Method, device, computer equipment and readable storage medium for determining automobile insurance premium

Abstract

The present application discloses a method, apparatus, computer device, and readable storage medium for determining auto insurance premiums, relating to the field of internet technology. By using a generative adversarial network to capture nonlinear features in historical driving behavior and generate more realistic sample data, the method then uses the sample data to train a subsequent premium determination model, enabling the model to determine a more reasonable premium strategy. The method comprises: obtaining historical claims data, reading historical driving behavior data from the historical claims data; constructing a sample generation model, inputting the historical driving behavior data into the sample generation model so that the sample generation model generates simulated driving behavior data; training an initial determination model using the historical driving behavior data and the simulated driving behavior data to obtain a target auto insurance premium determination model; and upon receiving driving behavior data to be predicted, inputting the driving behavior data to be predicted into the target auto insurance premium determination model to perform auto insurance premium prediction.

Description

Technical Field

[0001] The present application relates to the field of Internet technology, and in particular to a method, apparatus, computer equipment, and computer-readable storage medium for determining automobile insurance premiums. Background Art

[0002] With the continuous increase in car ownership and increasingly fierce competition in the auto insurance market, insurance companies are faced with the challenge of rationally determining auto insurance premiums to reduce risk and improve customer satisfaction. At the same time, the continuous development of internet technology and the widespread application of big data analysis, intelligent algorithms, and online service platforms are enabling insurance companies to obtain richer data on customer driving behavior and risk characteristics, providing strong technical support for determining auto insurance premiums.

[0003] In related technologies, insurance companies usually collect historical claims data through online service platforms, and then use big data analysis technology and intelligent algorithms to conduct statistical analysis on the historical claims data, treating drivers with similar characteristics as the same risk level, and thus calculating a unified premium.

[0004] In the process of implementing this application, the applicant discovered that the related technology has at least the following problems:

[0005] The traditional method of determining auto insurance premiums is to establish an auto insurance premium determination model through a large amount of labeled data. However, insurance data usually has complex nonlinear relationships. Traditional auto insurance premium determination models often find it difficult to capture the complex nonlinear relationship between driving behavior and accident risk, resulting in inaccurate results in auto insurance premium determination. Summary of the Invention

[0006] In view of this, the present application provides a method, apparatus, computer equipment and readable storage medium for determining automobile insurance premiums. The main purpose is to solve the problem that current insurance data usually has complex nonlinear relationships, and traditional automobile insurance premium determination models often find it difficult to capture the complex nonlinear relationship between driving behavior and accident risk, resulting in inaccurate results in automobile insurance premium determination.

[0007] According to a first aspect of the present application, a method for determining a motor vehicle insurance premium is provided, the method comprising:

[0008] Acquiring historical claims data, and reading historical driving behavior data from the historical claims data;

[0009] constructing a sample generation model, inputting the historical driving behavior data into the sample generation model, so that the sample generation model captures nonlinear driving characteristics in the historical driving behavior data based on a generative adversarial network, and generates simulated driving behavior data according to the nonlinear driving characteristics;

[0010] Building an initial determination model based on a deep reinforcement learning framework, and using the historical driving behavior data and the simulated driving behavior data to train the initial determination model to obtain a target auto insurance premium determination model;

[0011] When the driving behavior data to be predicted is received, the driving behavior data to be predicted is input into the target motor vehicle insurance premium determination model to perform motor vehicle insurance premium prediction to obtain a motor vehicle insurance premium prediction result.

[0012] Optionally, the constructing of the sample generation model, inputting the historical driving behavior data into the sample generation model so that the sample generation model captures nonlinear driving characteristics in the historical driving behavior data based on a generative adversarial network, and generates simulated driving behavior data according to the nonlinear driving characteristics, includes:

[0013] Constructing the sample generation model based on a generative adversarial network, and initializing model parameters of the sample generation model, the model parameters including generator parameters of a generator and discriminator parameters of a discriminator, wherein the generator and the discriminator are respectively constructed based on deep neural networks;

[0014] Continuously extracting a preset number of historical driving behavior data from the historical driving behavior data and merging them with random noise to generate data to be simulated;

[0015] The data to be simulated is input into the sample generation model, the generator and discriminator of the sample generation model are used to capture the nonlinear driving characteristics in the historical driving behavior data, and simulated driving behavior data is generated according to the nonlinear driving characteristics.

[0016] Optionally, inputting the to-be-simulated data into the sample generation model, using a generator and a discriminator of the sample generation model to capture nonlinear driving features in the historical driving behavior data, and generating simulated driving behavior data based on the nonlinear driving features includes:

[0017] Inputting the data to be simulated into the generator, receiving the data to be simulated through an input layer of the generator, passing the data to be simulated to multiple hidden layers of the generator, capturing the nonlinear driving characteristics through the multiple hidden layers, and generating simulation data according to the nonlinear driving characteristics, and outputting the simulation data as the simulated driving behavior data through an output layer of the generator, and passing the simulated driving behavior data to the discriminator;

[0018] The discriminator receives the data to be simulated and the simulated driving behavior data, labels the simulated driving behavior data and the data to be simulated, and obtains a labeling result. The labeling result is used to calculate the generator loss of the generator and the discriminator loss of the discriminator, so that the generator updates the model parameters of the generator based on the generator loss, and the discriminator updates the model parameters of the discriminator based on the discriminator loss, wherein the discrimination label is a discrimination result indicating whether the data is true or false.

