperature field sequence data.

[0087] In practice, the prediction duration can be set according to actual needs, such as 30 minutes or 1 hour. The predicted temperature field sequence data contains the temperature change trend of various locations on the ceiling of the wooden walkway over a future period. Based on this data, fire early warning information for the wooden walkway can be further generated. Specifically, by analyzing the predicted temperature field sequence data, the warning time when the temperature at any location on the ceiling of the wooden walkway exceeds the temperature threshold is determined. The method for calculating the warning time is to scan the predicted temperature field sequence data point by point, find the time point when the temperature first exceeds the temperature threshold at each location, and record these time points as the warning time. Subsequently, specific fire early warning information is generated based on the warning time, such as alerting relevant personnel through audible and visual alarm devices, or sending a warning notification to the fire department through a communication network.

[0088] In this step, to further improve prediction accuracy and system robustness, a multi-source data fusion mechanism can be introduced to dynamically correct the predicted temperature field sequence data output by the model. Specifically, after acquiring the predicted temperature field sequence data, the system simultaneously collects environmental parameters under the wooden walkway scenario, including key factors such as air circulation, wind speed, humidity, ambient temperature, and the current moisture content of the wood material. These environmental parameters are acquired in real time by various sensors (such as anemometers, hygrometers, and integrated temperature and humidity sensors) deployed around and inside the wooden walkway structure, and transmitted to the central data processing module for spatiotemporal alignment and fusion analysis. Environmental parameters have a significant impact on the fire spread rate, heat release rate, and the formation and development of roof jets. For example, under dry, high-temperature, and windy conditions, air entrainment is enhanced, the fire spreads faster, the temperature rises more rapidly, and the high-temperature area expands more widely; while in environments with high air humidity or near-no wind, thermal convection is weakened, and the fire develops relatively slowly. Based on this physical characteristic, a temperature field correction model integrating environmental parameters was constructed. This model is a lightweight neural network, trained through correlation analysis of a large amount of historical fire simulation data and environmental parameters. It can dynamically and nonlinearly adjust the predicted temperature field sequence according to real-time environmental data vectors. The correction process specifically includes: First, the real-time collected environmental parameters are preprocessed and normalized to form an environmental feature tensor that matches the spatiotemporal dimension of the predicted temperature field. Then, this environmental feature tensor and the original predicted temperature field sequence data are input into the correction model. The correction model, through an internally embedded attention mechanism, identifies the most critical environmental factors to the evolution of the current temperature field (such as identifying changes in the smoke transport path under strong winds) and outputs a temperature field correction coefficient matrix with spatiotemporal heterogeneity. This matrix accurately reflects the enhancing or inhibiting effects of different environmental factors on temperature field changes at different spatial locations and at different temporal stages. Finally, the correction coefficient matrix and the original predicted temperature field sequence data are fused point by point to obtain a corrected predicted temperature field sequence data that is closer to the real physical process. Through this refined fusion process, the system can significantly improve its predictive robustness under complex and variable weather conditions. In particular, under extreme or sudden weather conditions, it can more accurately reflect the actual fire development trend and effectively overcome the potential shortcomings of pure data-driven models in terms of physical consistency.

[0089] It should be noted that the content of this multi-source data integration can be used to correct the predicted temperature field sequence data after the model output, or it can be directly set in the model so that the model refers to this part of the multi-source data when outputting. This disclosure does not limit this aspect.

[0090] Based on the corrected predicted temperature field sequence data, the system further generates more forward-looking and reliable fire early warning information for wooden covered bridges. The corrected data not only includes the spatiotemporal evolution trend of the temperature field itself, but also incorporates the modulation effect of environmental factors on fire dynamics at a deeper level, making the generation logic of the early warning information more rigorous and scientific. The system performs in-depth analysis of the corrected predicted temperature field sequence data. In addition to performing the original point-by-point scanning to determine the warning time when each location first exceeds the temperature threshold, it also comprehensively evaluates the expansion rate of high-temperature areas, the merging trend of high-temperature clusters, and the time of reaching critical danger temperatures (such as the ignition point of wood materials). For example, under dry and windy conditions, the corrected model may show that a high-temperature point on the ceiling will reach the danger temperature in 8 minutes, instead of the original prediction of 12 minutes, and the system will trigger a higher-level warning accordingly. The generated early warning information is a structured data volume, which includes not only the warning level (such as attention, warning, alarm), the core warning time (i.e., the earliest time point to reach the danger threshold), but also the direction of fire spread, the expected scope of impact, and the identification of key risk locations. This information is clearly presented to monitoring personnel through a human-computer interaction interface in a visual manner (such as temperature field evolution animations and risk heat maps). Simultaneously, the core elements of the early warning information are automatically pushed to relevant emergency response units and responsible persons via communication networks, and can trigger on-site audible and visual alarm devices, emergency broadcast systems, and even automatically activate existing fire-fighting facilities (such as pre-filling pump units). This entire closed-loop process, from multi-source data fusion and model correction to intelligent early warning generation, greatly improves the timeliness, accuracy, and proactivity of fire prevention and control for wooden covered bridges, building a solid technological defense line for protecting irreplaceable historical and cultural heritage.

