Forest fire loss prediction method and device based on multi-modal data analysis

Abstract

The invention discloses a forest fire loss prediction method and device based on multi-modal data analysis, and relates to the technical field of fire prediction, and the method comprises the steps: obtaining historical fire data, and calibrating the historical fire grade; historical auxiliary data are obtained, period correction is conducted on the historical fire data, standard state historical fire data are generated, and the auxiliary data comprise historical element values, technical levels and expansion information. And based on the historical fire data and the fire grade, constructing a fire grade prediction model. And constructing and training a loss prediction model by using the historical fire grade as an input and the standard state historical fire data as an output. And combining the loss prediction model and the fire hazard grade prediction model, performing forest fire hazard loss prediction by using multi-mode monitoring data, and outputting a real-time fire hazard grade and prediction loss. Therefore, the technical effects of fusing the multi-modal data and improving the prediction timeliness and accuracy are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of fire prediction, and in particular to a method and device for predicting forest fire losses based on multimodal data analysis. Background Art

[0002] Forest resources are important natural resources, and the occurrence of forest fires has an impact on the ecological environment, economic property and human life. Analyzing the loss standards caused by forest fires is of great significance for forest fire prevention and reducing fire economic losses and casualties. Existing fire loss prediction methods mainly rely on single modal data, such as historical fire data and meteorological data, which makes it difficult to comprehensively and accurately predict fire losses. There are technical problems such as lack of dynamics in prediction and low prediction accuracy. Summary of the invention

[0003] The purpose of this application is to provide a forest fire loss prediction method and device based on multimodal data analysis, so as to solve the technical problems of lack of dynamics and low prediction accuracy in the prior art.

[0004] In view of the above technical problems, the present application provides a forest fire loss prediction method and device based on multimodal data analysis.

[0005] In a first aspect, the present application provides a forest fire loss prediction method based on multimodal data analysis, wherein the method comprises:

[0006] Historical fire data is acquired, and historical fire levels are calibrated based on the historical fire data.

[0007] Obtain historical auxiliary data, perform period correction on the historical fire data, and obtain standard historical fire data, wherein the historical auxiliary data includes historical element value information, historical technology level information, and historical inflation information.

[0008] Based on the historical fire data and the historical fire levels, a fire level prediction model is constructed.

[0009] The loss prediction model is constructed and trained by taking the historical fire levels as sample input and the loss data in the standard historical fire data as sample output.

[0010] The loss prediction model is combined with the fire grade prediction model to predict forest fire losses based on multimodal monitoring data, and the real-time forest fire grade and predicted forest fire losses are output.

[0011] In a second aspect, the present application also provides a forest fire loss prediction device based on multimodal data analysis, wherein the device comprises:

[0012] A data acquisition module is used to acquire historical fire data and calibrate historical fire levels based on the historical fire data.

[0013] An associated acquisition module is used to obtain historical auxiliary data, perform period correction on the historical fire data, and obtain standard historical fire data, wherein the historical auxiliary data includes historical element value information, historical technical level information, and historical inflation information.

[0014] The grade prediction model module is used to construct a fire grade prediction model based on the historical fire data and the historical fire grades.

[0015] A loss prediction model module is used to construct and train a loss prediction model using the historical fire level as sample input and the loss data in the standard historical fire data as sample output.

[0016] A multimodal prediction module is used to combine the loss prediction model with the fire grade prediction model, predict forest fire losses based on multimodal monitoring data, and output real-time forest fire grades and predicted forest fire losses.

[0017] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0018] By obtaining historical fire data and calibrating historical fire levels based on historical fire data; obtaining historical auxiliary data, correcting the historical fire data for a period, and obtaining standard historical fire data, where historical auxiliary data includes historical element value information, historical technical level information, and historical inflation information; building a fire level prediction model based on historical fire data and historical fire levels; using historical fire levels as sample inputs and loss data in standard historical fire data as sample outputs, building and training a loss prediction model; combining the loss prediction model with the fire level prediction model, predicting forest fire losses based on multimodal monitoring data, and outputting real-time forest fire levels and forest fire predicted losses. This achieves the technical effect of integrating multimodal data and improving prediction timeliness and accuracy.

