阅读说明：本技术 一种基于rbf神经网络的农村住房台风灾害预估方法 (RBF neural network-based rural house typhoon disaster estimation method ) 是由 李强 郏鸿韬 陆泳竹 裘晴 章根良 冯传庆 于 2021-07-30 设计创作，主要内容包括：本发明公开了一种基于RBF神经网络的农村住房台风灾害预估方法,包括：筛选出对评估地区有影响的台风信息；对台风关键参数标定,并利用风场模型模拟确定台风风速；根据评估地区内农村住房保险承保与理赔数据,获取各台风灾害影响下各乡镇农村住房对应的保额损失率；建立预测模型,将得到的台风风速数据和保额损失率数据代入预测模型进行训练,获取最优的RBF神经网络结构；通过预测模型预测得到不同台风风速下各乡镇保额损失率,根据预测结果对评估地区进行乡镇级分辨率的农村住房台风灾害评估。本发明借助公共巨灾保险数据和RBF神经网络可实现乡镇级高分辨率的农村住房台风灾情预评,评估结果可靠,对防灾减灾具有重要意义。(The invention discloses a rural house typhoon disaster estimation method based on a RBF neural network, which comprises the following steps: screening typhoon information which influences the evaluation area; calibrating typhoon key parameters, and determining typhoon wind speed by using a wind field model; acquiring the corresponding premium loss rate of each rural housing of the villages and towns under the influence of each typhoon disaster according to the insurance underwriting and claim settlement data of the rural housing in the evaluation area; establishing a prediction model, substituting the obtained typhoon wind speed data and the obtained quota loss rate data into the prediction model for training to obtain an optimal RBF neural network structure; and predicting the loss rate of the reserves of each village and town under different typhoon wind speeds through a prediction model, and performing village-and-town-level resolution rural house typhoon disaster assessment on the assessment area according to the prediction result. The method can realize rural house typhoon disaster situation pre-evaluation with high resolution at a village-tranquility level by means of public disaster insurance data and the RBF neural network, has reliable evaluation results, and has important significance for preventing and reducing disasters.)

1. A rural house typhoon disaster prediction method based on an RBF neural network is characterized by comprising the following steps:

step 1: screening typhoon information which influences the evaluation area from a typhoon historical information database;

step 2: calibrating typhoon key parameters of the screened typhoon information, and determining the typhoon wind speed by using a wind field model;

and step 3: acquiring the corresponding premium loss rate of each rural housing of the villages and towns under the influence of each typhoon disaster according to the insurance underwriting and claim settlement data of the rural housing in the evaluation area;

and 4, step 4: establishing a prediction model based on the RBF neural network, substituting the obtained typhoon wind speed data as an input variable of the prediction model and the obtained premium loss rate data as an output variable of the prediction model into the prediction model for training to obtain an optimal RBF neural network structure;

and 5: and predicting the loss rate of the reserves of each village and town under different typhoon wind speeds through a prediction model, and performing village-and-town-level resolution rural house typhoon disaster assessment on the assessment area according to the prediction result.

2. The RBF neural network-based rural house typhoon disaster estimation method according to claim 1, wherein: in step 4, the RBF neural network structure comprises an input layer, a hidden layer and an output layer;

each input sample of the input layer is:wherein x is_pRepresenting the typhoon wind speed, m representing the number of input samples, and T representing the transpose of a matrix formed by the input samples;

the input layer realizes x by radial basis function_p→R(x_p-c_i) The activation function of the spatial non-linear mapping transformation of (2) is:wherein R (x)_p-c_i) For the mapped value, exp () represents an exponential function with e as the base, | x_p-c_iI is the Euclidean distance, c_iIs the hidden layer node center, and sigma is the variance of the basis function;

the output layer realizes R (x) through weighted combination_p-c_i)→y_jIs expressed as:(n is the number of output layers), where ω is_ijFor the connection weight from hidden layer to output layer, i is 1,2,3, h (h is the number of hidden layer nodes corresponding to the training sample), y_jAnd outputting a value for the j (th) quota loss rate corresponding to the data.

3. The RBF neural network-based rural house typhoon disaster estimation method according to claim 2, wherein: in step 4, the method specifically comprises the following steps:

step 41: randomly selecting I vectors different from each other as initial clustering center c_i(0) (i ═ 1,2,3.. h), presetting the value of a threshold epsilon, and obtaining the final hidden layer center c through calculation_i(h)；

Step 42: solving for the variance σ, expressed as(h is the number of hidden layer nodes corresponding to the training sample), wherein c_maxSelecting the maximum distance between centers;

step 43: calculating a connection weight omega_ijComprises the following steps:(h represents the number of hidden layer nodes corresponding to the training sample), where c_maxThe maximum distance between the centers is chosen.

