Ancient building fire intelligent monitoring and early warning system based on multi-source information fusion

Abstract

The invention relates to the technical field of safety monitoring, and discloses an ancient building fire intelligent monitoring and early warning system based on multi-source information fusion, and the system comprises a multi-source heterogeneous sensing and preprocessing module which is used for obtaining and preprocessing multi-source heterogeneous data of an ancient building; the pyrolysis feature identification and inversion module is used for identifying pyrolysis features by using a deep convolutional neural network and a thermodynamic evolution model; the dynamic risk assessment architecture construction module is used for constructing a fire risk dynamic assessment architecture and generating local abnormal probability distribution; the multi-source evidence fusion and probability calculation module is used for feature fusion and comprehensive fire probability calculation; and the graded early warning cooperative execution module is used for converting the probability calculation result into an early warning execution signal to realize intelligent fire monitoring and early warning. The system can break through the limitation that a traditional single monitoring means is lagged in response and high in false alarm rate under the complex environment of the ancient building, and achieves the active intelligent early warning and monitoring of the fire of the ancient building.

Description

Technical Field

[0001] This application relates to the field of safety monitoring technology, and in particular to an intelligent monitoring and early warning system for fires in ancient buildings based on multi-source information fusion. Background Technology

[0002] Ancient buildings, as important carriers of historical and cultural heritage, are mostly constructed of pure wood or brick and wood. After hundreds of years of natural weathering, the moisture content of the wooden components is extremely low, and the surfaces are often covered with flammable paints and decorations, resulting in an extremely high fire load density. Furthermore, the interiors of ancient buildings often suffer from complex aging electrical wiring problems and are exposed to multiple fire-prone factors such as incense offerings and visitor activities. Once a fire breaks out, it can easily spread rapidly, causing irreversible damage. However, the unique spatial structure of ancient buildings (such as tall halls and blind spots created by beams and columns) and their semi-open environment often make the early signs of a fire very weak and concealed.

[0003] Existing fire monitoring technologies for ancient buildings mostly rely on traditional point-based smoke or heat detectors. In the tall, open interior spaces of these buildings, smoke takes a considerable amount of time to rise and accumulate to the detector's threshold, resulting in significant alarm delays. Furthermore, the presence of incense dust, changing light and shadow, and natural airflow within these buildings year-round easily interferes with single photoelectric smoke detectors, leading to frequent false alarms or missed alarms. On the other hand, relying solely on surface temperature monitoring is insufficient to detect smoldering fires within wooden structures or overheating hazards in electrical wiring. In complex scenarios involving multiple coupled fire-causing factors, single-dimensional sensor data cannot form an effective chain of evidence. In practical applications, due to the lack of accurate identification of early-stage pyrolysis characteristics and a deep fusion mechanism for multi-source data, alarms are often only triggered after open flames have formed, missing the optimal window for fire suppression. Therefore, there is an urgent need to construct an intelligent monitoring and early warning system capable of comprehensive perception, deep fusion of multi-source information, and high reliability. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an intelligent monitoring and early warning system for ancient building fires based on multi-source information fusion, which can overcome the limitations of traditional single monitoring methods in the complex environment of ancient buildings, such as slow response and high false alarm rate, and realize proactive intelligent early warning and monitoring for ancient building fires.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In the first aspect, this application provides an intelligent monitoring and early warning system for ancient building fires based on multi-source information fusion, including: a multi-source heterogeneous sensing and preprocessing module, used to acquire heterogeneous sensor network data and environmental meteorological data of the ancient building area. The heterogeneous sensor network data includes visible light image data, infrared thermal imaging data, multi-gas concentration data and electrical circuit status data, and performs spatiotemporal alignment and multi-scale denoising preprocessing on the heterogeneous sensor network data.

[0008] The pyrolysis feature identification and inversion module is used to identify the early pyrolysis features of wooden components of ancient buildings in real time based on preprocessed data, using deep convolutional neural networks and thermodynamic evolution models. The early pyrolysis features include surface temperature rise rate, smoke texture features and characteristic gas production rate.

[0009] The dynamic risk assessment architecture construction module is used to build a dynamic assessment architecture for fire risk of ancient buildings that maps virtual and real. The architecture includes a data perception layer, a feature fusion layer and a decision analysis layer. The data perception layer maps the three-dimensional spatial structure of ancient buildings based on digital twin technology and generates local anomaly probability distributions based on early pyrolysis features.

[0010] The multi-source evidence fusion and probability calculation module is used to perform spatiotemporal correlation analysis and full confidence calculation on the local anomaly probability distribution at the feature fusion layer using a multi-source information fusion algorithm based on the improved Dempster-Schafer evidence theory, so as to obtain the comprehensive fire occurrence probability of each monitoring zone of the ancient building.

[0011] The graded early warning and collaborative execution module is used by the decision analysis layer to receive the comprehensive fire occurrence probability and convert it into graded early warning execution signals for the audible and visual alarm system, automatic fire monitor system, electrical cut-off device and emergency evacuation indication system, driving the coordinated action of each subsystem to achieve intelligent fire monitoring, early warning and control.

[0012] As a preferred embodiment of the intelligent monitoring and early warning system for ancient building fires based on multi-source information fusion according to the present invention, the multi-source heterogeneous sensing and preprocessing module includes:

[0013] The visible light acquisition unit is used to acquire texture changes in the beam and column structure inside ancient buildings and visible smoke flow fields through visible light image data.

[0014] The infrared thermal imaging acquisition unit is used to acquire the temperature field distribution on the surface of wooden components and the heat radiation intensity of concealed fire sources through infrared thermal imaging data.

[0015] The multi-gas monitoring unit is used to collect real-time concentration values ​​of carbon monoxide, carbon dioxide, and volatile organic compounds through multi-gas concentration data.

[0016] The electrical condition monitoring unit is used to collect residual current, fault arcs, and cable temperatures of aging electrical circuits by collecting electrical circuit condition data.

[0017] The spatiotemporal registration and fusion unit is used to fuse heterogeneous sensor network data using a heterogeneous data registration algorithm based on timestamp interpolation and spatial coordinate transformation. This eliminates differences in sampling frequency and spatial perspective deviation between different sensors, and establishes a unified measurement equation for multi-source data. The calculation formula is as follows:

[0018] ;

[0019] in, for Multi-source fused data vectors that have completed spatiotemporal alignment at any given moment. This represents the total number of sensor types. For the first The reliability weighting coefficients for these sensors are dynamically adjusted based on the sensor's aging level and the level of environmental interference. For the first The raw sampling data stream of the sensor-like device. For the first Transmission delay compensation value for sensor-like sensors This is a spatial coordinate mapping and resampling function used to map data of different resolutions onto a unified digital twin mesh for ancient buildings. This represents the residual noise term in the data alignment process, and it follows a Gaussian distribution with a mean of zero.

[0020] The wavelet denoising and enhancement unit is used to enhance the signal of the multi-source fused data vector using an adaptive wavelet threshold denoising algorithm. The calculation formula is as follows:

[0021] ;

[0022] in, These are the denoised wavelet coefficients. For the original signal at the 1st The first layer decomposition scale Wavelet coefficients, For symbolic functions, For the first The adaptive threshold at the layer scale is proportional to the noise variance estimate at that scale, and is dynamically updated based on the diurnal variation of the background noise of the ancient building environment.

[0023] As a preferred embodiment of the intelligent monitoring and early warning system for ancient building fires based on multi-source information fusion according to the present invention, the pyrolysis feature identification and inversion module includes:

[0024] The model building sub-unit is used to construct a thermodynamic evolution model that includes a wood pyrolysis kinetics sub-model, a flue gas diffusion fluid dynamics sub-model, and a visual image feature extraction sub-model.