[0019] Optionally, the capturing of the nonlinear driving characteristics through the multiple hidden layers, generating simulated data according to the nonlinear driving characteristics, outputting the simulated data as the simulated driving behavior data through the output layer of the generator, and passing the simulated driving behavior data to the discriminator includes:

[0020] Receiving the data to be simulated transmitted by the input layer through a first hidden layer, so that the first hidden layer captures the nonlinear characteristics in the data to be simulated based on a rectified linear unit (ReLU) function, and linearly transforms the data to be simulated using a weight matrix and a bias, introducing the nonlinear characteristics into the linear transformation result, and obtaining simulated data corresponding to the first hidden layer;

[0021] Receiving the analog data transmitted by the first hidden layer through the second hidden layer, so that the second hidden layer captures the nonlinear characteristics in the analog data based on the rectified linear unit (ReLU) function, and linearly transforming the analog data using a weight matrix and a bias, introducing the nonlinear characteristics into the linear transformation result, and obtaining analog data corresponding to the second hidden layer; receiving the analog data transmitted by the second hidden layer through the third hidden layer, determining the analog data corresponding to the third hidden layer, until reaching the last hidden layer, and obtaining analog data corresponding to each hidden layer, wherein the second hidden layer is the next layer after the first hidden layer;

[0022] The output layer of the generator receives the simulated data transmitted by the last hidden layer, maps the simulated data into the simulated driving behavior data and transmits it to the discriminator.

[0023] Optionally, the constructing of an initial determination model based on a deep reinforcement learning framework and the use of the historical driving behavior data and the simulated driving behavior data to perform model training on the initial determination model to obtain a target auto insurance premium determination model include:

[0024] Aggregating the historical driving behavior data and the simulated driving behavior data into a target data set, and continuously extracting a preset number of target data from the target data set to generate a target training set, wherein the age characteristics in the target data meet a preset age condition;

[0025] constructing an initial determination model based on a deep reinforcement learning framework, inputting target data from the target training set into the initial determination model, so that the initial determination model determines a driving state corresponding to the target data based on a Q network, and selects a specific auto insurance premium pricing strategy for the driving state in a preset action space, wherein the driving state is used to indicate driving characteristics in the target data, and the driving characteristics include age characteristics, driving record characteristics, vehicle type characteristics, and geographic location characteristics;

[0026] obtaining feedback information associated with the specified auto insurance premium pricing strategy from the target data based on the initial determination model, so that the initial determination model calculates a reward value corresponding to the feedback information based on the Q network, and updates a weight of the specified auto insurance premium pricing strategy corresponding to the driving state according to the reward value;

[0027] Target training data is continuously acquired to train the Q network, weights of different auto insurance premium pricing strategies under each driving state are updated, and a target auto insurance premium pricing strategy corresponding to each driving state is determined based on the weights to obtain the target auto insurance premium determination model.

[0028] Optionally, obtaining feedback information associated with the specified auto insurance premium pricing strategy from the target data based on the initial determination model, so that the initial determination model calculates a reward value corresponding to the feedback information based on the Q network, and updates a weight of the specified auto insurance premium pricing strategy corresponding to the driving state according to the reward value, includes:

[0029] Determining a reward function based on the initial determination model and obtaining feedback information associated with the specified auto insurance premium pricing strategy from the target data, so that the initial determination model uses the Q network, determines a reward type associated with the feedback information through the reward function, and obtains a reward value indicated by the reward type;

[0030] Based on the initial determination model, the weight of each motor vehicle insurance premium pricing strategy corresponding to the driving state is determined, the weight corresponding to the specified motor vehicle insurance premium pricing strategy is read, and the weight corresponding to the specified motor vehicle insurance premium pricing strategy is updated using the sum of the weight and the reward value.

[0031] Optionally, when receiving the driving behavior data to be predicted, inputting the driving behavior data to be predicted into the target vehicle insurance premium determination model to perform vehicle insurance premium prediction to obtain a vehicle insurance premium prediction result, including:

[0032] When receiving the driving behavior data to be predicted, inputting the driving behavior data to be predicted into the target vehicle insurance premium determination model;

[0033] Based on the target auto insurance premium determination model, the driving state corresponding to the driving behavior data to be predicted and the weights of different auto insurance premium pricing strategies corresponding to the driving state are determined, and the target auto insurance premium pricing strategy with a higher weight value than other auto insurance premium pricing strategies is used as the auto insurance premium prediction result.

[0034] According to a second aspect of the present application, a vehicle insurance premium determination device is provided, the device comprising:

[0035] An acquisition module, configured to acquire historical claims data and read historical driving behavior data from the historical claims data;

[0036] a generation module, configured to construct a sample generation model, input the historical driving behavior data into the sample generation model, so that the sample generation model captures nonlinear driving characteristics in the historical driving behavior data based on a generative adversarial network, and generates simulated driving behavior data based on the nonlinear driving characteristics;

[0037] A construction module is used to construct an initial determination model based on a deep reinforcement learning framework, and use the historical driving behavior data and the simulated driving behavior data to train the initial determination model to obtain a target auto insurance premium determination model;

[0038] The determination module is used to input the driving behavior data to be predicted into the target motor vehicle insurance premium determination model to perform motor vehicle insurance premium prediction when receiving the driving behavior data to be predicted, so as to obtain a motor vehicle insurance premium prediction result.

[0039] Optionally, the generation module is used to construct the sample generation model based on a generative adversarial network and initialize the model parameters of the sample generation model, wherein the model parameters include generator parameters of the generator and discriminator parameters of the discriminator, wherein the generator and the discriminator are respectively constructed based on deep neural networks; continuously extract a preset number of historical driving behavior data from the historical driving behavior data and merge them with random noise to generate data to be simulated; input the data to be simulated into the sample generation model, use the generator and discriminator of the sample generation model to capture the nonlinear driving characteristics in the historical driving behavior data, and generate simulated driving behavior data based on the nonlinear driving characteristics.

