An enterprise park meteorological fire risk assessment method based on an adaptive random forest model

Abstract

The present invention relates to an enterprise park meteorological fire risk assessment method based on an adaptive random forest model, which includes establishing a risk assessment model and a visualization warning platform. The risk assessment model collects alarm data, meteorological data, enterprise data, and park division data to form a risk assessment data set, and then integrates the internal characteristics and external data of the park into the data set. Relying on real-time data stream technology, new data is input into the adaptive random forest model to instantly update and evaluate the fire risk index value of the park, and optimize the model decision tree to automatically adjust the risk assessment model to adapt to changes in the environment and operations, provide accurate warning information to the visualization warning platform, display the park risk level in real time, and issue risk alarms in a timely manner.

Description

Technical Field

[0001] The present invention belongs to the technical field of machine learning, and in particular relates to a method for assessing meteorological fire risk in an enterprise park based on an adaptive random forest model. Background Art

[0002] Existing enterprise park fire risk assessment uses two methods: rule-based risk assessment and risk assessment relying on traditional statistical models. These methods often cannot accurately predict and assess fire risks in complex environments, and are difficult to achieve real-time updates and dynamic management.

[0003] Rule-based campus fire risk assessment lacks real-time data processing capabilities and relies heavily on regularly updated data. It is unable to integrate the latest meteorological or campus operation data in real time, resulting in the assessment results possibly not reflecting the current risk status. In addition, the flexibility and adaptability of risk assessment are low. Because it is based on fixed rules, it is difficult to adapt to rapidly changing environmental conditions, such as climate change or special events within the campus. Therefore, the accuracy of risk prediction in a dynamic environment is limited.

[0004] However, traditional statistical models used for fire risk assessment are not effective when dealing with complex data with nonlinear relationships. They cannot accurately capture the interactions between various risk factors. Feature selection and model updating also require manual intervention. When new data arrives, the model cannot be automatically updated, which is inefficient.

[0005] Therefore, designing an enterprise park meteorological fire risk assessment method based on an adaptive random forest model that can automatically select features and adapt to learning new data in real time has become a technical problem that needs to be solved urgently. Summary of the invention

[0006] In order to solve the above problems, the present invention provides an enterprise park meteorological fire risk assessment method based on an adaptive random forest model.

[0007] The technical solution of the present invention is a method for assessing the meteorological fire risk of an enterprise park based on an adaptive random forest model, comprising establishing a risk assessment model and a visual early warning platform. The steps for establishing the risk assessment model include:

[0008] Data collection: collect and organize alarm data, meteorological data, enterprise data, and park division data to form a risk assessment data set;

[0009] Multi-source data integration, further integrating the park's internal characteristics and external data into data sets, using real-time data streaming technology, and transmitting to the central server through a high-speed data interface;

[0010] Data cleaning and preprocessing, first fill in missing data or delete the entire data, then introduce the standard deviation method and box plot method to identify and process outliers, and finally normalize the format of the data set, which includes date, time format and measurement unit;

[0011] Feature engineering analysis: analyze the features to conduct correlation analysis of data elements, reduce data dimensions through principal component analysis (PCA), and retain the main change trends in the data;

[0012] Build an adaptive random forest model, input a data set and extract data features, initialize the model for basic training, use the data in the data set to learn and build an initial set of decision trees; receive new data that has been cleaned and preprocessed in real time, adjust the decision tree parameters for incremental learning, use the new data as incremental input, and update the decision tree at the tree level and node level; set performance indicators for regular evaluation of the random forest model, feedback to adjust the model parameters, and guide future tree generation and node updates; finally, the model outputs the park fire risk index value;

[0013] The visual early warning platform realizes dynamic update by setting up a visual risk map and associating the output of the risk assessment model to display the risk level of the park in real time; when the risk level of the park reaches the preset warning line, the early warning signal is pushed to relevant personnel and the automation system to handle the fire risk.

[0014] As a further improvement of the present invention, the alarm data collects fire alarm records for the past 10 years, which include the specific date, time, type of fire and specific location of the fire, and classifies the fire types into electrical fires, chemical fires and general fires; the meteorological data collects historical disaster meteorological characteristics, including maximum temperature, minimum humidity, daily cumulative precipitation, cumulative precipitation days, seasons and weather conditions; the enterprise data collects enterprise information including business types and product characteristics; the park division data collects environment and facility configuration information of enterprise parks.