[0025] The pyrolysis rate inversion unit is used to calculate the thermal anomaly gradient on the surface of wooden components based on infrared thermographic sequences, and to invert the combustion state during the smoldering stage by combining multi-gas concentration data. The pyrolysis rate characterization formula is as follows:

[0026] ;

[0027] in, The instantaneous pyrolysis reaction rate of the wooden components is given. The factor is the pre-existing factor, which depends on the tree species and degree of weathering of the timber used in ancient buildings. The activation energy for pyrolysis, Let be the ideal gas constant. Obtained from infrared thermal imaging data The absolute temperature of the wood surface at all times. The degree of reaction in wood transformation, with a value ranging from 0 to 1. The reaction order is... The characteristic gas correction function depends on the coupling relationship between the rate of change of carbon monoxide concentration and the concentration of smoke particles;

[0028] The smoke and fire feature recognition unit is used to address visual false alarms caused by changes in lighting and dust interference in the ancient architectural environment. It establishes a neural network model for smoke and fire recognition based on a multi-channel attention mechanism. The feature discrimination formula is as follows:

[0029] ;

[0030] in, Confidence score for visual characteristics of fire. It is the sigmoid activation function. For the first Attention weights for each feature channel, For convolutional feature extraction operations, The input is a visible light image frame. This is a feature vector concatenation operation. The fusion gain coefficient is the temperature change rate. The time derivative of the wood surface temperature is used to distinguish between real flames and non-heat-generating disturbances (such as fluttering fabric or insects).

[0031] As a preferred embodiment of the intelligent monitoring and early warning system for ancient building fires based on multi-source information fusion according to the present invention, the dynamic risk assessment architecture construction module includes:

[0032] The data interface unit is used to receive data on pyrolysis rate, fire visual feature confidence level, and electrical circuit status at the data perception layer.

[0033] The three-dimensional mesh mapping unit is used to create a full-element three-dimensional topological mesh containing beams, columns, brackets, rafters, and electrical pipelines based on the information model of ancient buildings.

[0034] The risk index calculation unit is used to calculate the static fire load density and dynamic fire spread risk index of each spatial grid node of the ancient building. The calculation formula is as follows:

[0035] ;

[0036] in, For the first The comprehensive fire risk index of each spatial grid node. This is the static load weighting coefficient. For the first in this grid Mass of combustible materials For the first The calorific value of combustible materials This represents the effective base area of ​​the grid node. For dynamic environment weighting coefficients, for The local wind speed at each grid point was obtained from environmental meteorological data through computational fluid dynamics simulation. The distance of this grid node to the nearest potential fire source is the Euclidean distance.

[0037] The probability distribution generation unit is used to spatially map the comprehensive fire risk index with early pyrolysis characteristics to generate local anomaly probability distributions corresponding to the global motion control target. The local anomaly probability distributions include open flame probability layers, smoldering probability layers, and electrical overheating probability layers.

[0038] As a preferred embodiment of the intelligent monitoring and early warning system for ancient building fires based on multi-source information fusion according to the present invention, the multi-source evidence fusion and probability calculation module includes:

[0039] The identification frame definition unit is used to define the identification frame, corresponding to open flame, smoldering, electrical fault, and normal states, respectively. The frame representation function is:

[0040] ;

[0041] in, To identify the frame, Represents an open flame state. This represents a smoldering state. Represents an electrical fault condition. This represents a normal state;

[0042] The basic probability allocation unit is used to construct the basic probability allocation function for each sensor source for the recognition framework, and the sensor reliability factor is introduced to modify the basic probability allocation function.

[0043] The evidence combination operation unit is used to perform orthogonal sum operations on multi-source evidence using the improved Dempster combination rule. The calculation formula is as follows:

[0044] ;

[0045] in, For the integrated proposition The basic probability assignment, i.e., the overall probability of fire occurrence. The total number of information sources participating in the integration. For the first Information source on the proposition The basic probability allocation value; The coefficient of evidence conflict characterizes the degree of contradiction between monitoring results from different sensors, and is calculated as follows: ;

[0046] The conflict correction update unit is used to introduce a time-dimensional evidence update mechanism. It uses the historical fusion results within the sliding time window to smooth the evidence conflict coefficient at the current moment. If the evidence conflict coefficient exceeds the preset trust threshold, the weighted average strategy is activated to replace the orthogonal sum operation in order to eliminate the interference of highly conflicting evidence on the fusion result.

[0047] The diffusion trend prediction unit is used to calculate the diffusion trend vector of a fire over time based on the comprehensive fire occurrence probability and the spatial connectivity of ancient buildings. The diffusion trend vector includes the predicted growth rate of the burned area and the settling rate of the smoke layer.

[0048] As a preferred embodiment of the intelligent monitoring and early warning system for ancient building fires based on multi-source information fusion according to the present invention, the hierarchical early warning and collaborative execution module includes:

[0049] The extremely high risk response unit is used for a Level 1 extremely high risk warning. When the confidence level of the open flame state in the overall fire occurrence probability exceeds the first preset threshold, it immediately generates an automatic aiming and spraying command for the automatic fire monitor system and simultaneously triggers a full-area audible and visual alarm.

[0050] The high-risk response unit is used to generate a specific circuit power-off command for the electrical cut-off device and generate precise location guidance information for inspection personnel when the confidence level of smoldering state or electrical fault state in the comprehensive fire occurrence probability exceeds the second preset threshold but is lower than the first preset threshold for a level 2 high-risk early warning.

[0051] The medium-risk response unit is used for a level 3 medium-risk early warning. When the characteristic gas production rate is detected to be increasing and the overall fire probability has not reached the second preset threshold, the sampling frequency of the heterogeneous sensor network data is adjusted to a high-frequency mode, and the PTZ camera in the nearby area is driven to focus and verify.

[0052] The IoT collaborative command unit is used to synchronously send automatic aiming and spraying commands, specific circuit power-off commands, and graded early warning execution signals to the corresponding underlying execution controllers via a low-power wide-area IoT bus dedicated to ancient buildings, and dynamically generate the optimal evacuation route to avoid the fire source on the emergency evacuation instruction system.

[0053] As a preferred embodiment of the intelligent monitoring and early warning system for ancient building fires based on multi-source information fusion of the present invention, the system further includes an adaptive background learning and false alarm suppression module, which includes:

[0054] The environmental parameter database establishment unit is used to establish a seasonal environmental parameter database. The seasonal environmental parameter database is integrated into the intelligent monitoring and early warning process of ancient building fires. Based on the geographical location of the ancient building and historical meteorological records, the seasonal environmental parameter database stores the light intensity benchmark, wood equilibrium moisture content benchmark and background temperature fluctuation range under different seasons.

[0055] The environmental baseline verification unit is used to automatically retrieve the environmental baseline parameters of the current season for secondary verification after the decision analysis layer issues an early warning signal. If the fluctuation range of the monitoring data is within the background temperature fluctuation range, it is determined to be environmental interference and the comprehensive fire occurrence probability is automatically reset to zero. At the same time, the false alarm feature is stored in the negative sample database for online incremental training of the neural network model.

[0056] As a preferred embodiment of the intelligent monitoring and early warning system for ancient building fires based on multi-source information fusion of the present invention, the system further includes an air-ground collaborative three-dimensional patrol module, which includes:

[0057] The mobile node deployment unit is used to deploy several micro inspection drones with autonomous navigation capabilities as mobile sensor nodes. The mobile sensor nodes are equipped with portable thermal imaging modules and high-sensitivity gas sniffers.

[0058] The blind zone supplementary measurement scheduling unit is used to calculate the optimal inspection path and drive the mobile sensor node to fly to the area for close-range multi-angle supplementary measurement when the local anomaly probability distribution generated by the data perception layer shows that there is a blind zone in a certain area or the data confidence is lower than the preset standard.

[0059] The fault takeover unit is used to take over the monitoring task of the area by moving sensor nodes if the fixed sensor nodes fail. It uses the data stream collected by the UAV to fill the missing parts of the heterogeneous sensor network data in real time and maintain the input integrity of the improved Dempster-Schaffer evidence theory algorithm.