[0019] The above description is only an overview of the technical solution of the present application. In order to more clearly illustrate the technical means of the present application and to implement it in accordance with the contents of the specification, and to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The embodiments of the present invention and the following brief description are illustrated in conjunction with the accompanying drawings, which are described as follows:

[0021] Figure 1 This is a flowchart of a forest fire loss prediction method based on multimodal data analysis in this application;

[0022] Figure 2 This is a schematic diagram of the structure of a forest fire loss prediction device based on multimodal data analysis in this application.

[0023] Explanation of the accompanying drawings: data collection module 11, association collection module 12, level prediction model module 13, loss prediction model module 14, multimodal prediction module 15. DETAILED DESCRIPTION

[0024] The present application solves the technical problems faced by the prior art of lack of dynamics and low prediction accuracy by providing a forest fire loss prediction method and device based on multimodal data analysis.

[0025] The solution in the embodiment of the present technology, in order to solve the above problems, adopts the following overall ideas:

[0026] First, historical fire data is obtained, and the historical fire level is calibrated based on the historical fire data. Then, historical auxiliary data is obtained, the historical fire data is period-corrected, and standard historical fire data is obtained, wherein the historical auxiliary data includes historical element value information, historical technical level information, and historical inflation information. Then, a fire level prediction model is constructed based on the historical fire data and the historical fire level. Next, a loss prediction model is constructed and trained with the historical fire level as sample input and the standard historical fire data as sample output. Furthermore, the loss prediction model is combined with the fire level prediction model to predict forest fire losses based on multimodal monitoring data. Finally, the real-time forest fire level and the predicted forest fire loss are output.

[0027] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the drawings and specific implementation methods of the specification. It should be noted that the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments of the present application. It should be understood that the present application is not limited to the example embodiments described herein. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. It should also be noted that for the sake of ease of description, only the parts related to the present invention are shown in the drawings, rather than all of them.

[0028] Embodiment 1

[0029] like Figure 1 As shown, the present application provides a forest fire loss prediction method based on multimodal data analysis, the method comprising:

[0030] Historical fire data is acquired, and historical fire levels are calibrated based on the historical fire data.

[0031] Specifically, historical fire data refers to fire data corresponding to forest fires that have occurred in the target scene, and historical fire levels refer to fire intensities or fire severity levels corresponding to the times of multiple forest fires in the historical fire data. Exemplarily, the historical fire level includes a quantified fire level coefficient or a predefined fixed fire level.

[0032] In some embodiments, obtaining historical fire data, and calibrating historical fire levels based on the historical fire data, includes:

[0033] An interactive forest fire data source is used to extract the historical fire data based on a preset collection window, wherein the historical fire data includes the time of occurrence, location of occurrence, scope of impact, scale of loss, and historical monitoring data; a forest fire evaluation strategy series is obtained, and based on the historical fire data, the historical fire level is determined and calibrated.

[0034] Specifically, data sources include fire departments, insurance companies, news media, academic research institutions, etc. Historical fire data include the time of fire, location (geographic location of fire, such as longitude and latitude), impact area (area or range affected by fire), loss (economic loss caused by fire), casualties, historical monitoring data (temperature, humidity, wind speed, etc. when fire occurred), etc.

[0035] Specifically, the forest fire assessment strategy series is a collection of fire assessment strategies in multiple historical periods, and the forest fire assessment strategy series corresponds to the historical periods of forest fire events in historical fire data. Among them, forest fire assessment strategies in different historical periods change with the passage of time and the development of social environment. Exemplarily, fire assessment standards and strategies in different periods are obtained by consulting historical documents, research reports and announcement archives.

[0036] Specifically, the data structure of the forest fire assessment strategy series includes: the specific time or period for implementation of the strategy, the specific standards and indicators for fire grade classification, the geographical scope and fire type to which the strategy applies, etc.

[0037] Specifically, based on the historical fire data and evaluation strategies obtained, the fire levels of multiple forest fire events are re-identified and calibrated. First, the historical fire data is matched with the forest fire evaluation strategies of the corresponding period to ensure that the correct evaluation criteria are used for each fire event. Then, the historical fire data is analyzed and identified using the specific criteria in the matched forest fire evaluation strategy. This is because due to the inconsistency of data updates and the timeliness of the evaluation strategy, the levels of forest fire event records in the historical fire data may be biased. If the rough statistics of losses in the early stages of a forest fire event are incomplete, it may cause deviations in the originally defined fire levels. Re-identifying and calibrating historical fire levels through evaluation strategies of different historical periods will help improve the accuracy of fire level classification.