[0040] Optionally, the generation module is configured to input the data to be simulated into the generator, receive the data to be simulated through the input layer of the generator, pass the data to be simulated to multiple hidden layers of the generator, capture the nonlinear driving characteristics through the multiple hidden layers, generate simulated data based on the nonlinear driving characteristics, and output the simulated data as the simulated driving behavior data through the output layer of the generator, and pass the simulated driving behavior data to the discriminator; based on the discriminator receiving the data to be simulated and the simulated driving behavior data, label the simulated driving behavior data and the data to be simulated to obtain a labeling result, and the labeling result is used to calculate the generator loss of the generator and the discriminator loss of the discriminator, so that the generator updates the model parameters of the generator based on the generator loss, and the discriminator updates the model parameters of the discriminator based on the discriminator loss, wherein the discrimination label is a discrimination result indicating whether the data is true or false.

[0041] Optionally, the generation module is configured to receive the data to be simulated transmitted by the input layer through a first hidden layer, so that the first hidden layer captures the nonlinear features in the data to be simulated based on a rectified linear unit (ReLU) function, and uses a weight matrix and a bias to perform a linear transformation on the data to be simulated, introduce the nonlinear features into the linear transformation result, and obtain simulated data corresponding to the first hidden layer; receive the simulated data transmitted by the first hidden layer through a second hidden layer, so that the second hidden layer captures the nonlinear features in the simulated data based on a rectified linear unit (ReLU) function, and uses a weight matrix and a bias to perform a linear transformation on the simulated data, introduce the nonlinear features into the linear transformation result, and obtain simulated data corresponding to the second hidden layer; receive the simulated data transmitted by the second hidden layer through a third hidden layer, determine the simulated data corresponding to the third hidden layer, until reaching the last hidden layer, and obtain simulated data corresponding to each hidden layer, wherein the second hidden layer is the next layer of the first hidden layer; receive the simulated data transmitted by the last hidden layer through the output layer of the generator, map the simulated data into the simulated driving behavior data, and transmit it to the discriminator.

[0042] Optionally, the construction module is used to aggregate the historical driving behavior data and the simulated driving behavior data into a target data set, continuously extract a preset number of target data from the target data set to generate a target training set, and the age characteristics in the target data meet the preset age conditions; construct an initial determination model based on a deep reinforcement learning framework, input the target data in the target training set into the initial determination model, so that the initial determination model determines the driving state corresponding to the target data based on the Q network, and selects a specified auto insurance premium pricing strategy for the driving state in a preset action space, wherein the driving state is used to indicate the driving characteristics in the target data, and the The driving characteristics include age characteristics, driving record characteristics, vehicle type characteristics and geographic location characteristics; based on the initial determination model, feedback information associated with the specified auto insurance premium pricing strategy is obtained in the target data, so that the initial determination model calculates the reward value corresponding to the feedback information based on the Q network, and updates the weight of the specified auto insurance premium pricing strategy corresponding to the driving state according to the reward value; continuously obtains target training data to train the Q network, updates the weights of different auto insurance premium pricing strategies under each driving state, and determines the target auto insurance premium pricing strategy corresponding to each driving state according to the weight, thereby obtaining the target auto insurance premium determination model.

[0043] Optionally, the construction module is used to determine the reward function based on the initial determination model, and obtain feedback information associated with the specified auto insurance premium pricing strategy in the target data, so that the initial determination model adopts the Q network, determines the reward type associated with the feedback information through the reward function, and obtains the reward value indicated by the reward type; determines the weight of each auto insurance premium pricing strategy corresponding to the driving state based on the initial determination model, reads the weight corresponding to the specified auto insurance premium pricing strategy, and uses the sum of the weight and the reward value to update the weight corresponding to the specified auto insurance premium pricing strategy.

[0044] Optionally, the determination module is used to input the driving behavior data to be predicted into the target motor vehicle insurance premium determination model when receiving the driving behavior data to be predicted; determine the driving state corresponding to the driving behavior data to be predicted and the weights of different motor vehicle insurance premium pricing strategies corresponding to the driving state based on the target motor vehicle insurance premium determination model, and take the target motor vehicle insurance premium pricing strategy with a higher weight value than other motor vehicle insurance premium pricing strategies as the motor vehicle insurance premium prediction result.

[0045] According to a third aspect of the present application, a computer device is provided, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the methods described in the first aspect when executing the computer program.

[0046] According to a fourth aspect of the present application, a readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of any one of the methods in the first aspect are implemented.

[0047] By means of the above technical solution, the present application provides a method, apparatus, computer device and computer-readable storage medium for determining automobile insurance premiums. The present application first obtains historical claims data and reads historical driving behavior data from the historical claims data. Furthermore, a sample generation model is constructed, and the historical driving behavior data is input into the sample generation model, so that the sample generation model captures the nonlinear driving characteristics in the historical driving behavior data based on a generative adversarial network, and generates simulated driving behavior data based on the nonlinear driving characteristics. Next, an initial determination model is constructed based on a deep reinforcement learning framework, and the initial determination model is trained using historical driving behavior data and simulated driving behavior data to obtain a target automobile insurance premium determination model. Finally, when the driving behavior data to be predicted is received, the driving behavior data to be predicted is input into the target automobile insurance premium determination model to perform automobile insurance premium prediction, and obtain an automobile insurance premium prediction result. In this embodiment of the present application, by using a generative adversarial network (GAN) to capture the nonlinear characteristics of historical driving behavior, the model can generate richer and more diverse simulated driving behavior data. This rich driving behavior data is then used to train the subsequent premium determination model, enabling the model to more accurately assess the driver's risk level based on their actual performance (such as driving habits and risky behaviors), thereby formulating more reasonable premiums. In addition, customizing auto insurance premiums based on individual driving behavior can better match premiums with actual driving risks, thereby improving user satisfaction.