[0091] Furthermore, to verify the model's practical application effectiveness, a validation temperature field dataset needs to be constructed. The method for obtaining the validation temperature field dataset is similar to that of the target temperature field dataset, but it must be ensured that there is no data overlap between the two. The validation temperature field dataset contains the validation temperature field sequence data of the ceiling of the wooden covered bridge fire model over a certain period of time, along with the corresponding validation time. The model's output accuracy is evaluated using the validation temperature field dataset, calculating the error between the model's prediction and the validation time. If the error exceeds the allowable range, further optimization of the model is required, such as adjusting model parameters or increasing the amount of training data. For example, by deploying the above system in a wooden covered bridge in a scenic area, the temperature changes of the covered bridge ceiling can be monitored in real time, and early warning information can be issued in time before a fire occurs. This not only provides a scientific basis for the scenic area management department but also provides strong protection for the safety of tourists. In addition, this method can be extended to other similar open space buildings, such as wooden ancient buildings and exhibition halls, and has broad applicability and promotional value.

[0092] As can be seen from the above detailed description, the method for predicting the temperature field of a wooden covered bridge fire provided by this invention has a high degree of technical completeness and operability. From data acquisition to model training, and then to predictive analysis and early warning generation, each step has been carefully designed and rigorously verified to ensure the system's efficiency and reliability. Furthermore, the detailed explanation of each component and its collaborative relationships further enhances the feasibility of the technical solution, providing a clear operational guide for those skilled in the art.

[0093] In one example, the fire temperature field prediction system described in this invention was deployed in a wooden covered bridge in a scenic area. This wooden covered bridge is a typical mortise and tenon structure, with its ceiling constructed from multiple pieces of wood, featuring a large span and good ventilation. Because the wooden covered bridge is located in a rural area lacking adequate fire-fighting facilities, its fire prevention and control requirements are significantly higher. By deploying this system, the temperature changes of the wooden covered bridge's ceiling can be monitored in real time, and early warning information can be issued before a fire occurs, thus providing scientific evidence for scenic area management and ensuring the safety of tourists.

[0094] First, multiple temperature sensors were installed at key locations on the ceiling of the wooden walkway. These sensors were positioned in areas where fire sources might occur, near ventilation openings, and at structural weak points. For example, sensor 1 was placed at the edge of the ceiling near the visitor rest area, sensor 2 was placed directly below a ventilation opening, and sensor 3 was placed at the structural connection point in the middle of the wooden walkway. These sensors are wirelessly connected to a data acquisition module to ensure real-time data transmission. The data acquisition module preprocesses the received raw data, including filtering and noise reduction, time synchronization, and formatting, ultimately generating a temperature field sequence data that can be used for subsequent analysis. During this process, the data acquisition module has a time resolution of 1 second and a spatial resolution of 0.5 m × 0.5 m to ensure the accuracy and usability of the data.

[0095] Next, the temperature field sequence data is input into a pre-trained spatiotemporal prediction model for the temperature field of a fire on a wooden covered bridge. When the temperature field sequence data is input into the model, the highest temperature distribution value on the ceiling of the wooden covered bridge needs to be determined first. This process is achieved by scanning the temperature field sequence data point by point to find the highest temperature value on the ceiling at the current moment and its corresponding location. For example, if the temperature detected by sensor 2 at a certain moment is 85℃, which is higher than the detected values ​​of other sensors and exceeds the preset temperature threshold of 70℃, then the prediction process is triggered. Otherwise, the system continues to monitor the temperature field data. This design avoids redundant calculations for low-risk scenarios and improves the system's operating efficiency.

[0096] After triggering the prediction process, the model outputs a predicted temperature field sequence data representing a certain time period after the current moment. The prediction duration is set to 30 minutes. The predicted temperature field sequence data contains the temperature change trend of various locations on the ceiling of the wooden walkway over a future period. For example, the model prediction results show that the temperature at the location of sensor 2 will reach 120°C in 15 minutes, and the temperature at the location of sensor 3 will reach 90°C in 20 minutes. Based on this data, fire early warning information for the wooden walkway can be further generated. Specifically, by analyzing the predicted temperature field sequence data, the warning time when the temperature at any location on the ceiling of the wooden walkway exceeds the temperature threshold is determined. The method for calculating the warning time is to scan the predicted temperature field sequence data point by point, find the time point when the temperature first exceeds the temperature threshold at each location, and record these time points as the warning time. For example, the warning time at the location of sensor 2 is 15 minutes later, and the warning time at the location of sensor 3 is 20 minutes later. Subsequently, specific fire early warning information is generated based on the warning time, such as alerting relevant personnel through audible and visual alarm devices, or sending a warning notification to the fire department through a communication network.

[0097] In addition, to verify the model's practical application effectiveness, a validation temperature field dataset needs to be constructed. The method for obtaining the validation temperature field dataset is similar to that of the target temperature field dataset, but it must be ensured that there is no data overlap between the two. The validation temperature field dataset contains the validation temperature field sequence data of the ceiling of the wooden covered bridge fire model over a certain period of time, along with the corresponding validation times. The model's output accuracy is evaluated using the validation temperature field dataset by calculating the error between the model's prediction results and the validation times. If the error exceeds the allowable range, further optimization of the model is required, such as adjusting model parameters or increasing the amount of training data.

