CN-122024388-A - Unified probability propagation prediction method and system suitable for WUI fire disaster

CN122024388ACN 122024388 ACN122024388 ACN 122024388ACN-122024388-A

Abstract

The invention discloses a unified probability propagation prediction method and a system suitable for WUI fire, wherein the method comprises the steps of constructing a unified state space and discretizing, constructing an external driving model of building layer fire propagation, constructing a dynamic Bayesian network of a building layer by taking each building as a node, parameterizing a conditional probability table of the dynamic Bayesian network of the building layer based on a physical driving mechanism, establishing a mapping relation between an internal state of the building layer and an external observation category, realizing state correction and rolling prediction based on Bayesian update, namely carrying out Bayesian update on the system state by combining observation information after obtaining the external observation, obtaining posterior state distribution, taking the posterior distribution as an initial state of next time step prediction, forming a rolling assimilation closed loop, and finally outputting a multi-state probability risk field of building scale evolution along with time. The invention can realize the unified coupling of landscape, building and flying fire, obtain the state probability output of evolution along with time, and assimilate and correct deviation.

Inventors

WANG NAIYU
WANG CAN
ZHOU YANG
SHI KUAN
BAI YE

Assignees

浙江大学

Dates

Publication Date: 20260512
Application Date: 20260413

Claims (10)

1. The unified probability propagation prediction method suitable for the WUI fire is characterized by comprising the following steps of: constructing a unified state space and discretizing, namely constructing the unified state space comprising a building layer state, a landscape layer state and a flying fire layer state aiming at the WUI fire spreading process, and defining the discrete fire state of each building in the building layer; Building an external driving model for building fire spread of a building layer, wherein the external driving model comprises two parts of landscape layer fire scene evolution and flying fire transport deposition; building and parameterizing a dynamic Bayesian network of a building layer, namely building the dynamic Bayesian network of the building layer by taking each building as a node to represent the time sequence propagation relation of building states; Establishing an observation model, namely establishing a mapping relation between the internal state of the building layer and the external observation category; And (3) realizing state correction and rolling prediction based on Bayesian updating, namely carrying out Bayesian updating on the system state by combining the observation information after obtaining external observation to obtain posterior state distribution, taking the posterior distribution as an initial state of next time step prediction to form a rolling assimilation closed loop, and finally outputting a multi-state probability risk field of building scale evolution along with time.
2. The method of claim 1, wherein the discrete fire conditions of each of the buildings in the building layer include four of an unburned condition, a lit condition, a fully burned condition, and an burned-out condition for characterizing the building's stepwise evolution during the propagation of the fire.
3. The method of claim 1, wherein the landscape layer fire evolution is based on terrain, fuel, wind field and initial fire field data, and the time sequence spreading process of the wild fire outside the community on the landscape layer is calculated in a pushing mode, so that the combustion state, fire wire pushing information and relevant heat release characteristics of the landscape grid under each time step are obtained.
4. A method according to claim 1 or 3, characterized in that the flying fire transportation is deposited on the scene evolution result of the landscape layer, the units in active combustion state in the landscape layer and the building layer are identified as flying fire source items, and the migration and deposition processes of flying fire particles in the space are calculated at least by combining the wind speed, the wind direction, the transportation distance and the deposition probability, so that the flying fire exposure of each building in the current time step is obtained.
5. The method according to claim 1, wherein building a building layer dynamic bayesian network with each building as a node specifically comprises: And constructing a dynamic Bayesian network of a building layer by taking each building as a node, wherein a father node set of the state nodes of the current time step of each building comprises the state nodes of the previous time step of the building, the state nodes of the current time step of the neighborhood building and the flying fire exposure nodes of the current time step acting on the building.
6. The method according to claim 1, characterized in that the parameterizing of the conditional probability tables of the building layer dynamic bayesian network is based on a physical driving mechanism, in particular comprising: the physical quantity driving mode is adopted to parameterize a conditional probability table of the dynamic Bayesian network of the building layer, and the conditional probability table specifically comprises a flying fire ignition probability driven by flying fire exposure, a radiation ignition probability driven by neighborhood building thermal radiation and a state transition probability of internal development after the building is ignited.
7. The method of claim 6, wherein the step of providing the first layer comprises, The calculation of the flying fire ignition probability driven by the flying fire exposure comprises the steps of calculating the probability of transition from an unburned state to an ignition state by utilizing a preset ignition probability function based on the accumulated flying fire exposure born by the building in a single time step as the driving quantity; The calculation of the radiation ignition probability driven by the heat radiation of the neighborhood building specifically comprises the steps of calculating the probability of the neighborhood building causing the ignition of the target building based on the accumulated radiation energy or equivalent exposure of the neighborhood combustion building to the target building as the driving quantity by utilizing a preset ignition probability function; the state transition probabilities of the internal development after the ignition of the building include, in particular, the probability of the building developing from the ignition state to the fully combusted state and the probability of the building developing from the fully combusted state to the burnout state.
8. The method according to claim 1, wherein establishing a mapping relationship between the internal states of the building layer and the external observation categories, in particular comprises: Mapping discrete fire states of the building nodes into external observation categories through an observation operator of the observation model so as to realize the butt joint of the internal states of the building layers and external observation information under a unified interface, wherein the external observation categories comprise undamaged and damaged categories.
9. A method according to any one of claims 1-3 or 5-8, wherein after external observations are obtained, the system state is bayesian updated in combination with the observation information to obtain a posterior state distribution, the posterior state distribution is used as an initial state of next time step prediction to form a rolling assimilation closed loop, and finally a multi-state probability risk field of building scale evolution along with time is output, which specifically comprises: After external observation information of the current time step is obtained, based on an external driving model, a building layer dynamic Bayesian network and an observation model, and in combination with the external observation information of the current time step, the system state is subjected to Bayesian updating to obtain posterior state distribution reflecting the real fire condition of the current time step, and the updated posterior distribution is used as an initial state of fire propagation prediction of the next time step and is continuously advanced, so that a rolling assimilation closed loop of prediction-observation-correction-re-prediction is formed, a multi-state probability risk field of building scale evolution along with time is finally output, and dynamic assimilation prediction of the WUI fire propagation process is realized.
10. A unified probability propagation prediction system for WUI fires, comprising: the state space definition module is used for constructing a unified state space comprising a building layer state, a landscape layer state and a flying fire layer state aiming at the WUI fire spreading process and defining discrete fire states of all buildings in the building layer; The external driving construction module is used for constructing an external driving model for building layer fire spreading, and the external driving force comprises two parts of landscape layer fire scene evolution and flying fire transport and deposition; The dynamic Bayesian network construction module is used for constructing a dynamic Bayesian network of a building layer by taking each building as a node to represent the time sequence propagation relation of the building state; The observation model construction module is used for establishing a mapping relation between the internal state of the building layer and the external observation category; And the state updating module is used for carrying out Bayesian updating on the system state by combining the observation information after obtaining the external observation to obtain posterior state distribution, taking the posterior state distribution as an initial state predicted by the next time step to form a rolling assimilation closed loop, and finally outputting a multi-state probability risk field of building scale evolution along with time.