[0060] The fire tracking unit is used to control mobile sensor nodes to track the fire in real time after the fire is confirmed to have occurred, and to project the real-time video stream of the core area of ​​the fire into the digital twin model to assist emergency command and decision-making.

[0061] Secondly, the present invention provides an intelligent monitoring and early warning method for fires in ancient buildings based on multi-source information fusion, including: acquiring heterogeneous sensor network data and environmental meteorological data of the ancient building area, wherein the heterogeneous sensor network data includes visible light image data, infrared thermal imaging data, multi-gas concentration data and electrical circuit status data, and performing spatiotemporal alignment and multi-scale denoising preprocessing on the heterogeneous sensor network data.

[0062] Based on the preprocessed data, the early pyrolysis characteristics of wooden components of ancient buildings are identified in real time using deep convolutional neural networks and thermodynamic evolution models. The early pyrolysis characteristics include surface temperature rise rate, smoke texture characteristics and characteristic gas production rate.

[0063] A dynamic assessment architecture for fire risk of ancient buildings based on virtual-real mapping is constructed. The architecture includes a data perception layer, a feature fusion layer, and a decision analysis layer. The data perception layer maps the three-dimensional spatial structure of ancient buildings based on digital twin technology and generates local anomaly probability distributions based on early pyrolysis features.

[0064] In the feature fusion layer, a multi-source information fusion algorithm based on the improved Dempster-Schafer evidence theory is used to perform spatiotemporal correlation analysis and full confidence calculation on the local anomaly probability distribution, so as to obtain the comprehensive fire occurrence probability of each monitoring zone of the ancient building.

[0065] The decision analysis layer receives the overall fire occurrence probability and converts it into graded early warning execution signals for the audible and visual alarm system, automatic fire monitor system, electrical cut-off device and emergency evacuation indication system, driving the coordinated action of each subsystem to achieve intelligent fire monitoring, early warning and control.

[0066] Thirdly, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements an intelligent monitoring and early warning system for ancient building fires based on multi-source information fusion.

[0067] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention can overcome the limitations of traditional single monitoring methods in complex environments of ancient buildings, such as slow response and high false alarm rate. By constructing a heterogeneous sensor network that includes visible light, infrared thermal imaging, multiple gases, and electrical states, and performing strict spatiotemporal alignment and multi-scale denoising preprocessing, it provides a comprehensive, high-fidelity multimodal data foundation for fire monitoring, effectively eliminating environmental noise and blind zone interference. By combining deep convolutional neural networks and thermodynamic evolution models, it achieves real-time identification of early pyrolysis characteristics of wooden components (such as temperature rise rate, smoke texture, and characteristic gases), and can detect early signs of fire in the early morning. The system detects danger signals during the smoldering stage before a fire breaks out, significantly improving the foresight of fire detection. By constructing a digital twin dynamic assessment architecture that maps virtual and real data and using a fusion algorithm based on improved evidence theory, discrete monitoring data is transformed into an intuitive three-dimensional spatial risk distribution. This intelligently resolves conflicts and uncertainties between multi-source data, yielding a high-confidence comprehensive fire probability. Finally, a tiered early warning mechanism drives precise coordinated actions of subsystems such as audible and visual alarms, automatic fire monitors, and electrical cut-off, achieving a leap from passive alarm to active defense and providing solid technical protection for the long-term preservation of ancient buildings and the safety of personnel. Attached Figure Description

[0068] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0069] Figure 1 This invention presents a schematic diagram of an intelligent monitoring and early warning system for ancient building fires based on multi-source information fusion.

[0070] Figure 2 This is a schematic diagram of an intelligent monitoring and early warning method for fires in ancient buildings based on multi-source information fusion proposed in this invention.

[0071] Figure 3 This is a schematic diagram of an electronic device proposed by the present invention. Detailed Implementation

[0072] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0073] Example 1, referring to Figure 1 The first embodiment of the present invention provides an intelligent monitoring and early warning system for fires in ancient buildings based on multi-source information fusion, comprising:

[0074] The multi-source heterogeneous sensing and preprocessing module is used to acquire heterogeneous sensor network data and environmental meteorological data of the ancient building area. The heterogeneous sensor network data includes visible light image data, infrared thermal imaging data, multi-gas concentration data and electrical circuit status data, and performs spatiotemporal alignment and multi-scale denoising preprocessing on the heterogeneous sensor network data.

[0075] The pyrolysis feature identification and inversion module is used to identify the early pyrolysis features of wooden components of ancient buildings in real time based on preprocessed data, using deep convolutional neural networks and thermodynamic evolution models. The early pyrolysis features include surface temperature rise rate, smoke texture features and characteristic gas production rate.

[0076] The dynamic risk assessment architecture construction module is used to build a dynamic assessment architecture for fire risk of ancient buildings that maps virtual and real. The architecture includes a data perception layer, a feature fusion layer and a decision analysis layer. The data perception layer maps the three-dimensional spatial structure of ancient buildings based on digital twin technology and generates local anomaly probability distributions based on early pyrolysis features.

[0077] The multi-source evidence fusion and probability calculation module is used to perform spatiotemporal correlation analysis and full confidence calculation on the local anomaly probability distribution at the feature fusion layer using a multi-source information fusion algorithm based on the improved Dempster-Schafer evidence theory, so as to obtain the comprehensive fire occurrence probability of each monitoring zone of the ancient building.

[0078] The graded early warning and collaborative execution module is used by the decision analysis layer to receive the comprehensive fire occurrence probability and convert it into graded early warning execution signals for the audible and visual alarm system, automatic fire monitor system, electrical cut-off device and emergency evacuation indication system, driving the coordinated action of each subsystem to achieve intelligent fire monitoring, early warning and control.

[0079] Specifically, firstly, this invention constructs a comprehensive intelligent protection system tailored to the unique environment of ancient buildings. At the system's bottom layer, a high-density heterogeneous sensor network is deployed, covering the entire space of the ancient building, from concealed electrical wiring to tall beams and columns. It utilizes visible light to capture smoke patterns, infrared sensors to detect heat accumulation, gas sensors to detect combustion products, and electrical monitoring devices to manage source risks. This not only solves the problem of single sensors being susceptible to interference from light, shadow, and dust, but also achieves advanced perception of the fire's incubation period (smoldering, overheating). At the data level, spatiotemporal registration and wavelet denoising techniques are used to clean multi-source data into a high-quality, unified spatiotemporal reference data stream, laying the foundation for subsequent analysis. The core processing layer introduces a dual-driven model combining deep learning and thermodynamic mechanisms. This utilizes the powerful feature extraction capabilities of neural networks to identify weak fire signs in complex scenarios, while also combining physical evolution models to verify the rationality of fire development, significantly reducing the false alarm rate. Furthermore, the system utilizes digital twin technology to map physical world risks onto a three-dimensional virtual space, and combines this with an improved DS evidence theory to handle uncertainties and conflicts between multi-source information, calculating a high-confidence fire probability. Ultimately, the execution layer intelligently dispatches various fire-fighting resources based on risk levels, from cutting off power to precisely spraying extinguishing agents, achieving a leap from "passive alarm" to "active intervention," ensuring the safety of the ancient building and its inhabitants throughout the entire chain.

[0080] The specific multi-source heterogeneous sensing and preprocessing module includes the following sub-units:

[0081] The visible light acquisition unit is used to acquire texture changes in the beam and column structure inside ancient buildings and visible smoke flow fields through visible light image data.

[0082] The infrared thermal imaging acquisition unit is used to acquire the temperature field distribution on the surface of wooden components and the heat radiation intensity of concealed fire sources through infrared thermal imaging data.

[0083] The multi-gas monitoring unit is used to collect real-time concentration values ​​of carbon monoxide, carbon dioxide, and volatile organic compounds through multi-gas concentration data.