[0048] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order 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

[0049] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present application. The same reference symbols are used throughout the drawings to represent the same components. In the drawings:

[0050] Figure 1 A flow chart of a method for determining a motor vehicle insurance premium provided by an embodiment of the present application is shown;

[0051] Figure 2 A flow chart of a method for determining a motor vehicle insurance premium provided by an embodiment of the present application is shown;

[0052] Figure 3 A schematic structural diagram of a vehicle insurance premium determination device provided in an embodiment of the present application is shown;

[0053] Figure 4 A schematic diagram of the device structure of a computer device provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0054] The following describes exemplary embodiments of the present application in more detail with reference to the accompanying drawings. Although exemplary embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present application and to fully convey the scope of the present application to those skilled in the art.

[0055] The embodiment of the present application provides a method for determining automobile insurance premiums, such as Figure 1 As shown, the method includes:

[0056] 101. Obtain historical claims data and read historical driving behavior data from the historical claims data.

[0057] Claims record systems typically contain a large number of records of insurance company payouts due to drivers' driving behavior. These records, known as historical claims data, contain behavioral characteristics that can reflect drivers' driving habits and risk preferences. By identifying this historical claims data, we can derive driving behavior data such as driver age, gender, driving weather conditions, mileage, and driving history. This allows subsequent sample generation models to capture the nonlinear characteristics of this driving behavior data and generate more realistic simulated driving behavior data.

[0058] 102. Construct a sample generation model and input the historical driving behavior data into the sample generation model so that the sample generation model captures the nonlinear driving characteristics in the historical driving behavior data based on the generative adversarial network and generates simulated driving behavior data based on the nonlinear driving characteristics.

[0059] The Generative Adversarial Network (GAN) consists of two parts: a generator and a discriminator. The generator is responsible for generating simulated data, while the discriminator is responsible for distinguishing between the generated simulated data and real data. Through this adversarial training method, the generator gradually learns to generate data that is as realistic as possible. Furthermore, driving behavior data contains many nonlinear characteristics, such as how emergencies are handled and how driving habits change with age. The GAN model can capture these complex nonlinear characteristics and generate more realistic simulated driving behavior data, which helps enrich the training samples and improve the model's generalization ability.

[0060] 103. An initial determination model is constructed based on a deep reinforcement learning framework, and the historical driving behavior data and simulated driving behavior data are used to train the initial determination model to obtain a target auto insurance premium determination model.

[0061] Specifically, the deep reinforcement learning framework, DQL (Deep Q-Learning), first builds an initial deterministic model based on the deep reinforcement learning framework. This model is then trained using historical driving behavior data and generated simulated driving behavior data. Through continuous iteration and optimization, the model gradually learns how to accurately predict auto insurance premiums.

[0062] 104. When the driving behavior data to be predicted is received, the driving behavior data to be predicted is input into the target vehicle insurance premium determination model to perform vehicle insurance premium prediction and obtain a vehicle insurance premium prediction result.

[0063] Specifically, the driving behavior data to be predicted is input into the trained target auto insurance premium determination model. The target auto insurance premium determination model will output the corresponding auto insurance premium prediction results based on the learned driving behavior characteristics and premium relationship, thereby helping insurance companies to conduct risk assessment and pricing more accurately.

[0064] The method provided in the embodiment of the present application first obtains historical claims data and reads historical driving behavior data from the historical claims data. Furthermore, a sample generation model is constructed and the historical driving behavior data is input into the sample generation model so that the sample generation model captures the nonlinear driving characteristics in the historical driving behavior data based on a generative adversarial network and generates simulated driving behavior data based on the nonlinear driving characteristics. Next, an initial determination model is constructed based on a deep reinforcement learning framework, and the initial determination model is trained using the historical driving behavior data and the simulated driving behavior data to obtain a target auto insurance premium determination model. Finally, when the driving behavior data to be predicted is received, the driving behavior data to be predicted is input into the target auto insurance premium determination model to perform auto insurance premium prediction and obtain an auto insurance premium prediction result. In the embodiment of the present application, by using a generative adversarial network (GAN) to capture the nonlinear characteristics in historical driving behavior, the model can generate richer and more diverse simulated driving behavior data, and then use the rich driving behavior data to train the subsequent premium determination model, so that the model can more accurately assess the driver's risk level based on his or her actual performance (such as driving habits, risky behavior, etc.), thereby determining a more reasonable auto insurance premium strategy. In addition, customizing auto insurance premiums based on individual driving behavior can make premiums more closely aligned with actual driving risks, thereby improving user satisfaction.

[0065] The embodiment of the present application provides a method for determining automobile insurance premiums, such as Figure 2As shown, the method includes:

[0066] 201. Obtain historical claims data, and read historical driving behavior data from the historical claims data.