[0038] Obtain historical auxiliary data, perform period correction on the historical fire data, and obtain standard historical fire data, wherein the historical auxiliary data includes historical element value information, historical technology level information, and historical inflation information.

[0039] Specifically, historical auxiliary data, including historical factor value information, historical technology level information and historical inflation information, are used to make period corrections to historical fire data, and standard historical fire data are obtained. The loss data between different historical periods are equated to the same value scale, which can then be compared, making the fire data comparable.

[0040] In some embodiments, obtaining historical auxiliary data and performing period correction on the historical fire data includes:

[0041] According to the occurrence time of the historical fire data, the historical element value information, the historical technology level information, and the historical inflation information are obtained; based on the historical inflation information, the historical fire data is traversed to perform inflation correction on the loss scale to obtain inflation-corrected data; based on the loss scale in the inflation-corrected data, the historical element value information is called to perform element value correction to obtain element value correction data; based on the loss scale in the element value correction data, the control level is corrected in combination with the historical technology level information to generate the standard historical fire data.

[0042] Specifically, the historical element value information includes the value of land, buildings, forest resources, etc. in different historical periods. The fire loss data is adjusted through the historical element value information, so that the loss data at different time points are comparable. Exemplarily, the historical element value information includes the value data of land, buildings and forest resources in different historical periods obtained through statistical platforms, forestry departments, relevant documents and archives, historical data of resource markets, etc.

[0043] Specifically, the historical technical level information includes the development of fire prevention, monitoring and firefighting technologies in different historical periods. The historical technical level information reflects the changes in fire response capabilities in different periods. For example, the historical technical level information is reflected in forest fire response time, remote sensing monitoring capabilities, improvement and updating of tools and equipment, personnel training and improvement of emergency response capabilities, etc. Optionally, the historical technical level information also includes the fire fighting efficiency value obtained after comprehensive evaluation.

[0044] Specifically, historical inflation information includes inflation rates in different periods. Inflation information is used to adjust fire loss data to make it comparable between different time points. Among them, historical inflation information includes annual inflation rate, consumer price index CPI, producer price index PPI, nominal GDP and real GDP, etc. Based on this historical inflation information, the impact of changes in currency value over time can be eliminated, and historical fire loss data can be adjusted to present value, making fire loss data in different years comparable.

[0045] Specifically, based on historical inflation information, we traverse historical fire data, obtain each piece of historical fire data, extract the year when the fire occurred and the scale of losses, and make inflation corrections year by year based on the difference between the year when the fire occurred and the current year.

[0046] Specifically, based on the loss scale in the inflation-corrected data, the historical factor value information is called to correct the factor value and obtain the factor value correction data. First, traverse each inflation-corrected data to extract the corrected loss scale; then, according to the year of the fire, find the factor value information of the corresponding year. Calculate the ratio of the factor value of the corresponding year to the factor value of the current year, obtain the factor value adjustment coefficient, and use the factor value adjustment coefficient to correct the loss scale.

[0047] Optionally, the above-mentioned historical element value information and current year element value are both value information of elements in the field related to forest fires.

[0048] Specifically, based on the loss scale in the factor value correction data, the control level is corrected in combination with the historical technical level information to generate standard historical fire data. First, each factor value correction data is traversed to extract the corrected loss scale. Then, based on the same method principle as the above factor value correction, according to the year of the fire, the technical level information of the corresponding year is found, and combined with the technical level information of the current year, the technical level adjustment coefficient is calculated to correct the control level. Among them, the loss after the control level correction is equal to the product of the loss after the factor value correction and the technical level adjustment coefficient.

[0049] Based on the historical fire data and the historical fire levels, construct a fire level prediction model;

[0050] In some embodiments, based on the historical fire data and the historical fire levels, a fire level prediction model is constructed, including:

[0051] Based on the historical fire data, features related to the fire level are extracted to obtain historical multimodal data; using the historical multimodal data as input samples and the historical fire level as output samples, multiple weak sensors are constructed and trained based on a decision tree; based on an integrated learning method, the fire level prediction model is generated according to the multiple weak sensors.