[0015] As a further improvement of the present invention, the internal characteristics include enterprise type, production process, hazardous goods storage, employee density and electrical layout information, and the enterprise park categories are divided according to the internal characteristics. The enterprise park categories include flammable and explosive goods processing enterprises, electrical equipment concentrated enterprises and high-risk chemical storage enterprises; the external data receives information from meteorological sensors, geographic information systems GIS and park systems in real time, and the external data includes temperature, humidity, wind speed and terrain data.

[0016] As a further improvement of the present invention, the analysis features perform correlation analysis on data elements according to meteorological characteristics, time distribution characteristics and spatial distribution characteristics; the meteorological characteristics include daily precipitation, relative humidity, daily maximum temperature, consecutive rainless days and maximum wind speed; the time distribution characteristics include summer electricity consumption, winter electricity consumption and power equipment load; the spatial distribution characteristics include site function, building structure and population density; the multivariate analysis method is used in combination with the alarm data to analyze the meteorological characteristics, and the correlation between different meteorological elements and the fire risk in the park is obtained. By superimposing the time distribution characteristics, the correlation between electricity consumption in different seasons and the fire risk in the park is obtained. The kernel density method is used in combination with the alarm data to analyze the spatial distribution characteristics, and the spatial distribution characteristics of historical fires in the park are screened to obtain the possibility, sensitivity, resistance and stress resistance of different park sites being affected.

[0017] As a further improvement of the present invention, the adjustment of decision tree parameters includes adjusting the number of decision trees, the maximum depth and the splitting condition. The incremental learning is based on the Learning without Forgetting method, maintaining the prediction ability of the original data, calculating the difference between the output of the new model after iteration and the output of the old model, and expressing the difference in terms of Euclidean distance to obtain the distillation loss. The calculation formula is: ,in, is the output of the old model, is the output of the new model; then the distillation loss is weighted and summed with the standard loss of the new learning task, and the model loss function is obtained as , where L1 is the loss of the new model, L2 is the standard loss of the new learning task, L3 is the distillation loss, and w is the weight used to balance the two loss amounts; finally, according to the gradient of the total loss function of the model relative to the weight value of the distillation loss, the new weight value of the distillation loss is obtained, and the calculation formula is ,in, is the old weight, is the new weight, is the gradient of the total loss with respect to the weights.

[0018] As a further improvement of the present invention, the visualized risk map is based on an integrated geographic information system GIS and a high-resolution map, and is associated with the output of a risk assessment model. The fire risk index value of each park is identified on the map, and the map area is divided according to districts and counties to obtain the district and county level risk level of each area, and different district and county level risk levels are represented by multiple colors.

[0019] As a further improvement of the present invention, the park risk level is calculated by statistically analyzing the park fire risk index value, assigning corresponding weights to each park according to the proportion of the index value, and then performing weighted cumulative average processing on the risk data of all parks in the district and county to obtain the district and county level risk level. The calculation formula is: ,in, Indicates the risk level at the district or county level. Indicates the park fire risk index value within the corresponding county level. Indicates the risk weight value of the corresponding park.

[0020] After adopting the above method, by obtaining real-time multi-source data streams containing meteorological, geographical, and park data, and integrating park data characteristics and meteorological data characteristics, the collected data is cleaned and normalized, including outlier processing and missing data filling, to ensure the consistency and reliability of data input, ensure the quality of data, and provide a solid data foundation for risk assessment; feature selection is performed through feature engineering technology, and principal component analysis PCA is used to reduce data dimensions and retain the main change trends in the data. It can automatically identify and optimize the data features that have the greatest influence on fire risk prediction, retain the most predictive features while excluding irrelevant features, and simplify the subsequent calculation process of the adaptive random forest model, which solves the shortcomings of existing technologies in real-time data processing, model adaptability, complex data processing capabilities and automated feature selection, and improves the overall efficiency of park fire risk assessment.

