Fire risk assessment and early warning system

Abstract

The invention discloses a fire risk assessment and early warning system, and relates to the technical field of machine learning. The fire risk assessment early warning system comprises the following steps: (1) dividing potential factors of a fire risk into physical factors and environmental factors, and collecting related data; (2) constructing a risk assessment data learning sample; (3) dividing fire risk grades; (4) determining a fire risk calculation and evaluation model; and (5) comparing the risk score obtained by the established model with the divided fire risk level to obtain a building risk level. According to the embodiment of the invention, the method can facilitate the detection and early warning of the fire condition, and improves the safety.

Description

Technical Field

[0001] The present invention relates to the technical field of machine learning, and in particular to a fire risk assessment and early warning system. Background Art

[0002] Fires pose a serious threat to life safety, property loss and environmental impact. Fires can cause casualties, especially in crowded places where fires spread quickly and escape is difficult, increasing the risk to victims. Once the structure of a building is destroyed by fire, it will not only cause huge economic losses, but also damage important facilities, affecting the normal operation of society. In addition, the harmful gases and smoke released by fires also pose potential hazards to the surrounding environment and ecosystem, affecting air quality and residents' health.

[0003] Fire prediction is a challenging task, and its complexity stems from the interaction of multiple factors. The occurrence of fire is affected by multiple factors such as environmental conditions, climate change, and human activities. For example, temperature, humidity, wind speed, and the type and distribution of combustibles may directly affect the occurrence and spread of fire. Secondly, the structure and material properties of buildings also have an important impact on the behavior of fire, and fire may show very different characteristics in different types of buildings.

[0004] Fire prediction also needs to consider the unpredictability of human behavior, such as confusion and delays during evacuation, which further increases the difficulty of prediction. Fire prediction not only involves complex physical and chemical processes, but also requires the integration of multi-faceted information and data. Therefore, the development of an accurate prediction model is a topic that urgently needs in-depth research. The present invention proposes a new evaluation and early warning system to alleviate the problems existing in the prior art. Summary of the invention

[0005] In order to evaluate and warn of fire situations, improve safety and rapidity of fire detection, the present invention provides a fire risk evaluation and warning system.

[0006] The present invention proposes a fire risk assessment and early warning system, in which potential factors of fire occurrence can be divided into physical factors and environmental factors. These factors will be captured through a sensor network and form part of the data sample of the early warning system.

[0007] The collected temperature, smoke concentration, CO2 concentration, and humidity sensor readings are used as feature inputs to form a population matrix.

[0008] Among the environmental factors, the material, layout and ventilation of the building will affect the speed and range of fire spread. High temperature and dry environment will increase the risk of fire. The system can connect to external meteorological data and add data such as ambient temperature, wind speed, and precipitation. In crowded places, the risk and loss of fire are greater. The system can combine monitoring data to analyze the density of people. Model the environmental factors as a feature vector.

[0009] All physical and environmental factors are integrated together to form a multi-dimensional feature matrix.

[0010] According to the size of the risk score, the fire risk level is divided into: Low risk (0-0.3); Medium risk (0.3-0.6); High risk (0.6-0.8); Very high risk (0.8-1).

[0011] The multivariate linear regression model is used as the fire risk assessment model. In order to improve the accuracy of the model, interaction terms and regularization are introduced to further optimize the model.

[0012] The calculated risk score is used to determine the fire risk level.

[0013] The present invention has the following main advantages: The multivariate linear regression model is suitable for fire warning because it can quickly evaluate the combined effect of various sensor data and provide a reasonable basis for decision making. After introducing interaction terms and regularization, the model can handle more complex situations while maintaining stability and accuracy.

[0014] By continuously collecting new data and updating the model online, the fire risk warning system can better adapt to environmental changes and achieve more accurate fire predictions. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the specific implementation of the present invention, the following will briefly introduce the drawings required for use in the specific implementation. The drawings described below are some implementations of the present application: Figure 1 It is a flow chart of a fire risk assessment and early warning system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solutions and advantages of this application more clear, the following Figure 1 It should be understood that the specific embodiments described below are only used to explain the present application and are not used to limit the present application.

[0017] In one embodiment, if Figure 1 As shown, a fire risk assessment and early warning system, the method comprises the following steps: S101, dividing the potential factors of fire risk into physical factors and environmental factors, and collecting relevant data; S102. Construct risk assessment data learning samples.

[0018] The physical factors include temperature, smoke concentration, CO2 concentration and humidity; the environmental factors include building structure, weather conditions and crowd density, as well as the unit's fire management, fire facility status, drill records and historical fire cases.