[0052] Specifically, the fire level prediction model is used to evaluate and predict the level of fire based on multimodal monitoring data. The fire level prediction model is built based on the random forest algorithm. First, features related to the fire level are extracted from historical fire data to obtain historical multimodal data. Among them, historical multimodal data include the time and location of the fire (geographic coordinates of the location of the fire, altitude, etc.), affected area (burned area, indirect impact range, etc.), weather conditions (temperature, humidity, wind speed, precipitation, etc.), vegetation characteristics (vegetation type, coverage, etc.), remote sensing images, etc.

[0053] Furthermore, the historical multimodal data is divided into a training set and a test set. The training set data is used to train multiple decision trees, each of which serves as a weak perceptron. Optionally, different training subsets are generated through different sampling methods (such as Bootstrap sampling) to increase the diversity of the model. Each generated weak perceptron can predict the fire level of the input sample. Then, based on ensemble learning methods, including Bagging, Boosting, and Random Forest, the training set data is used to train the final fire level prediction model.

[0054] Specifically, during the ensemble learning process, the test set data is used to evaluate the model performance, calculate indicators such as accuracy, precision, recall, F1 score, etc., adjust the model parameters according to the evaluation results, and perform model optimization.

[0055] In some embodiments, constructing a fire grade prediction model further includes:

[0056] Based on the standard historical fire data and the real-time forest fire assessment strategy, an update calibration is performed to obtain the standard historical fire level; based on the standard historical fire level and the standard historical fire data, an enhanced training set is constructed; based on the enhanced training set, enhanced training of the fire level prediction model is performed.

[0057] Specifically, through the real-time forest fire evaluation strategy, each standard historical fire data is traversed to calibrate the fire level and re-evaluate the level of each fire data. Among them, the real-time forest fire evaluation strategy is the fire level evaluation standard and strategy currently used. The standard historical fire level reflects the level of historical data under the fire level evaluation criteria of the current period.

[0058] Furthermore, the standard historical fire data and the standard historical fire level are integrated into one data set, and the integrated data set is used as an enhanced training set. Then, the enhanced training set is divided into a training set and a validation set, and the model is trained using the enhanced training set based on the previously constructed fire level prediction model. Among them, the enhanced training set introduces a real-time forest fire evaluation strategy, and at the same time expands the training samples of the fire level prediction model, which helps to improve the accuracy of fire level prediction for input samples.

[0059] Using the historical fire levels as sample input and the loss data in the standard historical fire data as sample output, a loss prediction model is constructed and trained;

[0060] Specifically, first, define the historical fire level as the sample input, and the loss data in the calibrated historical fire data as the sample output, extract the features related to loss prediction, such as fire level, fire area, fire cause, response time, casualties, etc., and standardize the features to ensure the effect of model training. Then, divide the data set into training set and test set according to the ratio of 8:2 or 7:3. Then select a suitable machine learning model, and use the training set to train the selected model and adjust the model parameters. Exemplary, include linear regression, support vector regression (SVR) and multi-layer perceptron (MLP). Then, use the test set to evaluate the trained model, including mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), etc., and adjust the model parameters for retraining according to the evaluation results until the model performance meets the preset performance requirements.

[0061] The above method steps, based on historical fire levels and calibrated loss data, construct a prediction model that can accurately predict the amount of fire losses and provide scientific decision-making support for relevant agencies.

[0062] The loss prediction model is combined with the fire grade prediction model to predict forest fire losses based on multimodal monitoring data, and the real-time forest fire grade and predicted forest fire losses are output.

[0063] In some embodiments, the loss prediction model is combined with the fire grade prediction model to predict forest fire losses based on multimodal monitoring data, and the real-time forest fire grade and the predicted forest fire losses are output, including:

[0064] Connect the output end of the fire level prediction model with the input end of the loss prediction model; activate the multimodal monitoring equipment, conduct real-time forest fire monitoring of the target scene, and obtain the multimodal monitoring data; transmit the multimodal monitoring data to the input end of the fire level prediction model, analyze and obtain the real-time forest fire level; the loss prediction model uses the real-time forest fire level as input, predicts the economic loss of the forest fire, and outputs the predicted forest fire loss.