[0021] By using the adaptive random forest model and combining it with real-time data streaming technology, it is possible to collect and process multi-source data from meteorological sensors, geographic information systems (GIS) and campus operating systems in real time, and dynamically adjust the risk assessment model in real time. Compared with traditional fire risk assessment methods, it can provide real-time predictive risk assessment updates, respond more quickly to changes in environmental and operating conditions, and maintain the timeliness and accuracy of assessment results. By implementing incremental learning technology, the random forest model does not need to be trained from scratch when receiving new data, but can be updated and optimized based on the existing model. During optimization, the number, maximum depth and splitting conditions of decision trees are adjusted to adapt to the new data environment to ensure the accuracy and timeliness of the assessment. Through GIS technology, the risk assessment results are combined with the geographic location, the risk map is displayed in real time, and warnings are automatically triggered according to the risk level, which can provide managers with intuitive risk information, speed up the decision-making process and allocate emergency resources in a timely manner. Risk visualization and real-time warnings based on geographic location improve the response speed and accuracy of spatial decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 Shown is the enterprise park fire risk assessment flow chart of the present invention. DETAILED DESCRIPTION

[0023] like Figure 1 The present invention shows a method for assessing the meteorological fire risk in an enterprise park based on an adaptive random forest model, including establishing a risk assessment model and a visual early warning platform.

[0024] The steps of establishing the risk assessment model include:

[0025] Data collection: collect and organize alarm data, meteorological data, enterprise data, and park division data to form a risk assessment data set;

[0026] Multi-source data integration, further integrating the park's internal characteristics and external data into data sets, using real-time data streaming technology, and transmitting to the central server through a high-speed data interface;

[0027] Data cleaning and preprocessing, first fill in missing data or delete the entire data, then introduce the standard deviation method and box plot method to identify and process outliers, and finally normalize the format of the data set, which includes date, time format and measurement unit;

[0028] Feature engineering analysis: analyze the features to conduct correlation analysis of data elements, reduce data dimensions through principal component analysis (PCA), and retain the main change trends in the data;

[0029] Build an adaptive random forest model, input a data set and extract data features, initialize the model for basic training, use the data in the data set to learn and build an initial set of decision trees; receive new data that has been cleaned and preprocessed in real time, adjust the decision tree parameters for incremental learning, use the new data as incremental input, and update the decision tree at the tree level and node level; set performance indicators for regular evaluation of the random forest model, feedback to adjust the model parameters, and guide future tree generation and node updates; finally, the model outputs the park fire risk index value;

[0030] The visual early warning platform realizes dynamic update by setting up a visual risk map and associating the output of the risk assessment model to display the risk level of the park in real time; when the risk level of the park reaches the preset warning line, the early warning signal is pushed to relevant personnel and the automation system to handle the fire risk.

[0031] The alarm data collects fire alarm records for the past 10 years, which include the specific date, time, type of fire and specific location of the fire, and classifies the fire types into electrical fires, chemical fires and general fires; the meteorological data collects historical disaster meteorological characteristics, including maximum temperature, minimum humidity, daily cumulative precipitation, cumulative precipitation days, seasons and weather conditions; the enterprise data collects enterprise information including business types and product characteristics; the park division data collects the environment and facility configuration information of the enterprise park.

[0032] The internal characteristics include enterprise type, production process, hazardous goods storage, employee density and electrical layout information. Enterprise park categories are divided according to the internal characteristics. The enterprise park categories include flammable and explosive goods processing enterprises, electrical equipment concentrated enterprises and high-risk chemical storage enterprises; the external data receives information from meteorological sensors, geographic information systems GIS and park systems in real time. The external data includes temperature, humidity, wind speed and terrain data.

[0033] The analysis features conduct correlation analysis on data elements based on meteorological features, time distribution features and spatial distribution features; the meteorological features include daily precipitation, relative humidity, daily maximum temperature, consecutive rainless days and maximum wind speed; the time distribution features include summer electricity consumption, winter electricity consumption and power equipment load; the spatial distribution features include site function, building structure and population density; the multivariate analysis method is used in combination with the alarm data to analyze the meteorological features, and the correlation between different meteorological elements and the fire risk in the park is obtained. The correlation between electricity consumption in different seasons and the fire risk in the park is obtained by superimposing the time distribution features. The kernel density method is used in combination with the alarm data to analyze the spatial distribution features, and the spatial distribution features of historical fires in the park are screened to obtain the possibility, sensitivity, resistance and resilience of different park sites being affected.