4. The RBF neural network-based rural house typhoon disaster estimation method according to claim 3, wherein: obtaining a final hidden layer center c_i(h) The specific calculation method comprises the following steps:

step 411: calculating the Euclidean distance between the wind speed of each typhoon and the clustering center point, and expressing as follows: | x_p-c_i(h) I |, wherein p ═ 1,2,3.. m, i ═ 1,2,3.. h;

step 412: according to the wind speed x of each typhoon_pThe minimum Euclidean distance from the cluster center determines its classification, i.e. when i^*(x_p)＝min||x_p-c_i(h) When | x_pIs classified as the ith^*Class, dividing the total sample into I subsets: u shape₁(n),U₂(n),...U_I(n)；

Step 413: adopting a competition learning rule to adjust, and expressing as:

step 414: calculate | c_i(h+1)-c_i(h) I, when c_i(h+1)-c_i(h) When | < epsilon, the center c of the final hidden layer is obtained_i(h) (ii) a Otherwise, let h be h +1, and jump to step 411.

5. The RBF neural network-based rural house typhoon disaster estimation method according to claim 1, wherein: in step 1, typhoon information is screened by using the principle that a typhoon path passes through the 500km range of an evaluation area, and the typhoon information is divided into three types according to the distance from a typhoon landing point to the evaluation area: type I, land climbing is the province where the evaluation area is located; type II, landing is adjacent provinces of the assessment area; type iii, landed far away or not but with a typhoon-centric route into the 500km range of the assessment area.

6. The RBF neural network-based rural house typhoon disaster estimation method according to claim 1, wherein: in the step 2, calculating typhoon key parameters by adopting a Vickery empirical model, and simulating by adopting a Yan Meng wind field model to obtain 10min average maximum wind speed at a position of 10m height close to the ground in an evaluation area under the influence of each typhoon.

7. The RBF neural network-based rural house typhoon disaster estimation method according to claim 1, wherein: the typhoon key parameter in step 2 comprises the maximum wind speed radius R_maxHolland pressure profile parameter B and center gas pressure.

8. The RBF neural network-based rural house typhoon disaster estimation method according to claim 1, wherein: in step 3, the township or the street where the safety accident of the rural housing occurs can be accurately positioned according to the emergency address so as to realize township resolution disaster assessment.

9. The RBF neural network-based rural house typhoon disaster estimation method according to claim 3, wherein: in step 4, the accuracy of the obtained prediction model is checked according to the convergence degree of the training curve and the precision requirement of the data error, and if the accuracy meets the requirement, the step 5 is continued; otherwise, jump to step 42 to reset the value of the threshold epsilon.

10. The RBF neural network-based rural house typhoon disaster estimation method according to claim 1, wherein: in step 5, the quota loss rate of each village and town under the typhoon wind speed of different designs is predicted through the prediction model, the maximum quota loss rate and the corresponding village and town under the influence of the typhoon wind speed are obtained according to the prediction result, and an evaluation report is generated.

Technical Field

The invention relates to the technical field of disaster weather prediction, in particular to a rural house typhoon disaster prediction method based on a RBF neural network.

Background

China is a typhoon-suffering country, and in recent years, with global warming, the average typhoon intensity generated by the western pacific is in a descending trend, but the average typhoon landing intensity in China is in a gradually increasing trend. Typhoon is one of the products of ocean interaction with the atmosphere, often accompanied by strong winds, heavy rain and storm tides. It has strong destructive power and can cause secondary disasters such as flood, debris flow and the like. According to the statistics of data, typhoon is the most influencing various meteorological disasters in coastal cities all the year round, the average number of collapsed houses and death persons is the highest, compared with cities, houses in rural areas have poorer wind resistance, and particularly, some old houses are extremely easy to be damaged by typhoon.

According to the third agricultural census data in China, the farmer houses are mainly of brick-concrete and brick (stone) wood structures. The housing is 13182 thousands of households with a brick-concrete structure, and accounts for 57.2%; 5993 ten thousands households of brick (stone) wood structure account for 26.0%; 2884 ten thousands households of the reinforced concrete structure account for 12.5 percent; 640 million households of the bamboo grass soil blank structure account for 2.8%; 329 of the other structures accounted for 1.4%. Except for reinforced concrete structures with good wind resistance, buildings with structures such as brick, concrete, masonry, bamboo and grass can collapse due to the coming of typhoons. Therefore, it is increasingly important to reasonably evaluate and predict the disaster of the agricultural houses.

Aiming at typhoon disaster, a plurality of scholars at home and abroad carry out research. In the aspect of disaster-causing factor dangerousness, scholars establish a typhoon disaster situation estimation model based on the comprehensive disaster situation correlation degree and the typhoon disaster-causing factor; the maximum wind speed of scholars in the above-sea years is taken as an application object, and a probability statistical method is adopted to carry out quantitative diagnosis on the risk degree of the disaster-causing factor; scholars establish a neural network based on a principal component analysis method, and perform fitting evaluation on 5 typhoons in Zhejiang province by taking a direct economic loss index as an evaluation index. In the aspect of vulnerability of a disaster-bearing body, a learner comprehensively considers disaster-causing factors and characteristics of the disaster-bearing body and analyzes typhoon disaster risks in coastal areas of China according to a natural disaster system theory; the method comprises the following steps that a scholars evaluate disaster risks caused by typhoons in coastal areas of China by establishing an index system based on an analytic hierarchy process; students combine typhoon disasters with social and economic factors to point out that the main reasons for disaster damage are closely related to population and economic currency; some scholars establish a typhoon vulnerability equation of the rural housing in Zhejiang province by using the insurance data of the rural housing in Zhejiang province and give related parameters.