[0065] Specifically, first, the output of the fire level prediction model is connected to the input of the loss prediction model so that the fire level prediction result can be used as the input of the loss prediction model. Then, the multimodal monitoring equipment is activated to collect relevant data of the forest fire in the target scene in real time. Exemplary data include temperature, humidity, wind speed, flame height, flame area, etc. Among them, the multimodal monitoring equipment includes remote sensing satellites, patrol drones, ground sensors, etc.

[0066] Furthermore, the collected multimodal monitoring data is transmitted to the input end of the fire grade prediction model, the fire grade prediction model receives and analyzes the multimodal monitoring data, and outputs the real-time grade of the forest fire. Then, the loss prediction model uses the real-time grade of the forest fire output by the fire grade prediction model as input, predicts the economic loss of the forest fire, and outputs the predicted loss amount of the forest fire. Through the above steps, based on the real-time data collection of the multimodal monitoring equipment, combined with the fire grade prediction model and the loss prediction model, the real-time prediction of the real-time grade of the forest fire and the economic loss is achieved.

[0067] Furthermore, the loss prediction model is combined with the fire grade prediction model to predict forest fire losses according to multimodal monitoring data, and the real-time forest fire grade and the predicted forest fire loss are output, which also includes:

[0068] Based on the predicted forest fire loss, the level calibration is performed in combination with the real-time forest fire evaluation strategy, and the real-time forest fire level is updated based on the calibration result; the updated real-time forest fire level is returned to the input end of the loss prediction model for iterative prediction; if the iterative prediction result is stable, the real-time forest fire level and the predicted forest fire loss are output.

[0069] Specifically, first, based on the real-time forest fire evaluation strategy, the level is calibrated according to the forest fire predicted loss, a new fire level is obtained, and the new fire level is used as the updated forest fire real-time level. Then, the updated forest fire real-time level is returned to the input end of the loss prediction model for iterative prediction.

[0070] Specifically, if in several consecutive iterative predictions, the changes in the real-time forest fire level and the predicted forest fire loss are within an acceptable range, that is, the changes are small or there is no change, then the iterative prediction results can be considered to be stable, and the output is the real-time forest fire level and the predicted forest fire loss.

[0071] Optionally, a threshold is set. If the change in the predicted results for several consecutive iterations is less than this threshold, the result is considered to be stable.

[0072] In the above method steps, the updated real-time forest fire level has a high accuracy. Through the iterative prediction method, the predicted real-time forest fire level and the predicted forest fire loss can be continuously approached to the true value, thereby improving the accuracy of forest fire loss prediction.

[0073] In summary, the forest fire loss prediction method based on multimodal data analysis provided by the present invention has the following technical effects:

[0074] Obtain historical fire data, and calibrate historical fire levels based on historical fire data; obtain historical auxiliary data, perform period correction on historical fire data, and obtain standard historical fire data, where historical auxiliary data includes historical element value information, historical technical level information, and historical inflation information; construct a fire level prediction model based on historical fire data and historical fire levels; use historical fire levels as sample inputs and loss data in standard historical fire data as sample outputs to construct and train a loss prediction model; combine the loss prediction model with the fire level prediction model, predict forest fire losses based on multimodal monitoring data, and output real-time forest fire levels and predicted forest fire losses. This achieves the technical effect of integrating multimodal data and improving prediction timeliness and accuracy.

[0075] Embodiment 2

[0076] Based on the same concept as the forest fire loss prediction method based on multimodal data analysis in the embodiment, Figure 2 As shown, the present application also provides a forest fire loss prediction device based on multimodal data analysis, the device comprising:

[0077] The data acquisition module 11 is used to obtain historical fire data and calibrate historical fire levels based on the historical fire data;

[0078] The associated acquisition module 12 is used to obtain historical auxiliary data, perform period correction on the historical fire data, and obtain standard state historical fire data, wherein the historical auxiliary data includes historical element value information, historical technology level information, and historical inflation information;

[0079] A fire grade prediction model module 13 is used to construct a fire grade prediction model based on the historical fire data and the historical fire grades;

[0080] The loss prediction model module 14 is used to construct and train a loss prediction model using the historical fire level as sample input and the loss data in the standard historical fire data as sample output;

[0081] The multimodal prediction module 15 is used to combine the loss prediction model with the fire grade prediction model, predict forest fire losses based on multimodal monitoring data, and output the real-time forest fire grade and the predicted forest fire loss.