[0084] The electrical condition monitoring unit is used to collect residual current, fault arcs, and cable temperatures of aging electrical circuits by collecting electrical circuit condition data.

[0085] The spatiotemporal registration and fusion unit is used to fuse heterogeneous sensor network data using a heterogeneous data registration algorithm based on timestamp interpolation and spatial coordinate transformation. This eliminates differences in sampling frequency and spatial perspective deviation between different sensors, and establishes a unified measurement equation for multi-source data. The calculation formula is as follows:

[0086] ;

[0087] in, for Multi-source fused data vectors that have completed spatiotemporal alignment at any given moment. This represents the total number of sensor types. For the first The reliability weighting coefficients for these sensors are dynamically adjusted based on the sensor's aging level and the level of environmental interference. For the first The raw sampling data stream of the sensor-like device. For the first Transmission delay compensation value for sensor-like sensors This is a spatial coordinate mapping and resampling function used to map data of different resolutions onto a unified digital twin mesh for ancient buildings. This represents the residual noise term in the data alignment process, and it follows a Gaussian distribution with a mean of zero.

[0088] For example, in specific calculations, the total number of sensors is set. , respectively representing visible light cameras Infrared thermal imager and composite gas detector The system at the current time The sampling rate of the visible light camera is Transmission delay The infrared thermal imager sampling rate is Transmission delay The gas detector sampling rate is Transmission delay The system evaluates the reliability weight of each sensor in the current environment based on historical operating data and sets... (Highly affected by sunlight) (Relatively stable thermal characteristics) (Gas diffusion has a hysteresis). The spatial mapping function projects visible light images (1920×1080 resolution) and infrared images (640×512 resolution) onto a unified grid coordinate system of the digital twin model of the ancient building. Gas data is directly mapped to the spatial properties of this grid. During calculation, the system extracts... Visible light frames at any given time Infrared frames at any moment and The gas readings at each time point are interpolated and aligned. If the normalized visible light pixel value is 0.8, the normalized infrared temperature is 0.6, and the normalized gas concentration is 0.2, then the residual noise term... Then, at that moment, the fused data vector value of the grid is approximately Through this precise millisecond-level spatiotemporal alignment, the system can ensure that subsequent algorithms process multidimensional states at the same physical moment and spatial location, avoiding logical misjudgments caused by data asynchrony;

[0089] The wavelet denoising and enhancement unit is used to enhance the signal of the multi-source fused data vector using an adaptive wavelet threshold denoising algorithm. The calculation formula is as follows:

[0090]

[0091] in, These are the denoised wavelet coefficients. For the original signal at the 1st The first layer decomposition scale Wavelet coefficients, For symbolic functions, For the first The adaptive threshold at the layer scale is proportional to the noise variance estimate at that scale, and is dynamically updated based on the diurnal variation of the background noise of the ancient building environment.

[0092] For example, in specific calculations, for the temperature time series signal output by the infrared thermal imaging sensor, the sym4 wavelet basis function in the approximately symmetric wavelet system is selected as the basis function. Layer decomposition. In the first layer... At the level decomposition scale, the original wavelet coefficients at a certain moment are calculated. The system dynamically estimates the noise variance at this scale based on the current nighttime environment (low background noise) of the ancient building, and sets an adaptive threshold, assuming the calculation results... .because Denoising wavelet coefficients If the noise figure at another location is -3.2, and its absolute value is less than the threshold of 5, it is set to 0. Through this adaptive processing, the low-frequency background noise generated by the infrared sensor due to ambient temperature fluctuations and the high-frequency random noise generated by electronic components are effectively filtered out, while retaining the weak but real temperature rise abrupt changes on the wood surface caused by potential smoldering. This improves the signal-to-noise ratio and provides a clean data source for subsequent calculations of minute temperature rise rates.

[0093] Specifically, the multi-source heterogeneous sensing and preprocessing module solves the problem of chaotic and inconsistent data sources in ancient building monitoring through integrated design. The visible light acquisition unit uses a high-resolution camera to capture the flow texture of smoke, overcoming the limitation of traditional smoke detectors that can only monitor a single point; the infrared thermal imaging acquisition unit can penetrate the surface decorative layer to detect abnormal heat sources inside the wooden structure; the multi-gas monitoring unit uses an electrochemical array to simultaneously detect carbon monoxide, carbon dioxide, and volatile organic compounds, enabling the identification of wood pyrolysis products before open flames occur; the electrical status monitoring unit focuses on monitoring electrical faults, a major cause of fires in ancient buildings, and analyzes current and fault arcs in real time. The spatiotemporal registration and fusion unit not only aligns sensor data of different frequencies on the time axis, but also maps two-dimensional image data and three-dimensional gas and electrical data onto a unified digital twin model coordinate system on the spatial axis, realizing the "three-dimensional" reconstruction of data. The wavelet denoising and enhancement unit dynamically adjusts the denoising strategy according to the difference in environmental noise between day and night in ancient buildings (e.g., more tourists during the day and quiet at night), preserving the edge features of real fire signals to the greatest extent and ensuring the signal quality input to subsequent models.

[0094] The specific pyrolysis feature identification and inversion module includes the following sub-units:

[0095] The model building sub-unit is used to construct a thermodynamic evolution model that includes a wood pyrolysis kinetics sub-model, a flue gas diffusion fluid dynamics sub-model, and a visual image feature extraction sub-model.

[0096] The pyrolysis rate inversion unit is used to calculate the thermal anomaly gradient on the surface of wooden components based on infrared thermographic sequences, and to invert the combustion state during the smoldering stage by combining multi-gas concentration data. The pyrolysis rate characterization formula is as follows:

[0097]

[0098] in, The instantaneous pyrolysis reaction rate of the wooden components is given. The factor is the pre-existing factor, which depends on the tree species and degree of weathering of the timber used in ancient buildings. The activation energy for pyrolysis, Let be the ideal gas constant. Obtained from infrared thermal imaging data The absolute temperature of the wood surface at all times. The degree of reaction in wood transformation, with a value ranging from 0 to 1. The reaction order is... The characteristic gas correction function depends on the coupling relationship between the rate of change of carbon monoxide concentration and the concentration of smoke particles;

[0099] For example, in specific calculations, the monitored objects are old camphor wood components commonly found in ancient buildings. Based on a material property library, a pre-exponential factor is set. pyrolysis activation energy ideal gas constant Reaction order Infrared thermal imager detected The absolute temperature of the wood surface rises to (about At this point, the wood is in a slight pyrolysis stage, and the degree of reaction is... The characteristic gas correction function is calculated based on sensor data. The current rate of change in carbon monoxide concentration is significant, with a calculated value of 1.2 (indicating that gas production contributes to the discrimination weight). Substituting these values ​​into the formula, the exponential term is calculated. Degree of Reaction The instantaneous pyrolysis reaction rate. This value quantifies the current rate of pyrolysis loss of the wood, and the system compares it with a preset safety threshold (such as...). The comparison revealed that the level had exceeded the standard, indicating a significant risk of smoldering in the area. Therefore, an early warning signal was issued in advance without any open flame.

[0100] The smoke and fire feature recognition unit is used to address visual false alarms caused by changes in lighting and dust interference in the ancient architectural environment. It establishes a neural network model for smoke and fire recognition based on a multi-channel attention mechanism. The feature discrimination formula is as follows:

[0101] ;

[0102] in, Confidence score for visual characteristics of fire. It is the sigmoid activation function. For the first Attention weights for each feature channel, For convolutional feature extraction operations, The input is a visible light image frame. This is a feature vector concatenation operation. The fusion gain coefficient is the temperature change rate. The time derivative of the wood surface temperature is used to distinguish between real flames and non-heat-generating disturbances (such as fluttering fabrics or insects).