[0067] With the continuous increase in car ownership and increasingly fierce competition in the auto insurance market, insurance companies face the challenge of rationally determining auto insurance premiums to reduce risk and improve customer satisfaction. At the same time, with the continuous development of internet technology and the widespread application of big data analysis techniques, intelligent algorithms, and online service platforms, insurance companies are able to obtain richer data on customer driving behavior and risk characteristics, providing strong technical support for determining auto insurance premiums. Currently, insurance companies typically collect historical claims data through online service platforms and then use big data analysis techniques and intelligent algorithms to perform statistical analysis on this historical claims data, treating drivers with similar characteristics as having the same risk level and calculating a uniform premium. However, the applicant recognizes that traditional auto insurance premium determination methods rely on large amounts of labeled data to establish auto insurance premium determination models. However, insurance data often exhibits complex nonlinear relationships, and traditional auto insurance premium determination models often struggle to capture the complex nonlinear relationship between driving behavior and accident risk, resulting in inaccurate auto insurance premium determination results. Therefore, the present application provides a method for determining auto insurance premiums. The application first obtains historical claims data and then reads historical driving behavior data from the historical claims data. Furthermore, a sample generation model is constructed, and historical driving behavior data is input into the sample generation model. This allows the sample generation model to capture the nonlinear driving characteristics in the historical driving behavior data based on a generative adversarial network, and to generate simulated driving behavior data based on the nonlinear driving characteristics. Next, an initial determination model is constructed based on a deep reinforcement learning framework, and the initial determination model is trained using the historical driving behavior data and the simulated driving behavior data to obtain a target auto insurance premium determination model. Finally, when the driving behavior data to be predicted is received, the driving behavior data to be predicted is input into the target auto insurance premium determination model to perform auto insurance premium prediction and obtain an auto insurance premium prediction result. In this embodiment of the present application, by using a generative adversarial network (GAN) to capture the nonlinear characteristics in historical driving behavior, the model can generate richer and more diverse simulated driving behavior data. This rich driving behavior data is then used to train the subsequent premium determination model, enabling the model to more accurately assess the driver's risk level based on their actual performance (such as driving habits, risky behaviors, etc.), thereby determining a more reasonable auto insurance premium strategy. In addition, customizing auto insurance premiums based on individual driving behavior can make premiums more closely aligned with actual driving risks, thereby improving user satisfaction.

[0068] In the embodiment of the present application, it is first necessary to obtain a certain amount of historical claims data, and the historical claims data can be obtained from the database of the claims system during actual operation. The historical claims data in this step uses data from all motor insurance claims from 22 to 24 years, which is about 13 million. It should be noted that historical claims data usually contains a large number of records of drivers' driving behaviors that resulted in insurance company payments, and the historical claims data contains historical driving behavior data such as the driver's age, gender, driving weather, driving mileage and driving records. Reading historical driving behavior data in historical claims data can specifically be based on natural language processing technology to identify historical claims data, read historical driving behavior data therefrom, and then use these historical driving behavior data that reveal the driver's driving habits and risk preferences to generate realistic simulated driving behavior data, to provide support for the subsequent determination of motor insurance premiums.

[0069] 202. Construct a sample generation model based on the generative adversarial network and initialize the model parameters of the sample generation model.

[0070] In an embodiment of the present application, a sample generation model is constructed based on a generative adversarial network (GAN), and the model parameters of the sample generation model are initialized. The generative adversarial network (GAN) consists of two parts: a generator and a discriminator. The generator is responsible for generating simulated data, and the discriminator is responsible for distinguishing between the generated simulated data and real data. Through this adversarial training method, the generator can gradually learn to generate data that is as realistic as possible.

[0071] 203. The generator and discriminator of the sample generation model are used to capture the nonlinear driving characteristics in the historical driving behavior data, and simulated driving behavior data is generated based on the nonlinear driving characteristics.

[0072] In the embodiment of the present application, in order to increase the diversity of generated data, ensure that the generated data has sufficient randomness and unpredictability, and enable the generator to generate more realistic simulated data during the training process, the generator of the sample generation model needs to accept a random noise z and the basic characteristics of the historical driving behavior data (age, driving experience, vehicle type) as input to generate simulated driving behavior data. The specific process of generating simulated driving behavior data is as follows:

[0073] First, a preset number of historical driving behavior data are continuously extracted from the historical driving behavior data and merged with random noise z to generate data to be simulated, which provides support for increasing the diversity of the generated simulated driving behavior data and ensuring that the generated simulated driving behavior data has sufficient randomness and unpredictability.

[0074] Next, the data to be simulated is input into the generator, and the generator is used to generate simulated driving behavior data. The formula for generating simulated driving behavior by the generator is shown in the following formula 1:

[0075] Formula 1: G(z;θ g )

[0076] Where z is random noise, θ g is the generator parameter of the generator. The generator and discriminator in GAN usually adopt deep neural networks, and these ReLU algorithms are used to capture complex nonlinear features. ReLU can better reflect driving behavior under certain specific conditions (emergency situations). In the embodiment of the present application, the generator and discriminator can be regarded as neural networks with multiple hidden layers. During actual operation, the age of the driver can be set to two categories: greater than 30 years old and less than or equal to 30 years old. The nonlinear characteristics of driving behavior under 30 years old are captured by GAN. The nonlinear characteristics can be reaction speed, steering angle, etc. in an emergency situation. Specifically, the data to be simulated is first received through the input layer of the generator G, and the data to be simulated is passed to the multiple hidden layers of the generator, the nonlinear driving features are captured through the multiple hidden layers, and the simulation data is generated according to the nonlinear driving features, and the simulation data is output as simulated driving behavior data through the output layer of the generator, and the simulated driving behavior data is passed to the discriminator. The specific formula for extracting nonlinear features is shown in the following formula 2:

[0077] Formula 2: h l =ReLU(W l ·h l-1 +b l )