[0034] The adjustment of decision tree parameters includes adjusting the number of decision trees, maximum depth and splitting conditions. The incremental learning is based on the Learning without Forgetting method, maintaining the prediction ability of the original data, calculating the difference between the output of the new model after iteration and the output of the old model, and expressing the difference in terms of Euclidean distance to obtain the distillation loss. The calculation formula is: ,in, is the output of the old model, is the output of the new model; then the distillation loss is weighted and summed with the standard loss of the new learning task, and the model loss function is obtained as , where L1 is the loss of the new model, L2 is the standard loss of the new learning task, L3 is the distillation loss, and w is the weight used to balance the two loss amounts; finally, according to the gradient of the total loss function of the model relative to the weight value of the distillation loss, the new weight value of the distillation loss is obtained, and the calculation formula is ,in, is the old weight, is the new weight, is the gradient of the total loss with respect to the weights.

[0035] The visualized risk map is based on an integrated geographic information system (GIS) and a high-resolution map, and is associated with the output of a risk assessment model. The fire risk index value of each park is identified on the map, and the map area is divided according to districts and counties to obtain the district and county risk level of each area, and different district and county risk levels are represented by multiple colors.

[0036] The park risk level is calculated by counting the park fire risk index value, assigning corresponding weights to each park according to the proportion of the index value, and then performing weighted cumulative average processing on the risk data of all parks in the district and county to obtain the district and county level risk level. The calculation formula is: ,in, Indicates the risk level at the district or county level. Indicates the park fire risk index value within the corresponding county level. Indicates the risk weight value of the corresponding park.

[0037] By acquiring real-time multi-source data streams containing meteorological, geographical, and park data, and integrating park data features with meteorological data features, the collected data is cleaned and normalized, including outlier processing and missing data filling, to ensure the consistency and reliability of data input, guarantee the quality of data, and provide a solid data foundation for risk assessment; feature selection is performed through feature engineering technology, and principal component analysis PCA is used to reduce data dimensions and retain the main change trends in the data. It can automatically identify and optimize the data features that have the greatest influence on fire risk prediction, retain the features with the most predictive value while excluding irrelevant features, and simplify the subsequent calculation process of the adaptive random forest model. It solves the shortcomings of existing technologies in real-time data processing, model adaptability, complex data processing capabilities, and automated feature selection, and improves the overall efficiency of park fire risk assessment.

[0038] By using the adaptive random forest model and combining it with real-time data streaming technology, it is possible to collect and process multi-source data from meteorological sensors, geographic information systems (GIS) and campus operating systems in real time, and dynamically adjust the risk assessment model in real time. Compared with traditional fire risk assessment methods, it can provide real-time predictive risk assessment updates, respond more quickly to changes in environmental and operating conditions, and maintain the timeliness and accuracy of assessment results. By implementing incremental learning technology, the random forest model does not need to be trained from scratch when receiving new data, but can be updated and optimized based on the existing model. During optimization, the number, maximum depth and splitting conditions of decision trees are adjusted to adapt to the new data environment to ensure the accuracy and timeliness of the assessment. Through GIS technology, the risk assessment results are combined with the geographic location, the risk map is displayed in real time, and warnings are automatically triggered according to the risk level, which can provide managers with intuitive risk information, speed up the decision-making process and allocate emergency resources in a timely manner. Risk visualization and real-time warnings based on geographic location improve the response speed and accuracy of spatial decisions.