[0103] For example, in a specific calculation, the system input... This represents a visible light image frame containing a region resembling a flame. A convolutional neural network extracts... Each key feature channel (texture, color, dynamic flicker) is weighted using an attention mechanism, with the texture channel weight set as follows: Color channel weights Dynamic channel weights After the convolution operation, the sum of the eigenvalues ​​is normalized to 2.5. Simultaneously, the infrared sensor detects that the surface temperature in this region changes from [value missing] to [value missing] within one second. Rise to Then the temperature-time derivative Fusion gain coefficient By combining visual features with temperature features: Finally, the sigmoid activation function is used to calculate... The score is close to 1, indicating a very high degree of certainty that the area is a real fire. If only visual features (such as red fabric fluttering) are present, and the temperature derivative is close to 0, then the input value is 2.5. If filtered using a threshold of 0.95, purely visual interference will be eliminated, and only visual targets with thermal effects will be identified as fires, thus effectively solving the problem of false alarms caused by red silk, lanterns, etc. in ancient buildings.

[0104] Specifically, the pyrolysis feature identification and inversion module utilizes a dual verification strategy of "mechanism + data". The model construction subunit mathematizes the physicochemical laws of wood combustion, making the monitoring process no longer a simple threshold comparison, but a real-time inversion of the combustion process. The pyrolysis rate inversion unit pays special attention to the pyrolysis stage before wood combustion. Through the coupling of infrared and gas data, it can quantify the degree of danger when the wood is only blackened and carbonized but has not yet ignited, which is something that traditional detectors cannot do. The smoke and fire recognition unit solves the problem of applying image recognition technology in the complex lighting environment of ancient buildings. By introducing the rate of temperature change as a "strong constraint", it forces visual targets to have thermal characteristics that conform to the laws of combustion, completely filtering out non-fire interference such as sunset illumination and the movement of red objects, greatly improving the robustness of recognition.

[0105] The specific dynamic risk assessment architecture building module includes the following sub-units:

[0106] The data interface unit is used to receive data on pyrolysis rate, fire visual feature confidence level, and electrical circuit status at the data perception layer.

[0107] The three-dimensional mesh mapping unit is used to create a full-element three-dimensional topological mesh containing beams, columns, brackets, rafters, and electrical pipelines based on the information model of ancient buildings.

[0108] The risk index calculation unit is used to calculate the static fire load density and dynamic fire spread risk index of each spatial grid node of the ancient building. The calculation formula is as follows:

[0109] ;

[0110] in, For the first The comprehensive fire risk index of each spatial grid node. This is the static load weighting coefficient. For the first in this grid Mass of combustible materials For the first The calorific value of combustible materials This represents the effective base area of ​​the grid node. For dynamic environment weighting coefficients, for The local wind speed at each grid point was obtained from environmental meteorological data through computational fluid dynamics simulation. The distance of this grid node to the nearest potential fire source is the Euclidean distance.

[0111] For example, in a specific calculation, a grid node near a pillar inside the main hall of an ancient building is selected, and the effective base area is... The grid contains the timber mass of the posts. (calorific value) ) and the quality of flammable curtains (calorific value) Static load weight The static part is... The dynamic component considers the spread of fire; the current moment is after the fire started. Dynamic environment weights Computational fluid dynamics simulations show the local wind speed at this location. And the distance of this grid from the nearest detected ignition point To simplify the integral calculation, assuming that wind speed and distance are constant over a short period of time, the dynamic term is: Ultimately, the comprehensive fire risk index of this grid node... The index is updated in real time. If the wind speed increases or the distance to the fire source shortens, the dynamic items will increase sharply, causing the risk index to soar. Based on this, the system judges the direction of fire spread and renders the grid as a dark red high-risk area on the digital twin interface.

[0112] The probability distribution generation unit is used to spatially map the comprehensive fire risk index with early pyrolysis characteristics to generate local anomaly probability distributions corresponding to the global motion control target. The local anomaly probability distributions include open flame probability layers, smoldering probability layers, and electrical overheating probability layers.

[0113] Specifically, the dynamic risk assessment architecture module serves as a bridge connecting physical perception and decision-making. The 3D mesh mapping unit utilizes Building Information Modeling (BIM) technology to deconstruct ancient buildings into minute spatial units, achieving component-level accuracy in risk positioning. The risk index calculation unit creatively combines "static load" with "dynamic environment." The static component reflects the basic combustion capacity of materials, while the dynamic component reflects the potential for spread under the influence of wind field and distance. This calculation method reflects the evolution trend of a fire more effectively than simple temperature monitoring. The probability distribution generation unit visualizes dry data into layered layers, allowing commanders to intuitively see where smoldering is occurring (smoldering probability layer highlighted) and where electrical circuits are overloaded (electrical overheating layer highlighted), thus achieving transparent control over the fire situation.

[0114] The specific multi-source evidence fusion and probability calculation module includes the following sub-units:

[0115] The identification frame definition unit is used to define the identification frame, corresponding to open flame, smoldering, electrical fault, and normal states, respectively. The frame representation function is:

[0116]

[0117] in, To identify the frame, Represents an open flame state. This represents a smoldering state. Represents an electrical fault condition. This represents a normal state;

[0118] The basic probability allocation unit is used to construct the basic probability allocation function for each sensor source for the recognition framework, and the sensor reliability factor is introduced to modify the basic probability allocation function.

[0119] The evidence combination operation unit is used to perform orthogonal sum operations on multi-source evidence using the improved Dempster combination rule. The calculation formula is as follows:

[0120] ;

[0121] in, For the integrated proposition The basic probability assignment, i.e., the overall probability of fire occurrence. The total number of information sources participating in the integration. For the first Information source on the proposition The basic probability allocation value; The coefficient of evidence conflict characterizes the degree of contradiction between monitoring results from different sensors, and is calculated as follows: ;

[0122] For example, in specific calculations, the system needs to fuse two sensors ( The sensor 1 (visible light smoke and fire detection) determines the state of a certain area and assigns a basic probability value to "open flame". Assigning a value to "normal" Sensor 2 (infrared thermal imaging) assigns a value to "open flame". Assigning a value to "normal" First, calculate the coefficient of conflict of evidence: that is... and Conflict, plus conflict. Normalization coefficient Next, calculate the probability of "open flame" after fusion. : Calculate the probability of "normal" after fusion. By fusing the data, the confidence level for detecting open flames increased from 0.7 or 0.6 with a single sensor to 0.778, significantly reducing uncertainty. If... Value too large (e.g.) This indicates a serious discrepancy between the sensors (e.g., one reporting a fire alarm while the other reports normal), and the system will initiate a subsequent conflict correction mechanism.

[0123] The conflict correction update unit is used to introduce a time-dimensional evidence update mechanism. It uses the historical fusion results within the sliding time window to smooth the evidence conflict coefficient at the current moment. If the evidence conflict coefficient exceeds the preset trust threshold, the weighted average strategy is activated to replace the orthogonal sum operation in order to eliminate the interference of highly conflicting evidence on the fusion result.

[0124] The diffusion trend prediction unit is used to calculate the diffusion trend vector of a fire over time based on the comprehensive fire occurrence probability and the spatial connectivity of ancient buildings. The diffusion trend vector includes the predicted growth rate of the burned area and the settling rate of the smoke layer.

[0125] Specifically, the multi-source evidence fusion and probability calculation module is the intelligent core of the system. In ancient building environments, single sensors often emit erroneous signals due to environmental interference (e.g., dust is mistaken for smoke). The identification framework definition unit clarifies the four core states that the system can distinguish, covering the entire life cycle of a fire. The evidence combination and operation unit utilizes the "orthogonal sum" property of DS evidence theory to aggregate multiple weak pieces of evidence (e.g., faint smoke, slight temperature rise) into a single strong piece of evidence, thereby achieving a high-confidence alarm at the very early stages of a fire. Simultaneously, the conflict correction and update unit resolves the paradox problem (i.e., the "veto" phenomenon) that may arise in DS theory when dealing with highly conflicting evidence. By introducing historical trust over time, if a sensor has consistently performed stably, the system tends to trust it during sudden conflicts, or a weighted average is used for a "compromise" when conflicts are extremely high. This ensures the robustness of the system under any extreme data conditions, preventing the entire early warning system from collapsing due to the failure of a single sensor.