[0078] Among them, h l is the output of layer l, W l is the weight matrix of layer l, b lis the bias of the lth layer. Specifically, the data to be simulated transmitted by the input layer is received through the first hidden layer, so that the first hidden layer captures the nonlinear features in the data to be simulated based on the rectified linear unit ReLU function, and uses the weight matrix and bias to perform a linear transformation on the data to be simulated, introduces nonlinear features in the linear transformation result, and obtains the simulated data corresponding to the first hidden layer. The simulated data transmitted by the first hidden layer is received through the second hidden layer, so that the second hidden layer captures the nonlinear features in the simulated data based on the rectified linear unit ReLU function, and uses the weight matrix and bias to perform a linear transformation on the simulated data, introduces nonlinear features in the linear transformation result, and obtains the simulated data corresponding to the second hidden layer. The simulated data transmitted by the second hidden layer is received through the third hidden layer, and the simulated data corresponding to the third hidden layer is determined until the last hidden layer is reached, and the simulated data corresponding to each hidden layer is obtained, wherein the second hidden layer is the next layer of the first hidden layer. The simulated data transmitted by the last hidden layer is received through the output layer of the generator, and the simulated data is mapped to simulated driving behavior data and passed to the discriminator. For example, h l It is the extracted driving behavior of drivers under 30 years old on rainy days. The formula needs to strengthen the weight of the nonlinear feature of rainy days. If the ReLU output is less than or equal to 0, it means that on rainy days, the driver's behavior is more inclined to slow down. When generating simulated driving behavior data, simulated driving behavior data with a tendency to slow down on rainy days is generated for drivers under 30 years old and passed to the discriminator.

[0079] Finally, the discriminator receives the data to be simulated and the simulated driving behavior data, labels the simulated driving behavior data and the data to be simulated, and obtains the labeling results. The formulas for the discriminator to label the simulated driving behavior data and the data to be simulated are shown in the following formulas 3 and 4:

[0080] Formula 3: D(x;θ) d )

[0081] Formula 4: D(G(z);θ d )

[0082] Among them, θ dis the discriminator parameter of the discriminator, x is the historical driving behavior data, G(z) is the simulated driving behavior data, and the discriminator will mark the historical driving behavior data as 1 and the simulated driving behavior data as 0. In the actual operation process, the generator and the discriminator are trained in an adversarial manner so that the simulated data generated by the generator are as close to the real data as possible, while the discriminator is required to distinguish between the real data and the simulated data as much as possible. In the embodiment of the present application, the labeling results are used to calculate the generator loss of the generator and the discriminator loss of the discriminator, so that the generator updates the model parameters of the generator based on the generator loss, and the discriminator updates the model parameters of the discriminator based on the discriminator loss. The adversarial loss function is defined as shown in the following formula 4:

[0083] Formula 4:

[0084] Among them, pdata(x) is the distribution of real data, p z (z) is the distribution of noise. Wherein, the discrimination mark is the discrimination result indicating whether the data is true or false.

[0085] 204. An initial determination model is constructed based on a deep reinforcement learning framework. The initial determination model is trained using historical driving behavior data and simulated driving behavior data to obtain a target auto insurance premium determination model.

[0086] In an embodiment of the present application, DQL (Deep Q-Learning) in deep reinforcement learning is combined to further optimize the pricing strategy to dynamically adapt to different driving behaviors and risk levels. Through continuous iteration and optimization, the model can gradually learn how to accurately predict auto insurance premiums. In this step, the goal of reinforcement learning is to find the optimal strategy through interactive learning to maximize long-term benefits. In this scenario, after obtaining age, driving behavior and other relevant factors, the insurance company's benefits can be maximized while reducing risks. Specifically, first, historical driving behavior data and simulated driving behavior data are aggregated into a target data set, and a preset number of target data are continuously extracted from the target data set to generate a target training set. The age feature in the target data meets the preset age condition, for example, greater than 30 years old. Subsequently, an initial determination model is constructed based on the deep reinforcement learning framework, and the state space and action space of the initial determination model are set. The state space S includes driving age, driving record, driving vehicle type and driving geographical location, where the driving record includes accident frequency and speeding record. Action space A includes multiple insurance pricing strategies, for example, Plan A: Base Premium of 2,000 yuan, Plan B: Base Premium of 2,500 yuan (including a high-risk surcharge), and Plan C: Base Premium of 1,800 yuan (for low-risk drivers). The target data from the target training set is then input into the initial determination model. The initial determination model determines the driving state corresponding to the target data based on the state space of the Q network and selects a specific auto insurance premium pricing strategy for the driving state in the preset action space. The driving state represents the driving characteristics in the target data, including age, driving record, vehicle type, and location. Next, a reward function R(s,a) is designed. This reward function R(s,a) can be tailored to the insurance company's objectives. For example, positive rewards may be given for successfully attracting low-risk customers (such as those with a clean record), while negative rewards may be given when the pricing strategy attracts high-risk customers or increases claims expenses. Furthermore, feedback information associated with the specified auto insurance premium pricing strategy is obtained from the target data based on the initial determination model, so that the initial determination model calculates a reward value corresponding to the feedback information based on the Q network, and updates the weight of the specified auto insurance premium pricing strategy corresponding to the driving state according to the reward value. Then, the weight of each auto insurance premium pricing strategy corresponding to the driving state is determined based on the initial determination model, the weight corresponding to the specified auto insurance premium pricing strategy is read, and the weight corresponding to the specified auto insurance premium pricing strategy is updated using the sum of the weight and the reward value. The specific formula for optimizing the specified auto insurance premium pricing strategy using the Q value update formula is shown in the following formula 5:

[0087] Formula 5:

[0088] Where s represents the current driving state, including characteristics such as age and driving behavior; s′ represents the driving state at the previous time step; a represents the current pricing strategy; a′ represents the pricing strategy at the previous time step; and R(s, a) represents the immediate reward obtained based on the current pricing strategy and customer type. γ represents a discount factor that weighs the importance of future returns. Finally, the Q network is trained by continuously acquiring target training data, updating the weights of different auto insurance premium pricing strategies under each driving state. Based on the weights, the target auto insurance premium pricing strategy corresponding to each driving state is determined, resulting in a target auto insurance premium determination model.