Claims

1. A method for assessing meteorological fire risk in enterprise parks based on an adaptive random forest model, characterized by: It includes establishing a risk assessment model and a visual early warning platform. The steps of establishing the risk assessment model include: Data collection: collect and organize alarm data, meteorological data, enterprise data, and park division data to form a risk assessment data set; Multi-source data integration, further integrating the park's internal characteristics and external data into data sets, using real-time data streaming technology, and transmitting to the central server through a high-speed data interface; Data cleaning and preprocessing, first fill in missing data or delete the entire data, then introduce the standard deviation method and box plot method to identify and process outliers, and finally normalize the format of the data set, which includes date, time format and measurement unit; Feature engineering analysis: analyze the features to conduct correlation analysis of data elements, reduce data dimensions through principal component analysis (PCA), and retain the main change trends in the data; Build an adaptive random forest model, input a data set and extract data features, initialize the model for basic training, use the data in the data set to learn and build an initial set of decision trees; Receive new data that has been cleaned and preprocessed in real time, adjust decision tree parameters and perform incremental learning, use new data as incremental input, and perform tree-level and node-level updates on the decision tree; Set performance indicators for regular evaluation of the random forest model, provide feedback to adjust model parameters, and guide future tree generation and node updates; finally, the model outputs the park fire risk index value; The visual early warning platform realizes dynamic update by setting a visual risk map and associating the output of the risk assessment model to display the risk level of the park in real time; When the risk level in the park reaches the preset warning line, an early warning signal is sent to relevant personnel and automation systems to handle fire risks; The adjustment of decision tree parameters includes adjusting the number of decision trees, maximum depth and splitting conditions. The incremental learning is based on the Learning without Forgetting method, maintaining the prediction ability of the original data, calculating the difference between the output of the new model after iteration and the output of the old model, and expressing the difference in terms of Euclidean distance to obtain the distillation loss. The calculation formula is: ,in, is the output of the old model, is the output of the new model; then the distillation loss is weighted and summed with the standard loss of the new learning task, and the model loss function is obtained as , where L1 is the loss of the new model, L2 is the standard loss of the new learning task, L3 is the distillation loss, and w is the weight used to balance the two loss amounts; finally, according to the gradient of the total loss function of the model relative to the weight value of the distillation loss, the new weight value of the distillation loss is obtained, and the calculation formula is ,in, is the old weight, is the new weight, is the gradient of the total loss with respect to the weights.

2. According to claim 1, a method for assessing enterprise park meteorological fire risk based on an adaptive random forest model is characterized by: The alarm data collects fire alarm records for the past 10 years, which include the specific date, time, type of fire and specific location of the fire, and classifies the fire types into electrical fires, chemical fires and general fires; the meteorological data collects historical disaster meteorological characteristics, including maximum temperature, minimum humidity, daily cumulative precipitation, cumulative precipitation days, seasons and weather conditions; the enterprise data collects enterprise information including business types and product characteristics; the park division data collects the environment and facility configuration information of the enterprise park.

3. According to claim 1, a method for assessing enterprise park meteorological fire risk based on an adaptive random forest model is characterized by: The internal characteristics include enterprise type, production process, hazardous goods storage, employee density and electrical layout information. Enterprise park categories are divided according to the internal characteristics. The enterprise park categories include flammable and explosive goods processing enterprises, electrical equipment concentrated enterprises and high-risk chemical storage enterprises; the external data receives information from meteorological sensors, geographic information systems GIS and park systems in real time. The external data includes temperature, humidity, wind speed and terrain data.

4. According to claim 1, a method for assessing enterprise park meteorological fire risk based on an adaptive random forest model is characterized by: The analysis features conduct correlation analysis on data elements based on meteorological features, time distribution features and spatial distribution features; the meteorological features include daily precipitation, relative humidity, daily maximum temperature, consecutive rainless days and maximum wind speed; the time distribution features include summer electricity consumption, winter electricity consumption and power equipment load; the spatial distribution features include site function, building structure and population density; the multivariate analysis method is used in combination with the alarm data to analyze the meteorological features, and the correlation between different meteorological elements and the fire risk in the park is obtained. The correlation between electricity consumption in different seasons and the fire risk in the park is obtained by superimposing the time distribution features. The kernel density method is used in combination with the alarm data to analyze the spatial distribution features, and the spatial distribution features of historical fires in the park are screened to obtain the possibility, sensitivity, resistance and resilience of different park sites being affected.

5. According to claim 1, a method for assessing enterprise park meteorological fire risk based on an adaptive random forest model is characterized by: The visualized risk map is based on an integrated geographic information system (GIS) and a high-resolution map, and is associated with the output of a risk assessment model. The fire risk index value of each park is identified on the map, and the map area is divided according to districts and counties to obtain the district and county risk level of each area, and different district and county risk levels are represented by multiple colors.

6. According to claim 5, a method for assessing enterprise park meteorological fire risk based on an adaptive random forest model is characterized by: The park risk level is calculated by counting the park fire risk index value, assigning corresponding weights to each park according to the proportion of the index value, and then performing weighted cumulative average processing on the risk data of all parks in the district and county to obtain the district and county level risk level. The calculation formula is: ,in, Indicates the risk level at the district or county level. Indicates the park fire risk index value within the corresponding county level. Indicates the risk weight value of the corresponding park.

Images