[0082] Wherein, the data acquisition module 11 includes:

[0083] A historical data collection unit, used for interacting with forest fire data sources, and extracting the historical fire data based on a preset collection window, wherein the historical fire data includes the occurrence time, occurrence location, impact range, loss scale, and historical monitoring data;

[0084] The historical data calibration unit is used to obtain a series of forest fire assessment strategies and to identify and calibrate the historical fire levels based on the historical fire data.

[0085] Furthermore, the association collection module 12 includes:

[0086] A period information acquisition unit, used to acquire the historical element value information, the historical technology level information, and the historical inflation information according to the occurrence time of the historical fire data;

[0087] An inflation correction unit, configured to perform inflation correction on the loss scale by traversing the historical fire data based on the historical inflation information, and obtain inflation correction data;

[0088] An element value correction unit, configured to call the historical element value information based on the loss scale in the inflation correction data, perform element value correction, and obtain element value correction data;

[0089] A control level correction unit is used to correct the control level based on the loss scale in the element value correction data and in combination with the historical technical level information to generate the standard historical fire data.

[0090] Furthermore, the grade prediction model module 13 includes:

[0091] A feature extraction unit, used to extract features related to the fire level based on the historical fire data, and obtain historical multimodal data;

[0092] A basic construction unit, used to construct and train multiple weak perceptrons based on a decision tree, using the historical multimodal data as input samples and the historical fire levels as output samples;

[0093] An integrated learning unit is used to generate the fire level prediction model based on the multiple weak sensors based on the integrated learning method.

[0094] Furthermore, the grade prediction model module 13 also includes:

[0095] An updating and calibration unit, used to perform updating and calibration based on the standard state historical fire data and the real-time forest fire assessment strategy to obtain the standard state historical fire level;

[0096] A unit, configured to construct an enhanced training set based on the standard historical fire level and the standard historical fire data;

[0097] The enhanced training unit is used to perform enhanced training of the fire grade prediction model according to the enhanced training set.

[0098] Furthermore, the multimodal prediction module 15 includes:

[0099] A unit, used to connect the output end of the fire grade prediction model with the input end of the loss prediction model;

[0100] A multimodal monitoring unit, used to activate the multimodal monitoring device, perform forest fire monitoring of the target scene in real time, and obtain the multimodal monitoring data;

[0101] A real-time grade analysis unit, used for transmitting the multimodal monitoring data to the input end of the fire grade prediction model, and analyzing and obtaining the real-time grade of the forest fire;

[0102] The loss prediction unit is used to predict the economic loss of forest fire by taking the real-time level of forest fire as input through the loss prediction model, and output the predicted loss of forest fire.

[0103] Furthermore, the multimodal prediction module 15 also includes:

[0104] A level updating unit, configured to perform level calibration based on the predicted forest fire loss in combination with the real-time forest fire assessment strategy, and update the real-time forest fire level based on the calibration result;

[0105] An iterative loss prediction unit, used for returning the updated real-time forest fire level to the input end of the loss prediction model for iterative prediction;

[0106] The prediction output unit is used to output the real-time forest fire level and the predicted forest fire loss if the iterative prediction result is stable.

[0107] It should be understood that the embodiments mentioned in this specification focus on their differences from other embodiments. The specific embodiments in the aforementioned embodiment one are also applicable to the forest fire loss prediction device based on multimodal data analysis described in embodiment two. For the sake of brevity of the specification, they will not be further elaborated here.

[0108] It should be understood that the embodiments disclosed in this application and the above description can enable those skilled in the art to use this application to implement this application. At the same time, this application is not limited to the above-mentioned embodiments, and obvious modifications, combinations and substitutions of the embodiments mentioned in this application also fall within the scope of protection of this application.

Claims

1. Forest fire loss prediction method based on multimodal data analysis, It is characterized in that The method comprises: Acquiring historical fire data, and calibrating historical fire levels based on the historical fire data; Acquire historical auxiliary data, perform period correction on the historical fire data, and acquire standard historical fire data, wherein the historical auxiliary data includes historical element value information, historical technology level information, and historical inflation information; Based on the historical fire data and the historical fire levels, construct a fire level prediction model; Using the historical fire levels as sample input and the loss data in the standard historical fire data as sample output, a loss prediction model is constructed and trained; The loss prediction model is combined with the fire grade prediction model to predict forest fire losses based on multimodal monitoring data, and the real-time forest fire grade and predicted forest fire losses are output.