[0089] For example, the state space SS includes the following features: driver age 30, driving behavior characteristics including two accidents, five speeding violations, and mileage 12,000 kilometers in the past year. The pricing strategies for the action space AA include: Plan A: base premium 2,000 yuan, Plan B: base premium 2,500 yuan (including a high-risk surcharge), and Plan C: base premium 1,800 yuan (for low-risk drivers). The reward function is designed to reward successful low-risk drivers with +1,000 yuan and -1,500 yuan for claims resulting from high-risk customers (e.g., claims resulting from accidents). A Q-network is created, with the Q-values ​​of all state-action pairs initialized to 0. Behavioral data on 30-year-old drivers is then collected from historical driving data to form a training set. Next, at each time step, the state SS is input and pricing strategy AA is selected. Customer feedback is then collected, and the reward R(s,a) is calculated. The Q-value is then updated according to the Q-value update formula. Through multiple iterations of training, the pricing strategy for 30-year-old drivers is continuously optimized. After multiple iterations, suppose the system finds that pricing plan A excels in attracting low-risk drivers, earning an average reward of +1,200 yuan. Pricing plan B, with its higher risk compensation, results in more negative rewards, with an average reward of -300 yuan. Pricing plan C performs well in some low-risk situations, earning an average reward of +800 yuan. Based on the DQL training results, the model recommends pricing plan B for 30-year-old drivers with more than three speeding offenses. It recommends pricing plan A for drivers with fewer than two accidents and fewer than three speeding offenses. For all other drivers, the model recommends pricing plan C.

[0090] 205. When the driving behavior data to be predicted is received, the driving behavior data to be predicted is input into the target vehicle insurance premium determination model to perform vehicle insurance premium prediction and obtain a vehicle insurance premium prediction result.

[0091] In an embodiment of the present application, when the driving behavior data to be predicted is received, the received data is first preprocessed to ensure the integrity and accuracy of the data, for example, outliers are removed and missing values ​​are filled. Further, the driving behavior data to be predicted is input into the target auto insurance premium determination model, and the driving state corresponding to the driving behavior data to be predicted and the weights of different auto insurance premium pricing strategies corresponding to the driving state are determined based on the target auto insurance premium determination model. The system compares the weights of different auto insurance premium pricing strategies and identifies the main strategy with a weight value higher than other strategies. Finally, the auto insurance premium pricing strategy with the highest weight value is selected as the final auto insurance premium prediction result. By introducing generative adversarial networks (GANs) and reinforcement learning, the embodiment of the present application can capture the nonlinear characteristics in real driving behavior data, thereby generating more realistic simulated driving behavior data, and effectively solving the problem of data sparsity. Then, the real driving behavior data and the generated realistic simulated driving behavior data are used to train the auto insurance premium determination model, so that the auto insurance premium training model can realize personalized selection of premium strategies, effectively reducing the risk of insurance companies and encouraging drivers to improve their driving habits.

[0092] The method provided in the embodiment of the present application first obtains historical claims data and reads historical driving behavior data from the historical claims data. Furthermore, a sample generation model is constructed, and the historical driving behavior data is input into the sample generation model so that the sample generation model captures the nonlinear driving characteristics in the historical driving behavior data based on a generative adversarial network, and generates simulated driving behavior data based on the nonlinear driving characteristics. Next, an initial determination model is constructed based on a deep reinforcement learning framework, and the initial determination model is trained using the historical driving behavior data and the simulated driving behavior data to obtain a target auto insurance premium determination model. Finally, when the driving behavior data to be predicted is received, the driving behavior data to be predicted is input into the target auto insurance premium determination model to perform auto insurance premium prediction and obtain an auto insurance premium prediction result. In the embodiment of the present application, by using a generative adversarial network (GAN) to capture the nonlinear characteristics in historical driving behavior, the model can generate richer and more diverse simulated driving behavior data, and then use the rich driving behavior data to train the subsequent premium determination model, so that the model can more accurately assess the driver's risk level based on his or her actual performance (such as driving habits, risky behavior, etc.), thereby formulating a more reasonable premium. In addition, customizing auto insurance premiums based on individual driving behavior can make premiums more closely aligned with actual driving risks, thereby improving user satisfaction.

[0093] Further, as Figure 1 In a specific implementation of the method, the present application provides a device for determining automobile insurance premiums, such as Figure 3As shown, the device includes: an acquisition module 301, a generation module 302, a construction module 303, and a determination module 304.

[0094] The acquisition module 301 is used to acquire historical claim settlement data and read historical driving behavior data from the historical claim settlement data;

[0095] The generation module 302 is configured to construct a sample generation model, input the historical driving behavior data into the sample generation model, so that the sample generation model captures nonlinear driving characteristics in the historical driving behavior data based on a generative adversarial network, and generates simulated driving behavior data based on the nonlinear driving characteristics;

[0096] The construction module 303 is used to construct an initial determination model based on a deep reinforcement learning framework, and use the historical driving behavior data and the simulated driving behavior data to perform model training on the initial determination model to obtain a target auto insurance premium determination model;

[0097] The determination module 304 is configured to, upon receiving the driving behavior data to be predicted, input the driving behavior data to be predicted into the target vehicle insurance premium determination model to perform vehicle insurance premium prediction and obtain a vehicle insurance premium prediction result.

[0098] In a specific application scenario, the generation module 302 is used to construct the sample generation model based on a generative adversarial network and initialize the model parameters of the sample generation model, wherein the model parameters include the generator parameters of the generator and the discriminator parameters of the discriminator, wherein the generator and the discriminator are respectively constructed based on deep neural networks; continuously extract a preset number of historical driving behavior data from the historical driving behavior data and merge them with random noise to generate data to be simulated; input the data to be simulated into the sample generation model, use the generator and discriminator of the sample generation model to capture the nonlinear driving characteristics in the historical driving behavior data, and generate simulated driving behavior data based on the nonlinear driving characteristics.