At present, emergency management departments in various regions in China mainly collect and collect typhoon disaster situation data through basic disaster information personnel, the data time sequence is long, the number of human interference factors is large, the quality of historical data is not ideal, the civil administration indexes such as the number of disaster-stricken residents, casualties, direct economic loss and the like are emphasized, the counted typhoon disaster situation data lack a unified standard, and the arrangement and the correction of the disaster situation data are difficult. And the departments of meteorological water conservancy and the like select related factor factors based on historical data and mainly focus on disaster-causing factor statistics, determine weight coefficients according to two measurement values of the maximum wind speed and the accumulated rainfall generated by typhoon and the influence degree of the typhoon disaster, and finally construct a typhoon disaster comprehensive risk zoning model to obtain risk zoning results, but the risk zoning results cannot be accurate to the risk zones of various towns and villages, so that the disaster situation data provided by the departments of meteorological water conservancy and the like has low spatial resolution, is lack of disaster-bearing body statistics and is lack of quantitative evaluation of the risk of the typhoon disaster.

Therefore, for business parts such as meteorology and government, the prediction of possible disaster loss caused by typhoon in the future is more realistic.

Disclosure of Invention

Aiming at the problems in the prior art, the RBF neural network-based method for estimating the typhoon disaster of the rural housing is provided, a typhoon disaster situation evaluation model from a disaster-causing factor to disaster-bearing body loss is effectively established, the problems that the traditional risk evaluation space resolution is low and the risk measurement is difficult are solved, the risk disaster estimation of various villages and towns is accurately performed, the typhoon disaster risk is quantitatively estimated, and scientific basis is provided for the follow-up establishment of applications such as typhoon disaster insurance policies, typhoon emergency plans and the like.

The specific technical scheme is as follows:

a rural house typhoon disaster estimation method based on an RBF neural network comprises the following steps:

step 1: screening typhoon information which influences the evaluation area from a typhoon historical information database;

step 2: calibrating typhoon key parameters of the screened typhoon information, and determining the typhoon wind speed by using a wind field model;

and step 3: acquiring the corresponding premium loss rate of each rural housing of the villages and towns under the influence of each typhoon disaster according to the insurance underwriting and claim settlement data of the rural housing in the evaluation area;

and 4, step 4: establishing a prediction model based on an RBF neural network, substituting the obtained typhoon wind speed data as an input variable of the prediction model and the obtained premium loss rate data as an output variable of the prediction model into the prediction model for training to obtain an optimal RBF neural network structure;

and 5: and predicting the loss rate of the reserves of each village and town under different typhoon wind speeds through a prediction model, and performing village and town resolution ratio rural house typhoon disaster assessment on the assessment area according to the prediction result.

Compared with the prior art, the method effectively establishes the typhoon disaster situation evaluation model from disaster-causing factors to disaster-bearing body loss by using the RBF neural network, can realize rural house typhoon disaster situation estimation with high rural resolution by means of public disaster insurance data, solves the problems of low spatial resolution and difficult risk measurement in the traditional risk evaluation, accurately estimates the risk disasters of various towns and villages, quantitatively evaluates the typhoon disaster risks, provides scientific basis for subsequent applications of typhoon disaster insurance policy, typhoon emergency plan and the like, and can effectively reduce casualties and economic loss caused by wind-induced collapse.

Further, in step 4, the RBF neural network structure includes an input layer, a hidden layer and an output layer;

each input sample of the input layer is:wherein x is_pRepresenting the typhoon wind speed, m representing the number of input samples, and T representing the transpose of a matrix formed by the input samples;

the input layer realizes x by radial basis function_p→R(x_p-c_i) The activation function of the spatial non-linear mapping transformation of (2) is:wherein R (x)_p-c_i) For the mapped value, exp () represents an exponential function with e as the base, | x_p-c_iI is the Euclidean distance, c_iIs the hidden layer node center, and sigma is the variance of the basis function;

the output layer realizes R (x) through weighted combination_p-c_i)→y_jIs expressed as:j is 1,2,.. n (n is the number of output layers), where ω is_ijA connection weight value from the hidden layer to the output layer, i ═ 1,2,3, h (h is the number of hidden layer nodes corresponding to the training sample), y_jAnd outputting a value for the j (th) quota loss rate corresponding to the data.

The RBF neural network provided by the invention has the advantages of simple structure, high convergence speed and small overall prediction error.

Further, in step 4, the method specifically includes the following steps:

step 41: randomly selecting I vectors different from each other as initial clustering center c_i(0) (i ═ 1,2,3.. h), presetting the value of a threshold epsilon, and obtaining the final hidden layer center c through calculation_i(h)；

Step 42: solving for the variance σ, expressed asi 1, 2., h (h is the number of hidden layer nodes corresponding to the training sample), wherein c_maxSelecting the maximum distance between centers;

step 43: calculating a connection weight omega_ijComprises the following steps:i 1, 2.. h (h represents the number of hidden layer nodes corresponding to the training sample), wherein c_maxThe maximum distance between the centers is chosen.