2. The forest fire loss prediction method based on multimodal data analysis according to claim 1, It is characterized in that Acquiring historical fire data, and calibrating historical fire levels based on the historical fire data, including: An interactive forest fire data source extracts the historical fire data based on a preset collection window, wherein the historical fire data includes the occurrence time, occurrence location, impact range, loss scale, and historical monitoring data; A series of forest fire assessment strategies are obtained, and based on the historical fire data, the historical fire levels are identified and calibrated.

3. The forest fire loss prediction method based on multimodal data analysis according to claim 2, It is characterized in that Obtain historical auxiliary data and perform period correction on the historical fire data, including: According to the occurrence time of the historical fire data, the historical element value information, the historical technology level information, and the historical inflation information are obtained; Based on the historical inflation information, traverse the historical fire data to perform inflation correction on the loss scale to obtain inflation correction data; Based on the loss scale in the inflation correction data, calling the historical factor value information, performing factor value correction, and obtaining factor value correction data; Based on the loss scale in the factor value correction data and combined with the historical technical level information, the control level is corrected to generate the standard historical fire data.

4. The forest fire loss prediction method based on multimodal data analysis according to claim 3, It is characterized in that Based on the historical fire data and the historical fire levels, a fire level prediction model is constructed, including: Based on the historical fire data, extract features related to the fire level to obtain historical multimodal data; Taking the historical multimodal data as input samples and the historical fire levels as output samples, a plurality of weak perceptrons are constructed and trained based on a decision tree; Based on an integrated learning method, the fire level prediction model is generated according to a plurality of the weak sensors.

5. The forest fire loss prediction method based on multimodal data analysis according to claim 4, It is characterized in that Building a fire grade prediction model also includes: Based on the standard historical fire data and the real-time forest fire assessment strategy, update and calibrate to obtain the standard historical fire level; Based on the standard historical fire level and the standard historical fire data, construct an enhanced training set; Based on the enhanced training set, enhanced training of the fire grade prediction model is performed.

6. The forest fire loss prediction method based on multimodal data analysis according to claim 5, It is characterized in that Combining the loss prediction model with the fire grade prediction model, predicting forest fire losses based on multimodal monitoring data, and outputting real-time forest fire grades and forest fire predicted losses, including: Connecting the fire grade prediction model output end to the loss prediction model input end; Activate the multimodal monitoring device to perform forest fire monitoring of the target scene in real time and obtain the multimodal monitoring data; Transmitting the multimodal monitoring data to the input end of the fire grade prediction model to analyze and obtain the real-time grade of the forest fire; The loss prediction model takes the real-time forest fire level as input, predicts the economic losses of forest fires, and outputs the predicted forest fire losses.

7. The forest fire loss prediction method based on multimodal data analysis according to claim 6, It is characterized in that Combining the loss prediction model with the fire grade prediction model, predicting forest fire losses based on multimodal monitoring data, and outputting the real-time forest fire grade and the predicted forest fire losses, further comprising: Based on the predicted forest fire loss, the level is calibrated in combination with the real-time forest fire assessment strategy, and the real-time forest fire level is updated based on the calibration result; Returning the updated forest fire real-time level to the input end of the loss prediction model for iterative prediction; If the iterative prediction result is stable, the real-time forest fire level and the predicted forest fire loss are output. 8.Forest fire loss prediction device based on multimodal data analysis, It is characterized in that The device comprises: A data acquisition module, the data acquisition module is used to obtain historical fire data and calibrate historical fire levels based on the historical fire data; An associated acquisition module, the associated acquisition module is used to obtain historical auxiliary data, perform period correction on the historical fire data, and obtain standard historical fire data, wherein the historical auxiliary data includes historical element value information, historical technology level information, and historical inflation information; A grade prediction model module, the grade prediction model module is used to construct a fire grade prediction model based on the historical fire data and the historical fire grades; A loss prediction model module, wherein the loss prediction model module is used to construct and train a loss prediction model using the historical fire level as a sample input and the loss data in the standard historical fire data as a sample output; A multimodal prediction module is used to combine the loss prediction model with the fire grade prediction model, predict forest fire losses based on multimodal monitoring data, and output real-time forest fire grades and predicted forest fire losses.

Images