[0098] This invention also provides a training method for a spatiotemporal prediction model of the temperature field in a wooden covered bridge fire. In this method, the spatiotemporal prediction model of the temperature field in a wooden covered bridge fire is a long-term discrete prediction model, and the training method is as follows: Figure 2 As shown, it includes:

[0099] S201: Obtain the target temperature field dataset of the ceiling of a fire model of a wooden covered bridge under different fire source locations through fire simulation.

[0100] In practice, the target temperature field dataset contains the target temperature field sequence data of the ceiling of the wooden corridor bridge fire model within a certain period of time, and the time corresponding to the target temperature field sequence data is the annotation information of the target temperature field dataset.

[0101] When acquiring the target temperature field dataset, a fire model of the wooden corridor bridge was constructed through fire simulation, and different locations were selected as fire source points. Considering the feasibility of deploying temperature detectors in a real-world scenario, the ceiling of the wooden corridor bridge is easy to install and maintain, and offers wide coverage. Furthermore, during a fire, the heat plume rises vertically, impacting the ceiling and forming a ceiling jet, causing the ceiling temperature to rise. The ceiling temperature can effectively reflect the dynamic characteristics of the fire. Therefore, temperature slices were set at the ceiling of the wooden corridor bridge, selecting the ceiling cross-section as the temperature field slice object, recording the ceiling temperature data during combustion, and outputting the corresponding time-temperature point data.

[0102] After obtaining the time-temperature point data, it needs to be preprocessed to obtain the target temperature field dataset. Specifically, the temperature point data is converted into a temperature field tensor sequence to construct training and testing datasets. The temperature field tensor sequence is then sampled to obtain discrete temperature field sequence data. The long-term discrete prediction model data preprocessing first employs downsampling, setting the sampling interval. Every Sample data of length are used to form a new sparse sampling sequence. Based on this, the input length is set to . The temperature field sequence is predicted to have a subsequent length of The temperature field sequence is composed of a sliding window length, starting from the first slice of the downsampled sequence and moving one slice at a time. , This refers to discrete temperature field sequence data.

[0103] In one example, the temperature data corresponding to the recorded time points are rearranged and transformed into a continuous temperature field tensor sequence. The starting temperature field tensor sequence is selected according to the set warning temperature of the wooden walkway. Data is sampled separately to construct training and test datasets. Among all wooden walkway models, 75% of the models are assigned to the training dataset and 25% to the test dataset.

[0104] Discrete temperature field sequence data is obtained by intermittently sampling the temperature field tensor sequence. The long-term discrete prediction model data preprocessing first employs downsampling, setting the sampling interval... Every Sample data of length are used to form a new sparse sampling sequence. Based on this, the input length is set to . The temperature field sequence is predicted to have a subsequent length of The temperature field sequence is composed of a sliding window length, starting from the first slice of the downsampled sequence and moving one slice at a time. .

[0105] S202: Input the target temperature field sequence data into the initialized long-term discrete prediction model, and compare the annotation information with the target time output by the initialized long-term discrete prediction model.

[0106] S203: Train and initialize the long-term discrete prediction model based on the difference between the target time and the labeled information to obtain the long-term discrete prediction model.

[0107] In practical implementation, a long-term discrete prediction model is constructed based on discrete temperature field sequence data, employing a pure convolutional neural network structure of spatial encoder-spatiotemporal translator-spatial decoder. The attention mechanism is a dynamic weighting mechanism that assigns higher weights to key regions in the input sequence by calculating the correlation scores of each spatiotemporal location, thereby enhancing the model's ability to express important information without increasing the number of parameters. In cases such as... Figure 3 In the gated spatiotemporal attention model shown, a large receptive field is obtained through large kernel convolution, capturing correlations in both temporal and spatial dimensions. The output of the large kernel convolution is divided into two parts, one of which is passed through a sigmoid function as an attention gate. The formula for gated spatiotemporal attention is:

[0108] ;

[0109] ;

[0110] ;

[0111] ;

[0112] in These are input features. It is a two-dimensional convolution. Represents depthwise convolution. This represents depthwise dilated convolution. These are intermediate features after convolution; From The two parts split from the middle Used to generate gated signals that adaptively change based on input characteristics. Used to preserve original feature information; It's the Sigmoid function, which generates the gating coefficients. For element-wise multiplication, It is a gating signal right The output features after weighted processing It will adaptively filter out unimportant features from a spatiotemporal perspective.

[0113] Based on the gated spatiotemporal attention formula, discrete temperature field sequence data... Load the long-time discrete prediction model for training and obtain the weights of the long-time discrete prediction model. Specifically, the long-term discrete prediction model starts with the warning temperature slice of the downsampled sequence, and moves by one sliding window length at a time. The weights are trained as weights for a long-term discrete prediction model. .

[0114] The tensor changes at each stage are as follows:

[0115] ;

[0116] ;

[0117] in, , Indicates the time step. Indicates the number of channels. Indicates altitude, Indicates width, This indicates the number of convolutional layers in the spatial encoder and spatial decoder.