Further, a final hidden layer center c is obtained_i(h) The specific calculation method comprises the following steps:

step 411: calculating the Euclidean distance between the wind speed of each typhoon and the clustering center point, and expressing as follows: | x_p-c_i(h) I |, wherein p ═ 1,2,3.. m, i ═ 1,2,3.. h;

step 412: according to the wind speed x of each typhoon_pThe minimum Euclidean distance from the cluster center determines its classification, i.e. when i^*(x_p)＝min||x_p-c_i(h) When | x_pIs classified as the ith^*Class, dividing the total sample into I subsets: u shape₁(n),U₂(n),...U_I(n)；

Step 413: adopting a competition learning rule to adjust, and expressing as:

step 414: calculate | c_i(h+1)-c_i(h) I, when c_i(h+1)-c_i(h) When | < epsilon, acquiring the center c of the final hidden layer_i(h) (ii) a Otherwise, let h be h +1, and jump to step 411.

According to the invention, through continuous iteration tests, the optimal weight of the RBF neural network is obtained, and the optimal RBF neural network is further rapidly obtained.

Further, in step 1, typhoon information is screened by using the principle that a typhoon path passes through the 500km range of the evaluation area, and is divided into three types according to the distance from a typhoon landing point to the evaluation area: type I, land climbing is the province where the evaluation area is located; type II, landing is adjacent provinces of the assessment area; type iii, landed far away or not but with a typhoon-centric route into the 500km range of the assessment area. By the method, the original typhoon information data are screened, and typhoon information influencing an evaluation area is selected for sorting, classifying and analyzing, so that the accuracy of input sample variables is ensured, and the prediction precision of a prediction model is improved.

Further, in the step 2, calculating typhoon key parameters by adopting a Vickery empirical model, and simulating by adopting a Yan Meng wind field model to obtain 10min average maximum wind speed at a position of 10m height close to the ground in the evaluation area under the influence of each typhoon. Therefore, typhoon wind speed data can be accurately obtained, accuracy of input sample variables is guaranteed, and prediction accuracy of the prediction model is improved.

Further, the typhoon key parameter in step 2 comprises the maximum wind speed radius R_maxHolland gas pressure profile parameter B and center gas pressure.

Further, in step 3, the township or street where the safety accident of the rural housing occurs can be accurately positioned according to the emergency address to realize township resolution disaster assessment.

Further, in step 4, the accuracy of the obtained prediction model is checked according to the convergence degree of the training curve and the precision requirement of the data error, and if the accuracy meets the requirement, the step 5 is continued; otherwise, jump to step 42 to reset the value of the threshold epsilon. Therefore, the accuracy of the prediction model is detected, and the optimal RBF neural network model is obtained.

Further, in step 5, the quota loss rate of each village and town under the typhoon wind speed of different designs is predicted through the prediction model, the maximum quota loss rate and the corresponding village and town under the influence of the typhoon wind speed are obtained according to the prediction result, and an evaluation report is generated.

The estimation method provided by the invention can be popularized to other areas for typhoon disaster risk assessment and analysis, is beneficial to the development of quakeproof and disaster reduction work of all villages and towns in cities, and has very important practical significance.

Drawings

FIG. 1 is a flow chart of a method for estimating typhoon disasters of rural houses based on an RBF neural network;

FIG. 2a is a diagram illustrating the loss rate of the Bay in villages and towns in Ningbo city under the influence of type I No. 1416 typhoon in the embodiment of the present invention;

FIG. 2b is a diagram illustrating the loss rate of the Bay of villages and towns in Ningbo city under the influence of type I No. 1509 typhoon in the embodiment of the present invention;

FIG. 2c is a diagram illustrating the loss rate of the type I1814 typhoon in each town of Ningbo city;

FIG. 2d is a diagram illustrating the loss rate of the Bay in villages and towns in Ningbo city under the influence of type I1909 typhoon in the embodiment of the present invention;

FIG. 2e is a diagram illustrating the loss rate of the Bay in villages and towns in Ningbo city under the influence of type I No. 1918 typhoon in the embodiment of the present invention;

FIG. 3a is a diagram illustrating the loss rate of the coverage of each town in Ningbo city under the influence of type II 1410 typhoon in the embodiment of the present invention;

FIG. 3b is a diagram illustrating the loss rate of the coverage of each town in Ningbo city under the influence of type II 1614 typhoon in the embodiment of the present invention;

FIG. 3c is a graph showing the loss rate of the Bao in Ningbo city under the influence of type II typhoon No. 1709 in the embodiment of the present invention;

FIG. 3d is a diagram illustrating the loss rate of the coverage of each town in Ningbo city under the influence of type II type 1710 typhoon in the embodiment of the present invention;

FIG. 3e is a diagram illustrating the loss rate of the coverage of each town in Ningbo city under the influence of type II 1808 typhoon in the embodiment of the present invention;