[0126] The specific hierarchical early warning collaborative execution module includes the following sub-units:

[0127] The extremely high risk response unit is used for a Level 1 extremely high risk warning. When the confidence level of the open flame state in the overall fire occurrence probability exceeds the first preset threshold, it immediately generates an automatic aiming and spraying command for the automatic fire monitor system and simultaneously triggers a full-area audible and visual alarm.

[0128] The high-risk response unit is used to generate a specific circuit power-off command for the electrical cut-off device and generate precise location guidance information for inspection personnel when the confidence level of smoldering state or electrical fault state in the comprehensive fire occurrence probability exceeds the second preset threshold but is lower than the first preset threshold for a level 2 high-risk early warning.

[0129] The medium-risk response unit is used for a level 3 medium-risk early warning. When the characteristic gas production rate is detected to be increasing and the overall fire probability has not reached the second preset threshold, the sampling frequency of the heterogeneous sensor network data is adjusted to a high-frequency mode, and the PTZ camera in the nearby area is driven to focus and verify.

[0130] The IoT collaborative command unit is used to synchronously send automatic aiming and spraying commands, specific circuit power-off commands and graded early warning execution signals to the corresponding underlying execution controller through the low-power wide-area IoT bus dedicated to ancient buildings, and dynamically generate the optimal evacuation route to avoid the fire source on the emergency evacuation indication system.

[0131] Specifically, the tiered early warning and collaborative execution module achieves a closed loop from "perception" to "action." Unlike the one-size-fits-all approach of traditional fire alarm systems, this module finely divides risks into three levels based on the fusion-calculated probability values. For extremely high risks (such as the probability of open flame),... The system no longer waits for manual confirmation and directly takes over the fire monitors for physical fire suppression. This is crucial for all-wood structure ancient buildings, as every second of delay could lead to a chain reaction of fire; in high-risk situations (such as when smoldering is likely to occur...), The focus is on cutting off the source (power outage) and human intervention to avoid secondary damage to cultural relics from water stains; in medium-risk situations, the focus is on information enhancement, using gimbal cameras for "close-up" observation to confirm whether it is a false alarm. The IoT collaborative command unit ensures that these complex commands can be reliably transmitted to every end-execution device through a low-power wide area network within the complex physical space of the ancient building, while intelligently planning evacuation routes to guide personnel to avoid high-temperature and dense smoke areas, maximizing the protection of life safety;

[0132] In some embodiments, an intelligent monitoring and early warning system for ancient building fires based on multi-source information fusion may further include an adaptive background learning and false alarm suppression module, wherein the adaptive background learning and false alarm suppression module includes:

[0133] The environmental parameter database establishment unit is used to establish a seasonal environmental parameter database. The seasonal environmental parameter database is integrated into the intelligent monitoring and early warning process of ancient building fires. Based on the geographical location of the ancient building and historical meteorological records, the seasonal environmental parameter database stores the light intensity benchmark, wood equilibrium moisture content benchmark and background temperature fluctuation range under different seasons.

[0134] The environmental baseline verification unit is used to automatically retrieve the environmental baseline parameters of the current season for secondary verification after the decision analysis layer issues an early warning signal. If the fluctuation range of the monitoring data is within the background temperature fluctuation range, it is determined to be environmental interference and the comprehensive fire occurrence probability is automatically reset to zero. At the same time, the false alarm feature is stored in the negative sample database for online incremental training of the neural network model.

[0135] Specifically, the adaptive background learning and false alarm suppression modules endow the system with the ability to "self-evolve." The environmental characteristics of ancient buildings vary significantly with the seasons. For example, summer is hot and humid, the wood has a high moisture content, making it less flammable but prone to mold and heat generation; winter is dry and cold, the background temperature is low but static electricity is high. The environmental parameter database building unit pre-loads this prior knowledge. When the system detects that the surface temperature of a beam or column reaches a certain level at noon in summer... In such cases, a traditional system might trigger an alarm, but the environmental benchmark verification unit will search the database and find that the benchmark surface temperature of wood under direct sunlight at midday in the region can reach [temperature value missing]. Furthermore, since the fluctuations are within the normal range, it is determined to be "environmental interference" and the alarm is suppressed. This mechanism greatly reduces false alarms caused by environmental factors. At the same time, the system automatically records the data characteristics that lead to "quasi-alarms" and uses them as negative samples to input into the neural network for training. This makes the system more adaptable to the specific ancient building environment and more accurate in its identification as the running time increases.

[0136] In some embodiments, an intelligent monitoring and early warning system for ancient building fires based on multi-source information fusion may further include an air-ground collaborative three-dimensional patrol module, which includes:

[0137] The mobile node deployment unit is used to deploy several micro inspection drones with autonomous navigation capabilities as mobile sensor nodes. The mobile sensor nodes are equipped with portable thermal imaging modules and high-sensitivity gas sniffers.

[0138] The blind zone supplementary measurement scheduling unit is used to calculate the optimal inspection path and drive the mobile sensor node to fly to the area for close-range multi-angle supplementary measurement when the local anomaly probability distribution generated by the data perception layer shows that there is a blind zone in a certain area or the data confidence is lower than the preset standard.

[0139] The fault takeover unit is used to take over the monitoring task of the area by moving sensor nodes if the fixed sensor nodes fail. It uses the data stream collected by the UAV to fill the missing parts of the heterogeneous sensor network data in real time and maintain the input integrity of the improved Dempster-Schaffer evidence theory algorithm.

[0140] The fire tracking unit is used to control mobile sensor nodes to track the fire in real time after the fire is confirmed to have occurred, and to project the real-time video stream of the core area of ​​the fire into the digital twin model to assist emergency command and decision-making.

[0141] Specifically, the air-ground collaborative three-dimensional inspection module addresses the pain point that fixed sensors struggle to cover the complex structural blind spots of ancient buildings (such as deep within roof caissons and inside bracket sets). The miniature inspection drones normally remain hidden in their charging docks. Once the system detects abnormal data in a certain area but cannot confirm it (blind spot or low data confidence), the blind spot supplementary monitoring and dispatch unit immediately dispatches the drone to that point. Using onboard thermal imaging and gas sniffers, it conducts close-up reconnaissance, essentially adding a pair of flexible "eyes" and "nose" to the system. The fault takeover unit demonstrates the system's fault tolerance. When fixed lines fail due to aging or rodent damage causing sensor malfunction, the drone can temporarily act as an aerial sensor node, ensuring the data link remains unbroken. During the emergency command phase after a fire, the fire tracking unit controls the drone to avoid high-temperature airflows, providing an aerial view of the fire's spread and projecting real-time video onto the digital twin model of the command center. This provides commanders with a bird's-eye view, assisting in developing scientific firefighting plans and demonstrating the advantages of integrated air-ground collaborative operations.