[0118] After training is completed, the ceiling temperature field data can be predicted based on the trained fusion model and the newly constructed test set of fire models of wooden corridor bridges with different fire sources, so as to obtain the long-term prediction results of the temperature field.

[0119] This invention also provides a training method for a spatiotemporal prediction model of the temperature field in a wooden covered bridge fire. In this method, the spatiotemporal prediction model of the temperature field in a wooden covered bridge fire includes a long-term discrete prediction sub-model and an instantaneous continuous prediction sub-model. The training method is as follows: Figure 4 As shown, it includes:

[0120] S401: Obtain the target temperature field dataset of the ceiling of a fire model of a wooden covered bridge under different fire source locations through fire simulation.

[0121] In practice, the target temperature field dataset contains the target temperature field sequence data of the ceiling of the wooden corridor bridge fire model within a certain period of time, and the time corresponding to the target temperature field sequence data is the annotation information of the target temperature field dataset.

[0122] When acquiring the target temperature field dataset, a fire model of the wooden corridor bridge was constructed through fire simulation, and different locations were selected as fire source points. Considering the feasibility of deploying temperature detectors in a real-world scenario, the ceiling of the wooden corridor bridge is easy to install and maintain, and offers wide coverage. Furthermore, during a fire, the heat plume rises vertically, impacting the ceiling and forming a ceiling jet, causing the ceiling temperature to rise. The ceiling temperature can effectively reflect the dynamic characteristics of the fire. Therefore, temperature slices were set at the ceiling of the wooden corridor bridge, selecting the ceiling cross-section as the temperature field slice object, recording the ceiling temperature data during combustion, and outputting the corresponding time-temperature point data.

[0123] After obtaining the time-temperature point data, it is necessary to preprocess it to obtain the target temperature field dataset. Specifically, the temperature point data is converted into a temperature field tensor sequence to construct the training dataset and the test dataset. The temperature field tensor sequence is sampled to obtain discrete temperature field sequence data and continuous temperature field sequence data.

[0124] The data preprocessing for the long-term discrete prediction sub-model first employs downsampling, setting the sampling interval. Every Sample data of length are used to form a new sparse sampling sequence. Based on this, the input length is set to . The temperature field sequence is predicted to have a subsequent length of The temperature field sequence is composed of a sliding window length, starting from the first slice of the downsampled sequence and moving one slice at a time. .

[0125] By sampling continuous data from the temperature field tensor sequence, a continuous temperature field sequence is obtained. This continuous data sampling is set to an input continuous length of... The temperature field is predicted to have a subsequent length of Continuous temperature field data. Starting from the first slice, each slice is composed of a sliding window that moves one position at a time. .

[0126] That is, discrete temperature field sequence data. This refers to continuous temperature field sequence data.

[0127] In one example, the temperature data corresponding to the recorded time points are rearranged and transformed into a continuous temperature field tensor sequence. The starting temperature field tensor sequence is selected according to the set warning temperature of the wooden walkway. Data is sampled separately to construct training and test datasets. Among all wooden walkway models, 75% of the models are assigned to the training dataset and 25% to the test dataset.

[0128] Discrete temperature field sequence data is obtained by discretely sampling the temperature field tensor sequence at intervals. The long-term discrete prediction sub-model data preprocessing first employs downsampling, setting the sampling interval... Every Sample data of length are used to form a new sparse sampling sequence. Based on this, the input length is set to . The temperature field sequence is predicted to have a subsequent length of The temperature field sequence is composed of a sliding window length, starting from the first slice of the downsampled sequence and moving one slice at a time. .

[0129] By sampling continuous data from the temperature field tensor sequence, a continuous temperature field sequence is obtained. This continuous data sampling is set to an input continuous length of... The temperature field is predicted to have a subsequent length of Continuous temperature field data. Starting from the first slice, each slice is composed of a sliding window that moves one position at a time. .

[0130] That is, discrete temperature field sequence data. This refers to continuous temperature field sequence data. Discrete temperature field sequence data. and continuous temperature field sequence data They will be used to train the long-term discrete prediction sub-model and the instantaneous continuous prediction sub-model, respectively.

[0131] S402: Input the target temperature field sequence data into the initialized spatiotemporal prediction model of the temperature field of the wooden covered bridge fire, and compare the annotation information with the target time output by the initialized spatiotemporal prediction model of the temperature field of the wooden covered bridge fire.

[0132] S403: Train and initialize the spatiotemporal prediction model of the temperature field of the wooden covered bridge fire based on the difference between the target time and the labeled information, and obtain the spatiotemporal prediction model of the temperature field of the wooden covered bridge fire.