[0019] Specifically include: Physical factors: The temperature sensor reading at time t is T t ; The smoke concentration sensor reading at time t is S t ; The reading of the carbon dioxide concentration sensor at time t is C t ; The humidity sensor reading at time t is H t ; Environmental factors: Building characteristics are factors related to the building, such as building materials, building layout, fire safety measures, etc.; Based on the fire resistance grade of the material (such as A, B, and C in the national standard), it is mapped into a numerical value (such as A=1, B=2, C=3, the higher the value, the worse the fire resistance); Building layout: evaluate the complexity and accessibility of evacuation routes, quantified as channel density (number of channels / building area) and exit reach time (calculated by simulated evacuation model); Weather conditions are factors external to the environment, such as wind speed and temperature, that may affect the spread of fire; Wind speed: Take the measured value (unit: m / s). If there is no specific classification standard, it can be classified according to the risk level: Low risk (0–1 m / s): 1 Medium risk (1–3 m / s): 2 High risk (>3m / s): 3 Temperature: The higher the temperature, the higher the fire risk; you can directly input the temperature value (℃) or classify it by risk: Low risk (<15°C): 1 Medium risk (15–30°C): 2 High risk (>30℃): 3 Crowd density is an indicator that reflects the density of people in a building or place. The more people there are, the faster the spread of fire may be. Density: unit number of people (people) / unit area (square meters), for example, 10 people / ㎡ means dense crowds; Through risk classification mapping: Low risk (<1 person / ㎡): 1 Medium risk (1–3 people / ㎡): 2 High risk (>3 people / ㎡): 3 For the management data in fire risk assessment, the following quantitative divisions are made: Daily management inspection: Inspection frequency (unit: times / month): The higher the value, the more frequent the daily management; The number and type of hidden dangers: can be set according to the type of hidden dangers. For example, a minor hidden danger is scored 1 point, a serious hidden danger is scored 3 points, and the hidden danger points of each inspection can be accumulated; Personnel training: Number of training sessions (unit: times / year): points will be awarded based on a fixed number of training sessions; Training participation rate (unit: %): A qualified standard can be set, and points will be deducted if it is below the standard; Exam pass rate: the proportion of employees who pass the exam. You can set a threshold score and give a corresponding weight. Fire equipment maintenance: intact rate (unit: %): 80% intact rate is worth 10 points, and 1 point will be deducted if it decreases by 5%; Maintenance records: The number of maintenance visits will be counted as bonus points based on the record interval and inspection type (routine, comprehensive, etc.); Fire drill record: Frequency (unit: times / year): additional points will be awarded if the specified number of times is reached annually; Success rate: points will be added if certain standards are met, and points will be deducted if the success rate is low; Historical fire cases: Fire frequency: Count the number of fires that occurred in a certain period of time, quantify it as the fire incidence rate in the past certain period of time (such as 1 year or 5 years), and adjust the weight according to the region or building type; Fire type and severity: Quantify the severity of different fire types. For example, more severe fire events are assigned higher values ​​based on the destructiveness of fire. This allows the model to focus more on high-risk events. Fire losses: Statistics of economic losses caused by fire (such as property losses, casualties, etc.), calculate average losses based on historical data, or quantify them by region or building characteristics.

[0020] S103. Classify fire risk levels; According to the size of the risk score, the fire risk level is divided into: Low risk (0-0.3): The system assesses that the possibility of fire is low. No special fire prevention measures are required, but basic fire safety management still needs to be maintained. Medium risk (0.3-0.6): The possibility of fire has increased, but the risk has not yet reached the level that requires immediate action. Monitoring and prevention measures should be strengthened, and attention should be paid to possible fire hazards; High risk (0.6-0.8): The possibility of fire is high, and early warning measures and preliminary protective preparations are needed to prevent the fire from expanding or spreading; Very high risk (0.8-1): The possibility of fire is very high, or the potential fire hazard is very obvious. At this time, comprehensive emergency measures must be taken immediately to ensure the safety of personnel and quickly carry out fire prevention work.

[0021] S104. Establish a fire risk calculation and assessment model.

[0022] The multivariate linear regression model can provide an intuitive and easy-to-interpret mathematical framework for fire risk assessment. It assumes that fire risk is a linear combination of multiple variables and determines the contribution of each factor to fire risk by training the regression coefficients of the model.

[0023] It is assumed that the fire risk (Risk(t)) is determined by a set of physical influencing factors, including temperature, smoke density, carbon dioxide concentration, and humidity.

[0024] Linear interpolation is a simple and commonly used method to deal with missing values ​​or non-uniform sampling in sensor data. The following is a linear interpolation algorithm for temperature T t , smoke concentration S t 、Carbon dioxide concentration C t and humidity H t The sensor data is linearly interpolated: Given a known data point of a certain feature: time point t i and t i+1 ; Corresponding sensor reading y i and i+1 (Here y can be T, S, C, H) For any t i and t i+1 At the time point t between them, the linear interpolation formula is:

[0025] Linear interpolation processes data quickly and is suitable for processing large amounts of sensor data. It only requires basic mathematical operations and is suitable for real-time systems or scenarios with limited computing resources.

[0026] A linear regression algorithm is used to fit the data. The goal is to adjust the regression coefficients in the model by minimizing the error between the predicted fire risk value and the actual fire occurrence.

[0027] In order to improve the accuracy of the model, interaction terms and regularization are introduced to further optimize the model.