FIG. 3f is a graph showing the percentage loss of the total amount of villages and towns in Ningbo city under the influence of type II No. 1810 typhoon in the embodiment of the present invention;

FIG. 3g shows the loss rate of the Bao in Ningbo city under the influence of type II 1812 typhoon in the embodiment of the present invention;

FIG. 3h is a graph showing the percentage loss of Ningbo city under the influence of type II 1818 typhoon in the embodiment of the present invention;

FIG. 4a shows the loss rate of the balance of villages and towns in Ningbo city under the influence of type III No. 1407 typhoon in the embodiment of the present invention;

FIG. 4b is a diagram illustrating the loss rate of the third type 1408 typhoon in the Ningbo city under the influence of the second embodiment of the present invention;

FIG. 4c is a diagram illustrating the loss rate of the coverage of each town in Ningbo city under the influence of type III 1412 typhoon in the embodiment of the present invention;

FIG. 4d is a diagram illustrating the loss rate of the coverage of each town in Ningbo city under the influence of type III 1616 typhoon in the embodiment of the present invention;

FIG. 4e is a diagram illustrating the loss rate of the coverage of each town in Ningbo city under the influence of type III 1618 typhoon in the embodiment of the present invention;

FIG. 4f is a diagram showing the loss rate of the balance of villages and towns in Ningbo city under the influence of type III No. 1703 typhoon in the embodiment of the present invention;

FIG. 4g shows the loss rate of the Bay in villages and towns in Ningbo city under the influence of type III 1718 typhoon in the embodiment of the present invention;

FIG. 4h shows the loss rate of the balance of villages and towns in Ningbo city under the influence of type III 1819 typhoon in the embodiment of the present invention;

FIG. 4i is a diagram illustrating the loss rate of the Bay of villages and towns in Ningbo city under the influence of type III 1825 typhoon in the embodiment of the present invention;

FIG. 4j is a diagram illustrating the loss rate of the coverage of each town in Ningbo city under the influence of type III 1905 typhoon in the embodiment of the present invention;

FIG. 4k is a diagram illustrating the loss rate of the balance of villages and towns in Ningbo city under the influence of type III type No. 1913 typhoon in the embodiment of the present invention;

FIG. 4l is a diagram illustrating the loss rate of the balance of villages and towns in Ningbo city under the influence of type III type No. 1917 typhoon in the embodiment of the present invention;

fig. 5 shows the total loss rate of the protection of the rural houses in various villages and towns of Ningbo caused by typhoon in 2014-2019 in the embodiment of the present invention;

FIG. 6 is a schematic diagram of the RBF neural network structure in an embodiment of the present invention;

FIG. 7 is a graph of RBF neural network model training curves;

FIG. 8 is a graph of training BP neural network model;

FIG. 9 is a graph comparing the predicted values of BP and RBF neural network models with the actual rate of loss of the warranty;

FIG. 10 is a graph showing the comparison of mean square error between BP and RBF model prediction results;

FIG. 11a is a graph of RBF neural network predicting loss rate at a design wind speed of 10 years;

FIG. 11b is a graph of RBF neural network prediction of loss rate at a design wind speed of 50 years;

FIG. 11c shows the RBF neural network predicting the loss rate at the design wind speed of 100 years.

Detailed Description

The technical solutions in the embodiments of the present invention will be described clearly and completely with reference to the accompanying drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making an invasive task, are within the scope of the present invention.

Please refer to fig. 1, which is a flowchart of the method for estimating typhoon disaster of rural houses based on RBF neural network of the present invention, the method for estimating typhoon disaster of rural houses based on RBF neural network comprises the following steps:

step 1: and screening typhoon information which has influence on the evaluation area from a typhoon historical information database.

The method comprises the following steps of screening typhoon information which influences an evaluation area from a meteorological station of the evaluation area and CMA-STI tropical cyclone optimal path data set on the basis that a typhoon path passes through the 500km range of the evaluation area, wherein the information comprises the following steps: typhoon number and name, typhoon starting time, typhoon ending time, typhoon influence assessment area time, typhoon wind speed observed by a meteorological station, typhoon path central air pressure and typhoon path longitude and latitude. The typhoon path information is divided into three types according to the distance from a landing point to an evaluation area: type I, land logging is the province where the assessment area is located; type II, landing is adjacent provinces of the assessment area; type iii, landed far away or not but with its central path going within 500km of the area under evaluation.

In one embodiment, on the principle that a typhoon path passes through the 500km range of a Ningbo area, 25 pieces of typhoon path information which has influence on the Ningbo area in 2014-2019 are screened out from the meteorological station and CMA-STI tropical cyclone optimal path data set, and the typhoon can be divided into three types according to the path information: type I, the login point is in the coastal areas of Zhejiang, Taizhou and Ningbo; type II, the landing points are Fujian, Shanghai and other adjacent provinces and cities; type III, the landing point is far away areas such as Guangdong and the like or the landing point is not landed but the central path of the landing point enters the range of Ningbo 500 km.