[0142] Example 2, refer to Figure 2This is the second embodiment of the present invention, which provides a method for intelligent monitoring and early warning of ancient building fires based on multi-source information fusion, including: S1 acquiring heterogeneous sensor network data and environmental meteorological data of the ancient building area, wherein the heterogeneous sensor network data includes visible light image data, infrared thermal imaging data, multi-gas concentration data and electrical circuit status data, and performing spatiotemporal alignment and multi-scale denoising preprocessing on the heterogeneous sensor network data; S2 based on the preprocessed data, using a deep convolutional neural network and a thermodynamic evolution model, identifying the early pyrolysis characteristics of the wooden components of the ancient building in real time, wherein the early pyrolysis characteristics include surface temperature rise rate, smoke texture characteristics and characteristic gas production rate; S3 constructing a virtual-real mapping dynamic assessment architecture for ancient building fire risk. The architecture comprises a data perception layer, a feature fusion layer, and a decision analysis layer. The data perception layer maps the three-dimensional spatial structure of the ancient building based on digital twin technology and generates local anomaly probability distributions based on early pyrolysis features. In the feature fusion layer (S4), a multi-source information fusion algorithm based on an improved Dempster-Schafer evidence theory is used to perform spatiotemporal correlation analysis and full confidence calculation on the local anomaly probability distributions to obtain the comprehensive fire occurrence probability of each monitoring zone of the ancient building. The decision analysis layer (S5) receives the comprehensive fire occurrence probability and converts it into graded early warning execution signals for the audible and visual alarm system, automatic fire monitor system, electrical cut-off device, and emergency evacuation indication system, driving the coordinated action of each subsystem to achieve intelligent fire monitoring, early warning, and control.

[0143] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements an intelligent monitoring and early warning system for fires in ancient buildings based on multi-source information fusion.

[0144] like Figure 3 As shown, the electronic device may include a processor 610, a communication interface 620, a memory 630, and a communication bus 640. The processor 610, communication interface 620, and memory 630 communicate with each other via the communication bus 640. The processor 610 can call logical instructions from the memory 630 to execute an intelligent monitoring and early warning system for ancient building fires based on multi-source information fusion.

[0145] Furthermore, the logical instructions in the aforementioned memory 630 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0146] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute an intelligent monitoring and early warning system for ancient building fires based on multi-source information fusion.

[0147] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements an intelligent monitoring and early warning system for fires in ancient buildings based on multi-source information fusion.

[0148] The device embodiments described above are merely illustrative. 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 network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0149] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0150] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0151] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A smart monitoring and early warning system for fires in ancient buildings based on multi-source information fusion, characterized in that, include: The multi-source heterogeneous sensing and preprocessing module is used to acquire heterogeneous sensor network data and environmental meteorological data of the ancient building area. The heterogeneous sensor network data includes visible light image data, infrared thermal imaging data, multi-gas concentration data and electrical circuit status data. The module performs spatiotemporal alignment and multi-scale denoising preprocessing on the heterogeneous sensor network data. The pyrolysis feature identification and inversion module is used to identify the early pyrolysis features of ancient building wooden components in real time based on the preprocessed data using a deep convolutional neural network and a thermodynamic evolution model. The early pyrolysis features include surface temperature rise rate, smoke texture features and characteristic gas production rate. A dynamic risk assessment architecture construction module is used to construct a dynamic assessment architecture for fire risk of ancient buildings that maps virtual and real data. The architecture includes a data perception layer, a feature fusion layer and a decision analysis layer. The data perception layer maps the three-dimensional spatial structure of ancient buildings based on digital twin technology and generates a local anomaly probability distribution based on the early pyrolysis features. The multi-source evidence fusion and probability calculation module is used in the feature fusion layer to perform spatiotemporal correlation analysis and full confidence calculation on the local anomaly probability distribution using a multi-source information fusion algorithm based on the improved Dempster-Schafer evidence theory, so as to obtain the comprehensive fire occurrence probability of each monitoring zone of the ancient building. The graded early warning and collaborative execution module is used by the decision analysis layer to receive the comprehensive fire occurrence probability and convert the comprehensive fire occurrence probability into graded early warning execution signals for the audible and visual alarm system, automatic fire monitor system, electrical cut-off device and emergency evacuation indication system, so as to drive the coordinated action of each subsystem to realize intelligent fire monitoring, early warning and control.

2. The intelligent monitoring and early warning system for ancient building fires based on multi-source information fusion according to claim 1, characterized in that, The multi-source heterogeneous sensing and preprocessing module includes: The visible light acquisition unit is used to acquire the texture changes of the beam and column structure inside the ancient building and the visible smoke flow field through the visible light image data; An infrared thermal imaging acquisition unit is used to acquire the temperature field distribution and heat radiation intensity of concealed fire sources on the surface of wooden components through the infrared thermal imaging data. A multi-gas monitoring unit is used to collect real-time concentration values ​​of carbon monoxide, carbon dioxide, and volatile organic compounds through the multi-gas concentration data; An electrical condition monitoring unit is used to collect residual current, fault arc, and cable temperature of aging electrical circuits through the electrical circuit condition data. The spatiotemporal registration and fusion unit is used to fuse the heterogeneous sensor network data using a heterogeneous data registration algorithm based on timestamp interpolation and spatial coordinate transformation. This eliminates differences in sampling frequency and spatial perspective deviations between different sensors, and establishes a unified measurement equation for multi-source data. The calculation formula is as follows: ； in, for Multi-source fused data vectors that have completed spatiotemporal alignment at any given moment. This represents the total number of sensor types. For the first The reliability weighting coefficients for these sensors are dynamically adjusted based on the sensor's aging level and the level of environmental interference. For the first The raw sampling data stream of the sensor-like device. For the first Transmission delay compensation value for sensor-like sensors This is a spatial coordinate mapping and resampling function used to map data of different resolutions onto a unified digital twin mesh for ancient buildings. This represents the residual noise term in the data alignment process, and it follows a Gaussian distribution with a mean of zero. The wavelet denoising and enhancement unit is used to enhance the signal of the multi-source fused data vector using an adaptive wavelet threshold denoising algorithm. The calculation formula is as follows: ； in, These are the denoised wavelet coefficients. For the original signal at the 1st The first layer decomposition scale Wavelet coefficients, For symbolic functions, For the first An adaptive threshold for the layer scale is proportional to the noise variance estimate for that scale, and the adaptive threshold is dynamically updated based on the diurnal variation pattern of the background noise of the ancient building environment.

3. The intelligent monitoring and early warning system for ancient building fires based on multi-source information fusion according to claim 2, characterized in that, The pyrolysis feature identification and inversion module includes: The model building subunit is used to build the thermodynamic evolution model, which includes a wood pyrolysis kinetics submodel, a flue gas diffusion fluid dynamics submodel, and a visual image feature extraction submodel. The pyrolysis rate inversion unit is used to calculate the thermal anomaly gradient on the surface of the wooden component based on the infrared thermographic sequence, and to invert the combustion state of the smoldering stage by combining the multi-gas concentration data. The pyrolysis rate characterization formula is as follows: ； in, The instantaneous pyrolysis reaction rate of the wooden components is given. The factor is the pre-existing factor, which depends on the tree species and degree of weathering of the timber used in ancient buildings. The activation energy for pyrolysis, Let be the ideal gas constant. Obtained from infrared thermal imaging data The absolute temperature of the wood surface at all times. The degree of reaction in wood transformation is represented by a value between 0 and 1. The reaction order is... The characteristic gas correction function depends on the coupling relationship between the rate of change of carbon monoxide concentration and the concentration of smoke particles; The smoke and fire feature recognition unit is used to address visual false alarms caused by changes in lighting and dust interference in the ancient architectural environment. It establishes a neural network model for smoke and fire recognition based on a multi-channel attention mechanism. The feature discrimination formula is as follows: ； in, Confidence score for visual characteristics of fire. It is the sigmoid activation function. For the first Attention weights for each feature channel, For convolutional feature extraction operations, The input is a visible light image frame. This is a feature vector concatenation operation. The fusion gain coefficient is the temperature change rate. The time derivative of the wood surface temperature is used to distinguish between real flames and non-heat-generating disturbances (such as fluttering fabric or insects).