[0028] Interaction terms refer to the interaction between variables in the model. For example, temperature and humidity may have independent effects on fire risk, but their combination (interaction) may be more significant in predicting risk. Interaction terms allow the model to capture these complex nonlinear relationships and enhance prediction accuracy.

[0029] A single feature may have a small impact on fire risk, but the interaction between different variables may have a greater impact. For example, under high temperature conditions, a decrease in humidity may significantly increase the risk of fire. Such complex relationships can be effectively captured by interaction terms.

[0030] Y represents the sensor data after linear interpolation (including T, S, C, and H). Considering the interaction between various features and introducing the interaction term, the model can be expressed as: Y j Y k Represents the interaction term of different sensor data, α0,α i ,α jk ,β i is the regression coefficient, Z i Represents quantified environmental factor data.

[0031] In this way, the model can capture the mutual influence between different features and improve the accuracy of risk assessment. By introducing interaction terms, the model can handle more complex data patterns and avoid the linear regression model ignoring important nonlinear effects.

[0032] For example, complex interactions between temperature, humidity, CO2 concentration and building characteristics may play a decisive role in the occurrence of fire.

[0033] By capturing more nuanced variable relationships, interaction terms can help models make more accurate fire risk predictions, especially on nonlinear and multidimensional data sets.

[0034] For example, under conditions of high temperature and low humidity, a rapid rise in smoke concentration may indicate the rapid spread of a fire. This complex situation cannot be modeled by a simple linear relationship, but the interaction term can solve this problem.

[0035] In order to prevent the model from overfitting, Ridge regression is introduced.

[0036] Ridge limits the size of the regression coefficient by adding an L2 regularization term to prevent the model from overfitting the training data.

[0037] Regularization is a technique to prevent the model from overfitting. Overfitting means that the model performs well on the training data but performs poorly on new test data. By introducing regularization terms, the complexity of the model can be constrained to make it more robust.

[0038] Ridge regression imposes square constraints on coefficients to prevent excessive coefficient values ​​and reduce model complexity and overfitting.

[0039] The model formula is: where λ is the regularization strength hyperparameter, and p and q are the number of features.

[0040] The regularization term limits the size of the regression coefficients, preventing the model from overfitting to the details of the training data. This allows the model to generalize better to new data and better predict unseen fire risk data.

[0041] Ridge regression reduces the impact of large coefficients and prevents the model from being overly dependent on certain variables. Especially when sensor data is imperfect or noisy, regularization can make the model more robust to small fluctuations in the data and avoid prediction errors caused by abnormal fluctuations in a single variable.

[0042] Combining interaction terms with regularization enables the model to capture the relationship between complex variables while maintaining the simplicity and robustness of the model through regularization.

[0043] By introducing interaction terms, the model can identify the potential complex factors of fire occurrence and combine regularization to prevent overfitting, thereby improving the accuracy of predicting fire risks.

[0044] Regularization makes the model more robust when dealing with new or different sensor data, avoiding the degradation of model performance due to accidental factors (such as noise or outliers) in a specific dataset.

[0045] The fire risk score New-Risk(t) calculated by the multivariate linear regression model is converted into the range of [0,1] by the Sigmoid function.

[0046] The Sigmoid function is defined as: Where z = New-Risk(t) is the output of the linear regression model.

[0047] This method compresses the original risk value to the range of [0,1], and is close to 1 when the value is high and close to 0 when the value is low. The result is compared with the risk level standard set in advance, and the four levels of fire risk are compared according to the size of the score, so as to determine the current fire risk level and take corresponding measures.

[0048] This approach combines sensor data with potential fire risk factors to help managers make timely decisions to avoid fires or minimize their damage.

Claims

1. A fire risk assessment and early warning system, characterized in that: include: S101. Classify potential fire risk factors into physical factors and environmental factors, and collect relevant data; S102, construct risk assessment data learning samples; S103. Classify fire risk levels; S104. Establish a fire risk calculation and assessment model; S105. Compare the risk score obtained by the established model with the classified fire risk level to obtain the building risk level.

2. A fire risk assessment and early warning system according to claim 1, characterized in that: The physical factors are temperature, smoke concentration, CO2 concentration and humidity, and the corresponding sensor readings are collected simultaneously. The environmental factors are building structure, weather conditions and human traffic density.

3. A fire risk assessment and early warning system according to claim 1, characterized in that: The collected temperature, smoke concentration, CO2 concentration, and humidity sensor readings are used as feature inputs to form an overall matrix, the environmental factors are modeled as a feature vector, and all physical and environmental factors are fused together to form a multi-dimensional feature matrix.

4. A fire risk assessment and early warning system according to claim 1, characterized in that: According to the size of the risk score, the fire risk level is divided into: Low risk (0-0.3); Medium risk (0.3-0.6); High risk (0.6-0.8); Very high risk (0.8~1).

5. A fire risk assessment and early warning system according to claim 1, characterized in that: The risk score obtained through the fire risk assessment model is compared with the classified fire risk level.

Images