[0133] In practical implementation, a long-term discrete prediction sub-model is constructed based on discrete temperature field sequence data, employing a pure convolutional neural network structure of spatial encoder-spatiotemporal translator-spatial decoder. The attention mechanism is a dynamic weighting mechanism that assigns higher weights to key regions in the input sequence by calculating the correlation scores of each spatiotemporal location, thereby enhancing the model's ability to express important information without increasing the number of parameters. In cases such as... Figure 3 In the gated spatiotemporal attention model shown, a large receptive field is obtained through large kernel convolution, capturing correlations in both temporal and spatial dimensions. The output of the large kernel convolution is divided into two parts, one of which is passed through a sigmoid function as an attention gate. The formula for gated spatiotemporal attention is:

[0134] ;

[0135] ;

[0136] ;

[0137] ;

[0138] in These are input features. It is a two-dimensional convolution. Represents depthwise convolution. This represents depthwise dilated convolution. These are intermediate features after convolution; From The two parts split from the middle Used to generate gated signals that adaptively change based on input characteristics. Used to preserve original feature information; It's the Sigmoid function, which generates the gating coefficients. For element-wise multiplication, It is a gating signal right The output features after weighted processing It will adaptively filter out unimportant features from a spatiotemporal perspective.

[0139] The long-term discrete prediction sub-model starts with a slice of the warning temperature from the downsampled sequence, and moves by one sliding window length at a time. The weights are trained as long-term discrete prediction sub-model weights. .

[0140] The tensor changes at each stage are as follows:

[0141] ;

[0142] ;

[0143] in, , Indicates the time step. Indicates the number of channels. Indicates altitude, Indicates width, This indicates the number of convolutional layers in the spatial encoder and spatial decoder.

[0144] An instantaneous continuous spatiotemporal prediction sub-model is constructed based on continuous temperature field sequence data. It maintains the same structure as the long-term discrete prediction sub-model, but with different pre-training weights. The instantaneous continuous spatiotemporal prediction sub-model is composed of samples taken from continuous data, starting from the warning temperature slice and moving one sliding window at a time. The weights are trained as instantaneous continuous spatiotemporal prediction sub-models. .

[0145] The long-term discrete prediction sub-model and the instantaneous continuous prediction sub-model were trained separately to obtain a fusion model combining the trained long-term discrete prediction sub-model and the instantaneous continuous prediction sub-model, namely the spatiotemporal prediction model of the temperature field of the wooden covered bridge fire.

[0146] In one example, the discrete temperature field sequence data is processed according to the gated spatiotemporal attention formula. Load the long-time discrete prediction sub-model for training and obtain the weights of the long-time discrete prediction sub-model. Continuous temperature field sequence data Load the instantaneous continuous prediction sub-model for training and obtain the weights of the instantaneous continuous spatiotemporal prediction sub-model. During training, the tensor shape changes at each stage as follows:

[0147] ;

[0148] ;

[0149] in, , Indicates the time step. Indicates the number of channels. Indicates altitude, Indicates width, This indicates the number of convolutional layers in the spatial encoder and spatial decoder.

[0150] The long-term discrete prediction sub-model and the instantaneous continuous prediction sub-model were loaded and trained on the temperature field sequence of the wooden corridor bridge ceiling. The AdamW optimizer was selected, with independent weight decay handled. The learning rate (LR) was 0.0005, and a cosine annealing learning rate adjuster with a warmup phase was used. Training was conducted for two epochs (50 epochs per cycle, 100 epochs in total). The warmup learning rate was applied in the first 5 epochs of the first cycle, and the mean squared error (MSE) was chosen as the loss function. Various combinations of model hyperparameters were experimented with during training to achieve the optimal prediction performance. Finally, a fusion model combining the trained long-term discrete prediction sub-model and the instantaneous continuous prediction sub-model was obtained.

[0151] After training is completed, the ceiling temperature field data can be predicted based on the trained fusion model and the newly constructed test set of fire models of wooden corridor bridges with different fire sources, so as to obtain the long-term prediction results of the temperature field.

[0152] A new fire model for wooden covered bridges with different fire source locations was constructed to predict the ceiling temperature field. The long-term discrete prediction sub-model, upon reaching the warning temperature, will predict the warning time. The temperature field sequence is used as input data into the long-term discrete prediction sub-model. After 11 iterations, the discrete predicted temperature field sequence is obtained. .

[0153] Discrete predicted temperature field sequence The instantaneous continuous prediction sub-model is loaded, with each discrete temperature field slice serving as input to predict the corresponding continuous temperature field sequence data. The output is the instantaneous predicted temperature field sequence. .

[0154] Based on the discrete and instantaneous predicted temperature field sequences, the missing parts of the long-term discrete temperature field sequence data are filled by inserting the corresponding continuous predicted temperature field between each long-term discrete temperature field sequence data, thus forming a complete continuous temperature field prediction sequence and obtaining the long-term temperature field prediction results. The accuracy of the prediction results is evaluated using two metrics: structural similarity index (SSIM) and peak signal-to-noise ratio (PSNR).

[0155] Two baseline models, ConvLSTM and PredRNN, were included in the tests to compare the fusion model. The test dataset included a new fire source in the wooden covered bridge scene to examine the generalization ability of different models in this new scenario.

[0156] In the prediction process of the long-term discrete prediction sub-model, the temperature field tensor of each group of wooden covered bridge scenes is: In the prediction process of the fusion model, the first set of data obtained after the warning temperature is reached is first input into the trained long-term discrete prediction sub-model. Output the model As new input data, it is iterated 11 times and combined with the first set of data to obtain... The long-term discrete prediction sequence is then used. Subsequently, within this long-term discrete prediction sequence, each of the 120 predicted temperature field sequences will be copied four times to meet the input requirements of the instantaneous continuous prediction sub-model. All adjusted tensors are then... The input is the instantaneous continuous prediction sub-model, and the output fills in the missing parts of the long-term discrete temperature field sequence data, forming a complete continuous temperature field prediction sequence. Finally, a complete... Long-term temperature field prediction results.