4. The intelligent monitoring and early warning system for ancient building fires based on multi-source information fusion according to claim 3, characterized in that, The dynamic risk assessment architecture construction module includes: The data interface unit is used to receive the pyrolysis rate, the confidence level of the fire visual features, and the electrical circuit status data in the data perception layer. The three-dimensional mesh mapping unit is used to create a full-element three-dimensional topological mesh containing beams, columns, brackets, rafters, and electrical pipelines based on the information model of ancient buildings. The risk index calculation unit is used to calculate the static fire load density and dynamic fire spread risk index of each spatial grid node of the ancient building. The calculation formula is as follows: ； in, For the first The comprehensive fire risk index of each spatial grid node. This is the static load weighting coefficient. For the first in this grid Mass of combustible materials For the first The calorific value of combustible materials This represents the effective base area of ​​the grid node. For dynamic environment weighting coefficients, for The local wind speed at each grid point was obtained from environmental meteorological data through computational fluid dynamics simulation. The distance of this grid node to the nearest potential fire source is the Euclidean distance. The probability distribution generation unit is used to spatially map the comprehensive fire risk index with the early pyrolysis characteristics to generate a local anomaly probability distribution corresponding to the global motion control target. The local anomaly probability distribution includes an open flame probability layer, a smoldering probability layer, and an electrical overheating probability layer.

5. The intelligent monitoring and early warning system for ancient building fires based on multi-source information fusion according to claim 4, characterized in that, The multi-source evidence fusion and probability calculation module includes: A frame definition unit is used to define the identification frames, corresponding to open flame, smoldering, electrical fault, and normal states, respectively. The frame representation function is: ； in, To identify the frame, This represents the open flame state. This represents the smoldering state. This represents the electrical fault state. This represents the normal state; The basic probability allocation unit is used to construct the basic probability allocation function for each sensor source for the recognition framework, and the sensor reliability factor is introduced to modify the basic probability allocation function. The evidence combination operation unit is used to perform orthogonal sum operations on multi-source evidence using the improved Dempster combination rule. The calculation formula is as follows: ； in, For the integrated proposition The basic probability assignment, i.e., the overall probability of fire occurrence. The total number of information sources participating in the integration. For the first Information source on the proposition The basic probability allocation value; The coefficient of evidence conflict characterizes the degree of contradiction between monitoring results from different sensors, and is calculated as follows: ; The conflict correction update unit is used to introduce a time-dimensional evidence update mechanism. It uses the historical fusion results within the sliding time window to smooth the evidence conflict coefficient at the current moment. If the evidence conflict coefficient exceeds a preset trust threshold, a weighted average strategy is activated to replace the orthogonal sum operation in order to eliminate the interference of highly conflicting evidence on the fusion result. The diffusion trend prediction unit is used to calculate the diffusion trend vector of the fire over time based on the comprehensive fire occurrence probability and the spatial connectivity of the ancient building. The diffusion trend vector includes the predicted growth rate of the burned area and the settling rate of the smoke layer.

6. The intelligent monitoring and early warning system for ancient building fires based on multi-source information fusion according to claim 5, characterized in that, The hierarchical early warning collaborative execution module includes: The extremely high risk response unit is used to generate an automatic aiming and spraying command for the automatic fire monitor system when the confidence level of the open flame state in the comprehensive fire occurrence probability exceeds the first preset threshold in response to the first extremely high risk warning, and simultaneously triggers the whole area audible and visual alarm. The high-risk response unit is used to generate a specific circuit power-off command for the electrical cut-off device and generate precise location guidance information for the inspection personnel when the confidence level of smoldering state or electrical fault state in the comprehensive fire occurrence probability exceeds the second preset threshold but is lower than the first preset threshold for a level 2 high-risk early warning. The medium-risk response unit is used to adjust the sampling frequency of the heterogeneous sensor network data to a high-frequency mode and drive the PTZ camera in the nearby area to perform focusing verification when the characteristic gas production rate shows an upward trend and the comprehensive fire occurrence probability does not reach the second preset threshold for a level 3 medium-risk early warning. The IoT collaborative command unit is used to synchronously send the automatic aiming and spraying command, the specific circuit power-off command, and the graded early warning execution signal to the corresponding underlying execution controller through a low-power wide-area IoT bus dedicated to ancient buildings, and dynamically generate the optimal evacuation path to avoid the fire source on the emergency evacuation indication system.

7. The intelligent monitoring and early warning system for ancient building fires based on multi-source information fusion according to claim 1, characterized in that, The system further includes an adaptive background learning and false alarm suppression module, which includes: An environmental parameter database establishment unit is used to establish a seasonal environmental parameter database. The seasonal environmental parameter database is integrated into the intelligent monitoring and early warning process for fires in ancient buildings. The seasonal environmental parameter database stores the light intensity benchmark, wood equilibrium moisture content benchmark, and background temperature fluctuation range for different seasons based on the geographical location of the ancient building and historical meteorological records. The environmental baseline verification unit is used to automatically retrieve the environmental baseline parameters of the current season for secondary verification after the decision analysis layer issues an early warning signal. If the fluctuation range of the monitoring data is within the background temperature fluctuation range, it is determined to be environmental interference and the comprehensive fire occurrence probability is automatically reset to zero. At the same time, the false alarm feature is stored in the negative sample database for online incremental training of the neural network model.

8. The intelligent monitoring and early warning system for ancient building fires based on multi-source information fusion according to claim 1, characterized in that, The system also includes an air-ground collaborative three-dimensional patrol module, which includes: A mobile node deployment unit is used to deploy several micro-inspection drones with autonomous navigation capabilities as mobile sensing nodes. The mobile sensing nodes are equipped with a portable thermal imaging module and a high-sensitivity gas sniffer. The blind spot supplementary measurement scheduling unit is used to calculate the optimal inspection path and drive the mobile sensor node to fly to the area to perform close-range multi-angle supplementary measurement when the local anomaly probability distribution generated by the data perception layer shows that there is a blind spot in a certain area or the data confidence is lower than the preset standard. The fault takeover unit is used to take over the monitoring task of the area through the mobile sensor node if the fixed sensor node fails. It uses the data stream collected by the UAV to fill the missing parts of the heterogeneous sensor network data in real time and maintain the input integrity of the improved Dempster-Schafer evidence theory algorithm. The fire tracking unit is used to control the mobile sensor node to track the fire in real time after the fire is confirmed to have occurred, and to project the real-time video stream of the core area of ​​the fire into the digital twin model to assist in emergency command and decision-making.

9. A method for intelligent monitoring and early warning of fires in ancient buildings based on multi-source information fusion, characterized in that, include: Heterogeneous sensor network data and environmental meteorological data of the ancient building area are acquired. The heterogeneous sensor network data includes visible light image data, infrared thermal imaging data, multi-gas concentration data and electrical circuit status data. Spatiotemporal alignment and multi-scale denoising preprocessing are performed on the heterogeneous sensor network data. Based on the preprocessed data, the early pyrolysis characteristics of wooden components of ancient buildings are identified in real time using a deep convolutional neural network and a thermodynamic evolution model. The early pyrolysis characteristics include surface temperature rise rate, smoke texture characteristics, and characteristic gas production rate. A dynamic assessment architecture for fire risk of ancient buildings based on virtual-real mapping is constructed. The architecture includes a data perception layer, a feature fusion layer and a decision analysis layer. The data perception layer maps the three-dimensional spatial structure of ancient buildings based on digital twin technology and generates a local anomaly probability distribution based on the early pyrolysis features. In the feature fusion layer, a multi-source information fusion algorithm based on the improved Dempster-Schafer evidence theory is used to perform spatiotemporal correlation analysis and full confidence calculation on the local anomaly probability distribution to obtain the comprehensive fire occurrence probability of each monitoring zone of the ancient building. The decision analysis layer receives the comprehensive fire occurrence probability and converts it into graded early warning execution signals for the audible and visual alarm system, automatic fire monitor system, electrical cut-off device and emergency evacuation indication system, driving the coordinated action of each subsystem to achieve intelligent fire monitoring, early warning and control.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the intelligent monitoring and early warning method for ancient building fires based on multi-source information fusion as described in claim 9.

Images