Step 2: and calibrating typhoon key parameters of the screened typhoon information, and determining the typhoon wind speed by utilizing a wind field model.

In one embodiment, a Yan Meng wind field model is used herein to simulate the near-ground typhoon wind speed in a Ningbo area, since the daily maximum 10min average wind speed observed by a meteorological station during typhoon exposure is found to be relatively small. Before wind field simulation, typhoon key parameters need to be calibrated, and a Vickery empirical model is adopted to determine the maximum wind speed radius R_maxKey parameter values such as Holland air pressure profile parameter B and central air pressure; and determining the typhoon wind speed by adopting a Yan Meng wind field model simulation, and obtaining the 10min average maximum wind speed at the 10m height position close to the ground in the Ningbo area under the influence of 25 typhoons.

Please refer to table 1, which is a table of 25 typhoon feature information screened according to an embodiment of the present invention.

Table 1: 25 pieces of typhoon characteristic information

As can be seen from table 1, the average typhoon affecting Ningbo area per year in 2014-2019 is 4 to 5. Among them, type I typhoons, which land on the coast of zhejiang and have serious influence on ningbo, occur on average 1 time per year. The 25 typhoons screened affected Ningbo for 3 to 4 days on average, the short duration of effect was only 1 day (No. 1918 Mina typhoon), and the long duration of effect reached 8 days (No. 1812 Yunlingtai).

And step 3: and acquiring the corresponding premium loss rate of each rural housing in the villages and towns under the influence of each typhoon disaster according to the insurance underwriting and claim settlement data of the rural housing in the evaluation area.

And (4) according to insurance underwriting and claim settlement data of rural houses in the evaluation area, counting to obtain the corresponding premium loss rate of each village and town under the influence of each typhoon disaster. The loss rate of the insurance sum of each village and town is equal to the ratio of the total real claim sum of each village and town to the total premium of each village and town.

The rural house insurance data content comprises underwriting and claim settlement data: the underwriting data is obtained by extracting indexes such as an insurance policy number, the number of insured households, an underwriting address, the amount of each household insurance, total insurance fee and the like from the insurance policy; the claim settlement data comprises insurance policy number, insurance date, insurance reason, insurance address, loss of insurance target, actual claim amount and the like. The insurance policy numbers in the two types of data correspond to each other, and the village and town or streets where the safety accidents of the rural houses occur can be accurately positioned by utilizing the emergency address so as to realize the disaster assessment of the village and town level resolution.

In an embodiment, the agricultural house quota loss rates of the Ningbo villages and towns under the influence of different typhoon paths in 2014-2019 are counted, representative typhoons are selected for analysis, and the quota loss rates of the Ningbo villages and towns under the influence of 25 typhoons are specifically shown in fig. 2a to 4 l. It can be seen from the figure that the difference of the disaster and the loss caused by typhoons on different landings is large, the total loss rate of the balance of the villages and the towns is in positive correlation with the wind speed of the typhoons affecting the Ningbo areas, that is, the loss of the typhoons (I-type typhoons) logged on the coast of Zhejiang is larger than that of the typhoons (II-type typhoons) logged on the farmhouses in the Ningbo areas, and particularly, the typhoons No. 1509, 'Brillian', and the typhoons No. 1909, 'Liqima', generate huge damage to the rural farmhouses in the villages and the towns. While the type III typhoon, such as the 1616 typhoon "Mallerka", has no landing and does not affect the wind speed in the Ningbo area, the central path of the typhoon enters the Ningbo 500km range in the evolution process, so that great loss is still caused to rural houses in various villages and towns of Ningbo, the main reason is that under the attack of the typhoon spiral wind and rain zone, secondary disasters such as flooding, landslide and debris flow occur in the Ningbo area, and the rural houses are easily damaged by the secondary disasters.

The data of the insuring in 2014-charge 2019 and the data of the claims due to typhoon are summarized and sorted according to the county scale, and the specific information is shown in table 2.

TABLE 2 summary of data on agricultural house insurances in counties and counties of Ningbo caused by typhoon

It can be seen from table 2 that the difference of the insurance application situation of the county and towns of Ningbo is large. Cixi is large in population base, so that the total amount of the insurable money is obviously higher than that of other counties. The county with higher total pay amount is Yuyao, Xiangshan, Ninghai and Cixi. The counties with higher percentage of loss of the balance are Xiangshan, Bei Lun, Ninghai and Yuyao.

In order to more intuitively analyze the relationship between typhoon disaster insurance and rural house risk, fig. 5 shows the total loss rate of the quota of each rural and rural house in Ningbo caused by typhoon in 2014-plus 2019. As can be seen from table 2, the highest percentage loss in the ningbo area is obtained in the ningbo xiangshan as the place where typhoon is repeatedly registered, and the percentage loss in the maojiang county is close to 0.01%. In addition, the loss rate of the balance in the areas of the northern Lun, the Ninghai and the Yuyao is higher than the average value in the Ningbo area, wherein the loss rate of the northern Lun Meishan village, the Ninghai tea yard village and the Yuyao deer pavilion village is 0.007252%, 0.003258% and 0.009502% respectively. According to the specific positions of the towns, the towns belong to mountain towns, and the buildings in the mountain areas mainly adopt a brick (stone) mixed structure, so that the structure system is difficult to resist the influence of strong wind and rainstorm caused by typhoon. Thus, these areas have a greater rate of loss of warranty and there is a greater risk of damage to the building structure.