[0157] Because ConvLSTM and PredRNN models lack long-term discrete prediction capabilities, a method similar to that used for instantaneous continuous prediction is employed. First, the first set of data obtained after the warning temperature is reached is input into the trained ConvLSTM and PredRNN models. Output the model As new input data, it was iterated 149 times and combined with the first set of data to obtain the complete set. Complete and continuous temperature field prediction sequence.

[0158] The predicted temperature field of the wooden covered bridge ceiling is shown in the table below.

[0159]

[0160] In the table, "*" indicates that ConvLSTM and PredRNN are the baseline models, and no long-term discrete prediction sub-model is used; instead, instantaneous continuous prediction is used. As shown in the table above, ConvLSTM, PredRNN, and the fusion model perform well in high temporal resolution scenarios such as wooden covered bridges. This section compares the performance metrics for predicting ceiling temperature field sequences. The 4-second instantaneous prediction refers to the ability to predict a subsequent 4-second temperature field sequence from a continuous 4-second temperature field on the test set. Regarding SSIM structural similarity, although the fusion model's metrics are slightly higher than the other two models, the differences between the three models are small. In terms of PSNR (Peak Signal-to-Noise Ratio), the fusion model performs slightly better. The 200-second, 400-second, and 600-second long-term predictions represent dividing a 600-second long-term sequence into 1-second to 200-second, 1-second to 400-second, and 1-second to 600-second intervals, respectively, examining the model's predictive ability for these different long-term sequences. The fusion model uses a combination of a long-term discrete prediction sub-model and an instantaneous continuous prediction sub-model, while the ConvLSTM and PredRNN models, unable to be predicted using this method, employ continuous sampling iterative prediction with a length of 4 seconds. The results show that in the wooden covered bridge scenario, the fusion model maintains high accuracy across the three prediction durations. The other benchmark models show some gaps compared to the fusion model. In the 600-second long-term prediction, the fusion model improves SSIM by 9.30% and PSNR by 18.28% compared to the PredRNN model. The fusion model performs better and has stronger instantaneous and long-term prediction capabilities.

[0161] The model was trained using data with a heat release rate per unit area (HRRPUA) of 125 kW / m² and a heat release rate (HRR) of 62.5 kW. To test the generalization ability of the fusion model, tests were conducted on test sets with HRRs of 50 kW, 75 kW, and 125 kW, and HRRPUAs of 100 kW / m², 150 kW / m², and 250 kW / m², respectively, as shown in the table below:

[0162]

[0163] The 62.5 kW heat release rate in the table represents the parameter setting for the training data scenario; heat release rates of 50 kW, 75 kW, and 125 kW did not appear in the training data. The prediction metrics in Table 2 show that the model possesses high generalization ability. In the three long-term predictions, compared to the original 62.5 kW heat release rate test set, the metrics for the test sets with a 20% increase (75 kW) and a 20% decrease (50 kW) heat release rate did not show a significant decrease, and the difference in prediction accuracy among the three test sets was small. The SSIM and PSNR metrics for the test set with a 100% increase (125 kW) heat release rate decreased significantly, with SSIM decreasing by 7.45% and PSNR decreasing by 19.64%. However, in the 600-second long-term prediction, SSIM remained above 0.85 and PSNR remained above 25, still yielding accurate prediction results. The results indicate that in the wooden covered bridge scenario, the fusion model maintained high accuracy in generalization tests across different heat release rates, demonstrating long-term prediction capability.

[0164] The above provides illustrative examples of the method embodiments according to this application. The present invention also provides a device for predicting the temperature field of a wooden covered bridge fire. Figure 5 This is a schematic diagram of a temperature field prediction device for a wooden covered bridge fire according to an embodiment of the present invention. (Refer to...) Figure 5 The 700 wooden covered bridge fire temperature field prediction device includes the following modules.

[0165] Acquisition unit 701 is used to acquire the temperature field sequence data of the wooden covered bridge ceiling in the current moment scene;

[0166] The processing unit 702 is used to input the temperature field sequence data into the pre-trained spatiotemporal prediction model of the temperature field of the wooden covered bridge fire, and obtain the predicted temperature field sequence data representing a certain time after the current moment.

[0167] The early warning unit 703 is used to generate fire early warning information for wooden covered bridges based on predicted temperature field sequence data.

[0168] In one optional implementation of this application embodiment, the processing unit 702 is specifically used for:

[0169] Determine the highest temperature distribution value on the ceiling of the wooden corridor bridge based on temperature field sequence data;

[0170] When the highest distribution value is greater than the preset temperature threshold, the temperature field sequence data is input into the pre-trained spatiotemporal prediction model of the temperature field of the wooden covered bridge fire to obtain the predicted temperature field sequence data representing a certain time after the current moment.