Description

Unified probability propagation prediction method and system suitable for WUI fire disaster Technical Field The invention relates to the technical field of fire risk early warning, in particular to a unified probability propagation prediction method and system suitable for WUI fires. Background The wild fire disaster has evolved from a traditional natural ecological event to a composite Urban disaster which causes huge damage to the built environment, particularly in suburb staggered areas (Wildland-Urban Interface, abbreviated as WUI), the fire is often expressed as a cascade process of 'wild fire-community fire-Urban big fire', namely, on one hand, a fire wire formed by vegetation burning is continuously propelled, and on the other hand, a community internal building group is subjected to non-local and multi-point concurrent ignition under the actions of heat radiation, flame contact and flying fire, so that the large-range structural loss and life and property risks are caused. The WUI fire disaster can also cause linkage influence on the operation of key infrastructures such as power supply, communication, traffic and medical treatment, and the disaster emergency and post-disaster recovery cost is obviously increased. Unlike some sudden disasters, wild fires often have observable and updatable features in the development process, namely, as information such as remote sensing, unmanned aerial vehicles, ground inspection, emergency reports and the like continuously enter, management departments often have opportunity windows for dynamically adjusting emergency response strategies on the time scale of a fire spread from hours to days. Therefore, fire risk prediction for WUI scenes not only needs to provide qualitative judgment of whether fire occurs, but also needs to continuously absorb new observation in the disaster, correct risk sequencing and spatial distribution in real time, and provide operable quantitative support for resource scheduling, important defense and evacuation decision. The WUI fire disaster is different from the traditional wild fire, the dominant loss occurs in the built environment, and the propagation mechanism is shown as a 'landscape-building-flying fire' multi-mechanism concurrent coupling (shown in figure 1), namely the landscape surface fire spreads along the combustible fuel, short-range radiation effect ignition exists between buildings, the flying fire is transported remotely under the action of a wind field and deposited in the downwind direction to form a new ignition source, and the three mutually feed back and rapidly change a risk field. Therefore, in the WUI fire emergency scenario, to implement decision-oriented disaster risk prediction, a prediction model capable of rapidly generating a future time-step fire influence range and exposure strength under the environmental driving of a given wind field, fuel, terrain and the like is generally required. In the prior art, the common technical route mainly comprises landscape propagation simulation based on a fire spreading mechanism and used for deducing the firing line propulsion and combustion intensity, and a building risk assessment model facing to the built environment and used for converting fire exposure into ignition or loss risk of a building layer. In order to characterize the remote multipoint ignition common in WUI scenes, some methods also introduce statistical or semi-physical models of flying fire transport and deposition for giving the ignition "pressure field" or probability of ignition for downwind areas. In general, by combining information such as landscape propagation, building ignition and flying fire ignition, time sequence risk output of building scale can be theoretically formed, and a quantitative basis is provided for emergency response such as important defense, resource scheduling and evacuation. However, the prior art still has a key gap in the prediction output level of the fire emergency decision oriented to the WUI. Firstly, the existing method is used for respectively modeling landscape spreading, inter-building ignition and flying fire ignition or splicing in a loose coupling mode, and lacks a framework capable of integrally and uniformly describing the multi-mechanism propagation process in the same state space, so that the influence among different mechanisms is difficult to be transmitted in a consistent manner in the model, and the output is also difficult to maintain the same semantic caliber. Secondly, many prediction results are more biased to live wire range, intensity field or static loss evaluation, and the building scale required by emergency action-oriented state probability (such as risk sequencing and hot evolution of a plurality of time steps in the future) and time evolution state probability are difficult to directly give, so that the supporting capability of the prediction results on resource scheduling, important defending and dynamic decision is limited. In the disaste