And 4, step 4: and establishing a prediction model based on the RBF neural network, substituting the obtained typhoon wind speed data as an input variable of the prediction model and the obtained premium loss rate data as an output variable of the prediction model into the prediction model for training to obtain an optimal RBF neural network structure.

The radial basis function neural network (hereinafter referred to as RBF neural network) is a three-layer feedforward neural network, can approximate any nonlinear function, has good global approximation performance and generalization capability, and has a fast learning convergence speed.

Please refer to fig. 6, which is a schematic diagram of an RBF neural network structure according to an embodiment of the present invention. In the RBF neural network, the first layer is an input layer, samples are input, and the number of nodes of the input layer is the dimension of the samples. The second layer is a hidden layer, also called a radial base layer, the activation function of the node uses a radial basis function, the hidden layer carries out nonlinear transformation on an input sample, and a low-dimensional space is mapped to a high-dimensional space, so that the problem that the low-dimensional space is not solved is solved. The third layer is an output layer, also called a linear layer, and the hidden layer and the output layer are connected by weight, which is to perform a linear transformation on the output of the hidden layer.

In the present invention, each sample is a vector, and each input sample of the input layer is:

wherein x is_pRepresenting the typhoon wind speed, m representing the number of input samples, and T representing the transpose of the matrix formed by the input samples.

The input layer realizes x by radial basis function_p→R(x_p-c_i) The activation function of the spatial non-linear mapping transformation of (2) is:

wherein R (x)_p-c_i) For the mapped value, exp () represents an exponential function with e as the base, | x_p-c_i| | is the European distance, c_iσ is the variance of the basis function for the hidden layer node center.

The output layer realizes R (x) through weighted combination_p-c_i)→y_jIs linearly mapped toComprises the following steps:

j ═ 1,2,. and n (n is the number of output layers)

Wherein, ω is_ijFor the connection weight from the hidden layer to the output layer, i is 1,2,3, h (h is the number of hidden layer nodes corresponding to the training sample), y_jAnd outputting a value for the j (th) quota loss rate corresponding to the data.

In the step 4, the method specifically comprises the following steps:

step 41: randomly selecting I vectors different from each other as initial clustering center c_i(0) (i ═ 1,2,3.. h), presetting the value of a threshold epsilon, and obtaining the final hidden layer center c through calculation_i(h)。

Wherein, a final hidden layer center c is obtained_i(h) The specific calculation method comprises the following steps:

step 411: calculating the Euclidean distance between the wind speed of each typhoon and the clustering center point, and expressing as follows: | x_p-c_i(h) I |, wherein p ═ 1,2,3.. m, i ═ 1,2,3.. h;

step 412: according to the wind speed x of each typhoon_pThe minimum Euclidean distance from the cluster center determines its classification, i.e. when i (x)_p)＝min||x_p-c_i(h) When | x_pClassified as class I, dividing all samples into I subsets: u shape₁(n),U₂(n),...U_I(n)；

Step 413: adopting a competition learning rule to adjust, and expressing as:

step 414: calculate | c_i(h+1)-c_i(h) I, when c_i(h+1)-c_i(h) When | < epsilon, acquiring the center c of the final hidden layer_i(h) (ii) a Otherwise, let h be h +1, and jump to step 411.

Step 42: solving for the variance σ, expressed asi 1, 2., h (h is the number of hidden layer nodes corresponding to the training sample), wherein c_maxThe maximum distance between the centers is chosen.

Step 43: calculating a connection weight omega_ijComprises the following steps:i 1, 2.. h (h represents the number of hidden layer nodes corresponding to the training sample), wherein c_maxThe maximum distance between the centers is chosen.

With typhoon wind speed as input variable x_pThe loss rate of the balance of each village and town is taken as an output variable y_jSelecting proper radial basis function expansion coefficients to construct a radial basis function network quota loss rate prediction model, and setting a corresponding training set and a corresponding training target, specifically, creating a radial basis function network with a radial basis function expansion coefficient of 1.2, and setting a mean square error target of 0.001. The model selects 80% of sample data to train, and the rest 20% of sample data is used for checking and calibrating. The model training curve and the allowance loss rate comparison graph are respectively shown in fig. 7 and 9, the accuracy of the obtained prediction model is checked according to the convergence degree of the model training curve and the precision requirement of the data error, if the requirement is met, the set mean square error target 0.001 is reached, the next step is continued, and if the requirement is not met, the model parameter value is adjusted to rebuild the neural network until the condition is met.

And 5: and predicting the loss rate of the reserves of each village and town under different typhoon wind speeds through a prediction model, and performing village and town resolution ratio rural house typhoon disaster assessment on the assessment area according to the prediction result.