[0171] In an optional implementation of this application embodiment, the processing unit 702 specifically trains the spatiotemporal prediction model of the temperature field for fires in wooden covered bridges using the following method:

[0172] Through fire simulation, the target temperature field dataset of the ceiling of the wooden bridge fire model under different fire source locations is obtained. The target temperature field dataset contains the target temperature field sequence data of the ceiling of the wooden bridge fire model within a certain time period. The time corresponding to the target temperature field sequence data is the annotation information of the target temperature field dataset.

[0173] Input the target temperature field sequence data into the initialized spatiotemporal prediction model of the temperature field of the wooden bridge fire, and compare the annotation information with the target time output by the initialized spatiotemporal prediction model of the temperature field of the wooden bridge fire.

[0174] The spatiotemporal prediction model of the temperature field of a wooden covered bridge fire is trained and initialized based on the difference between the target time and the labeled information, thus obtaining the spatiotemporal prediction model of the temperature field of a wooden covered bridge fire.

[0175] In one optional implementation of this application embodiment, the processing unit 702 is specifically used for:

[0176] The target temperature field dataset is sampled to obtain discrete temperature field sequence data and corresponding annotation information.

[0177] Discrete temperature field sequence data is used as target temperature field sequence data.

[0178] In one optional implementation of this application embodiment, the spatiotemporal prediction model for the temperature field of a wooden covered bridge fire includes a long-term discrete prediction sub-model and an instantaneous continuous prediction sub-model. The processing unit 702 is specifically used for:

[0179] The target temperature field dataset is sampled to obtain discrete temperature field sequence data, the first annotation information corresponding to the discrete temperature field sequence data, continuous temperature field sequence data, and the second annotation information corresponding to the continuous temperature field sequence data.

[0180] In one optional implementation of this application embodiment, the processing unit 702 is specifically used for:

[0181] Input the discrete temperature field sequence data into the initialized long-term discrete prediction sub-model, and compare the first annotation information with the first target time output by the initialized long-term discrete prediction sub-model.

[0182] The continuous temperature field sequence data is input into the initialized instantaneous continuous prediction sub-model, and the second annotation information is compared with the second target time output by the initialized instantaneous continuous prediction sub-model.

[0183] In an optional implementation of this application embodiment, the processing unit 702 is further configured to:

[0184] Through fire simulation, the verification temperature field dataset of the ceiling of the wooden bridge fire model under different fire source locations is obtained. The verification temperature field dataset contains the verification temperature field sequence data of the ceiling of the wooden bridge fire model over a certain period of time, as well as the verification time corresponding to the verification temperature field sequence data.

[0185] The accuracy of the spatiotemporal prediction model for temperature field in wooden covered bridge fires was verified using a validation temperature field dataset.

[0186] In one optional implementation of this application embodiment, the early warning unit 703 is specifically used for:

[0187] Based on the predicted temperature field sequence data, the warning time when the temperature at any location on the ceiling of the wooden covered bridge scene exceeds the temperature threshold is determined.

[0188] Fire warning information for wooden covered bridges is generated based on the warning time.

[0189] This application also provides a computer program product including computer program instructions that, when executed by a processor, cause the processor to perform the steps of the method for predicting the temperature field of a wooden covered bridge fire according to various embodiments of this specification as described in the "Exemplary Methods" section above.

[0190] The computer program product can be written in any combination of one or more programming languages ​​to perform the operations of the embodiments of this specification. The programming languages ​​include object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages.

[0191] This application also provides a computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor of the steps in the method for predicting the temperature field of a wooden covered bridge fire according to various embodiments of this specification as described in the "Exemplary Methods" section above.

[0192] This application also provides an electronic device, including a memory and a processor. The memory stores a method for predicting the temperature field of a wooden covered bridge fire, and the processor is used to employ the above-described method for predicting the temperature field of a wooden covered bridge fire when executing the method.

[0193] Specifically, such as Figure 6As shown, the electronic device includes a processor 100, at least one communication bus 200, a user interface 300, at least one external communication interface 400, and a memory 500. The communication bus 200 is configured to enable communication between these components. The user interface 300 may include a display screen, and the external communication interface 400 may include standard wired and wireless interfaces. The memory 500 stores a method for predicting the temperature field of a wooden walkway fire. The processor 100 is used to employ the method when executing the method stored in the memory 500 for predicting the temperature field of a wooden walkway fire.

[0194] The descriptions of the above computer program products, computer-readable storage media, and electronic devices are similar to those of the above method embodiments, and have similar beneficial effects. For any technical details not disclosed in the computer program products, computer-readable storage media, and electronic devices of this application, please refer to the descriptions of the method embodiments of this application for understanding.

[0195] The sequence numbers or order of description of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0196] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0197] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0198] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0199] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer, or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., digital versatile disc (DVD)), or a semiconductor medium (e.g., solid-state disk (SSD)). It is worth noting that the computer-readable storage medium mentioned in the embodiments of this application can be a non-volatile storage medium; in other words, it can be a non-transient storage medium.

[0200] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, stored data, displayed data, etc.), and signals involved in the embodiments of this application are all authorized by the user or fully authorized by all parties, and the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. For example, the scene data of the current frame in the 3D virtual scene involved in the embodiments of this application, the client's device information, and the scene interaction information are all obtained with full authorization.