Forecasting the damage rate of each village and town under the typhoon wind speed of different designs through a forecasting model, obtaining the maximum damage rate and the corresponding village and town under the influence of the typhoon wind speed according to a forecasting result, and generating an evaluation report, wherein the evaluation report comprises: the size of the typhoon wind speed, estimated premium loss and the distribution condition of the rural house disasters.

In one embodiment, the corresponding design wind speeds are calculated through the wind pressures of Ningbo city specified by the building structure load code of 10 years, 50 years and 100 years, and the values are 21.91m/s, 28.28m/s and 30.98m/s respectively. The trained neural network model is used to predict the loss rate of the sum of the villages and towns in Ningbo city at the designed wind speed, and the results are shown in FIGS. 11a to 11 c. It can be seen from the figure that the rate of loss of the protection of rural houses increases with the increase of the wind speed, which means that the possibility of occurrence of disasters increases. Under the design wind speed which is met once in 100 years, most rural houses in Ningbo areas suffer from large damage, and especially the loss rate of the protection of rural houses in rural towns in Ningbo southern vogue areas, Ninghai counties and Xiangshan counties is far higher than that in northern rural towns. Under the influence of typhoon wind speed, in villages and towns with large reserve loss rate predicted values, houses of the villages and towns are easy to damage when typhoons come, and property and personnel life safety are easy to threaten. According to the typhoon disaster situation prediction result, before the typhoon comes, the emergency management department can timely and pertinently carry out the reinforcement and the first-aid repair of the farm house, and casualties and economic losses caused by collapse caused by wind are reduced as much as possible.

The feasibility and the superiority of the method proposed by the invention are verified by experiments below.

(1) Neural network model construction

According to the existing data samples, Matlab software is utilized to respectively establish rural housing station wind disaster situation estimation models based on a BP (back propagation) neural network and a RBF (radial Basis Function, RBF) neural network. Wherein the BP neural network is designed as a typical layer 3 network model. The data of the input layer is typhoon wind speed, the data of the output layer is the loss rate of the balance of each village and town, and the statistical data covers 139 villages and towns of Ningbo. And determining the number of neurons in the hidden layer to be 5, and adopting tangent sigmoid transfer functions of 'tansig' and 'purelin' as transfer functions respectively. The learning training function adopts a default function of 'rainlm' of the system, the training times are set to be 1000, the training target is 0.001, and other parameters are set to be default values.

The construction of the RBF neural network is simpler than that of the BP neural network. According to the existing data samples, a neural network with a radial basis function expansion coefficient of 1.2 is created through a newrb function, and a mean square error target is set to be 0.001.

(2) Neural network model training comparison

80% of sample data is selected for training in both models, and the remaining 20% of sample data is used for checking and calibrating. In order to make the BP neural network closer to the training target during training, the number of hidden layer nodes in the model, the training function and other relevant parameters need to be adjusted, and these often take a long time to debug. In contrast, in the training process of the RBF neural network, the number of hidden layer nodes can be automatically increased by setting, and whether the error of the network meets the error requirement or not is checked from 0 node until the error requirement or the maximum number of hidden layer nodes is met.

At a training accuracy of 0.001, the two model training curves are shown in fig. 7 and 8. As can be seen from the figure, both models converge under given conditions, but the BP neural network model requires 110 steps to meet the precision requirement. The RBF neural network model can meet the requirement only by 12 steps, and the training speed is obviously superior to that of a BP neural network. In addition, by running the two models, the training results of the RBF model are basically the same under the condition of parameter determination, while the training results of the BP model are not next stable and sometimes have larger deviation.

Furthermore, the predicted value and actual value pairs of the BP model and the RBF model are shown in FIG. 9, and the mean square error value is used as the evaluation index of the simulation situation of the two models, as shown in FIG. 10, it can be known that the maximum mean square error value of the result predicted by the BP neural network model is 18 × 10^-6Compared with the maximum absolute error of 8 x 10^-6The RBF neural network model has low precision and large error range fluctuation. Although the mean square errors of the two methods in the same town and country are high and low, the mean square error of the predicted value of the RBF is smaller than that of the BP neural network in general.

Therefore, the results show that the RBF neural network established under the same conditions has higher prediction precision, and the training time and the convergence speed are better than those of the BP neural network. Therefore, the method for constructing the typhoon disaster situation estimation model by using the RBF neural network has more advantages than the BP neural network.

In conclusion, the invention effectively establishes a typhoon disaster situation evaluation model from disaster factors to disaster-bearing body loss by using the RBF neural network, can realize rural house typhoon disaster situation estimation with high rural resolution by means of public disaster insurance data, solves the problems of low spatial resolution and difficult risk measurement in the traditional risk evaluation, accurately estimates the risk disaster of each town and village, quantificationally evaluates the typhoon disaster risk, provides scientific basis for the subsequent formulation of typhoon disaster insurance policies, typhoon emergency plans and other applications, and can effectively reduce casualties and economic loss caused by wind collapse.

While the invention has been described with reference to a preferred embodiment, it will be understood by those skilled in the art that various changes in form and detail may be made therein without departing from the spirit and scope of the invention.