[0201] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for predicting the temperature field in a fire on a wooden covered bridge, characterized in that, include: Obtain the temperature field sequence data of the ceiling of the wooden covered bridge in the current moment; The temperature field sequence data is input into a pre-trained spatiotemporal prediction model for the temperature field of a fire on a wooden covered bridge, to obtain predicted temperature field sequence data representing a certain time after the current moment. Based on the predicted temperature field sequence data, fire early warning information for wooden covered bridges is generated; The spatiotemporal prediction model for the temperature field of the wooden covered bridge fire is trained using the following methods: Through fire simulation, the target temperature field dataset of the ceiling of the wooden corridor bridge fire model under different fire source locations is obtained. The target temperature field dataset contains the target temperature field sequence data of the ceiling of the wooden corridor bridge fire model within a certain time period. The time corresponding to the target temperature field sequence data is the annotation information of the target temperature field dataset. The target temperature field sequence data is input into the initialized spatiotemporal prediction model of the temperature field of the wooden covered bridge fire, and the annotation information is compared with the target time output by the initialized spatiotemporal prediction model of the temperature field of the wooden covered bridge fire. The initialized spatiotemporal prediction model of the temperature field of the wooden covered bridge fire is trained based on the difference between the target time and the labeled information, and the spatiotemporal prediction model of the temperature field of the wooden covered bridge fire is obtained. The spatiotemporal prediction model for the temperature field of a wooden covered bridge fire includes a long-term discrete prediction sub-model and an instantaneous continuous prediction sub-model. After obtaining the target temperature field dataset of the ceiling of the wooden covered bridge fire model at different fire source locations through fire simulation, the method includes: The target temperature field dataset is sampled to obtain discrete temperature field sequence data, first annotation information corresponding to the discrete temperature field sequence data, continuous temperature field sequence data, and second annotation information corresponding to the continuous temperature field sequence data. The step of inputting the target temperature field sequence data into the initialized spatiotemporal prediction model of the temperature field for a wooden covered bridge fire, and comparing the labeled information with the target time output by the initialized spatiotemporal prediction model of the temperature field for a wooden covered bridge fire, includes: The discrete temperature field sequence data is input into the initialized long-term discrete prediction sub-model, and the first annotation information is compared with the first target time output by the initialized long-term discrete prediction sub-model. The continuous temperature field sequence data is input into the initialized instantaneous continuous prediction sub-model, and the second annotation information is compared with the second target time output by the initialized instantaneous continuous prediction sub-model.

2. The method according to claim 1, characterized in that, The temperature field sequence data is input into a pre-trained spatiotemporal prediction model for the temperature field of a wooden covered bridge fire to obtain predicted temperature field sequence data representing a certain time period after the current moment, including: Based on the temperature field sequence data, determine the highest temperature distribution value on the ceiling of the wooden corridor bridge. When the highest distribution value is greater than the preset temperature threshold, the temperature field sequence data is input into the pre-trained spatiotemporal prediction model of the temperature field of the wooden covered bridge fire to obtain the predicted temperature field sequence data representing a certain time after the current moment.

3. The method according to claim 2, characterized in that, The spatiotemporal prediction model for the temperature field of a wooden covered bridge fire is a long-term discrete prediction model. Therefore, after obtaining the target temperature field dataset of the ceiling of the wooden covered bridge fire model at different fire source locations through fire simulation, the method includes: The target temperature field dataset is sampled to obtain discrete temperature field sequence data and the corresponding annotation information of the discrete temperature field sequence data; The discrete temperature field sequence data is used as the target temperature field sequence data.

4. The method according to claim 1, characterized in that, The method further includes: Through fire simulation, a verification temperature field dataset of the ceiling of a wooden bridge fire model under different fire source locations is obtained. The verification temperature field dataset includes the verification temperature field sequence data of the ceiling of the wooden bridge fire model over a certain period of time, as well as the verification time corresponding to the verification temperature field sequence data. The accuracy of the output of the spatiotemporal prediction model for the temperature field of the wooden covered bridge fire was verified using the aforementioned verification temperature field dataset.

5. The method according to claim 1, characterized in that, The generation of fire early warning information for wooden covered bridges based on the predicted temperature field sequence data includes: Based on the predicted temperature field sequence data, determine the warning time when the temperature at any location on the ceiling of the wooden covered bridge scene exceeds the temperature threshold; The fire warning information for the wooden covered bridge is generated based on the warning time.

6. A device for predicting the temperature field of a fire in a wooden covered bridge for implementing the method of any one of claims 1-5, characterized in that, include: The acquisition unit is used to acquire the temperature field sequence data of the ceiling of the wooden covered bridge in the current moment. The processing unit is used to input the temperature field sequence data into a pre-trained spatiotemporal prediction model of the temperature field of a wooden covered bridge fire, and obtain predicted temperature field sequence data representing a certain time after the current moment. The early warning unit is used to generate fire early warning information for the wooden covered bridge based on the predicted temperature field sequence data.

7. An electronic device, characterized in that, include: The system includes a memory and a processor, which are interconnected. The memory stores computer instructions, and the processor executes the computer instructions to perform the method for predicting the temperature field of a fire on a wooden covered bridge as described in any one of claims 1 to 5.

Images