[0099] In a specific application scenario, the generation module 302 is configured to input the data to be simulated into the generator, receive the data to be simulated through the input layer of the generator, pass the data to be simulated to multiple hidden layers of the generator, capture the nonlinear driving characteristics through the multiple hidden layers, generate simulated data based on the nonlinear driving characteristics, and output the simulated data as the simulated driving behavior data through the output layer of the generator, and pass the simulated driving behavior data to the discriminator; based on the discriminator receiving the data to be simulated and the simulated driving behavior data, label the simulated driving behavior data and the data to be simulated to obtain a labeling result, and the labeling result is used to calculate the generator loss of the generator and the discriminator loss of the discriminator, so that the generator updates the model parameters of the generator based on the generator loss, and the discriminator updates the model parameters of the discriminator based on the discriminator loss, wherein the discrimination label is a discrimination result indicating whether the data is true or false.

[0100] In a specific application scenario, the generation module 302 is configured to receive the data to be simulated transmitted by the input layer through a first hidden layer, so that the first hidden layer captures the nonlinear features in the data to be simulated based on a rectified linear unit (ReLU) function, and uses a weight matrix and a bias to perform a linear transformation on the data to be simulated, introduce the nonlinear features into the linear transformation result, and obtain simulated data corresponding to the first hidden layer; receive the simulated data transmitted by the first hidden layer through a second hidden layer, so that the second hidden layer captures the nonlinear features in the simulated data based on a rectified linear unit (ReLU) function, and uses a weight matrix and a bias to perform a linear transformation on the simulated data, introduce the nonlinear features into the linear transformation result, and obtain simulated data corresponding to the second hidden layer; receive the simulated data transmitted by the second hidden layer through a third hidden layer, determine the simulated data corresponding to the third hidden layer, until reaching the last hidden layer, and obtain simulated data corresponding to each hidden layer, wherein the second hidden layer is the next layer of the first hidden layer; receive the simulated data transmitted by the last hidden layer through the output layer of the generator, map the simulated data into the simulated driving behavior data, and transmit it to the discriminator.

[0101] In a specific application scenario, the construction module 303 is used to aggregate the historical driving behavior data and the simulated driving behavior data into a target data set, continuously extract a preset number of target data from the target data set to generate a target training set, and the age characteristics in the target data meet the preset age conditions; construct an initial determination model based on a deep reinforcement learning framework, input the target data in the target training set into the initial determination model, so that the initial determination model determines the driving state corresponding to the target data based on the Q network, and selects a specified auto insurance premium pricing strategy for the driving state in the preset action space, wherein the driving state is used to indicate the driving state in the target data. characteristics, wherein the driving characteristics include age characteristics, driving record characteristics, vehicle type characteristics and geographic location characteristics; based on the initial determination model, feedback information associated with the specified auto insurance premium pricing strategy is obtained in the target data, so that the initial determination model calculates a reward value corresponding to the feedback information based on the Q network, and updates the weight of the specified auto insurance premium pricing strategy corresponding to the driving state according to the reward value; continuously obtains target training data to train the Q network, updates the weights of different auto insurance premium pricing strategies under each driving state, and determines the target auto insurance premium pricing strategy corresponding to each driving state according to the weight, thereby obtaining the target auto insurance premium determination model.

[0102] In a specific application scenario, the construction module 303 is used to determine the reward function based on the initial determination model, and obtain feedback information associated with the specified auto insurance premium pricing strategy in the target data, so that the initial determination model adopts the Q network, determines the reward type associated with the feedback information through the reward function, and obtains the reward value indicated by the reward type; determines the weight of each auto insurance premium pricing strategy corresponding to the driving state based on the initial determination model, reads the weight corresponding to the specified auto insurance premium pricing strategy, and uses the sum of the weight and the reward value to update the weight corresponding to the specified auto insurance premium pricing strategy.

[0103] In a specific application scenario, the determination module 304 is used to input the driving behavior data to be predicted into the target motor vehicle insurance premium determination model when receiving the driving behavior data to be predicted; determine the driving state corresponding to the driving behavior data to be predicted and the weights of different motor vehicle insurance premium pricing strategies corresponding to the driving state based on the target motor vehicle insurance premium determination model, and use the target motor vehicle insurance premium pricing strategy with a weight value higher than other motor vehicle insurance premium pricing strategies as the motor vehicle insurance premium prediction result.

[0104] The device provided in the embodiment of the present application first obtains historical claims data and reads historical driving behavior data from the historical claims data. Furthermore, a sample generation model is constructed, and the historical driving behavior data is input into the sample generation model so that the sample generation model captures the nonlinear driving characteristics in the historical driving behavior data based on a generative adversarial network, and generates simulated driving behavior data based on the nonlinear driving characteristics. Next, an initial determination model is constructed based on a deep reinforcement learning framework, and the initial determination model is trained using the historical driving behavior data and the simulated driving behavior data to obtain a target auto insurance premium determination model. Finally, when the driving behavior data to be predicted is received, the driving behavior data to be predicted is input into the target auto insurance premium determination model to perform auto insurance premium prediction and obtain an auto insurance premium prediction result. In the embodiment of the present application, by using a generative adversarial network (GAN) to capture the nonlinear characteristics in historical driving behavior, the model can generate richer and more diverse simulated driving behavior data, and then use the rich driving behavior data to train the subsequent premium determination model, so that the model can more accurately assess the driver's risk level based on his or her actual performance (such as driving habits, risky behavior, etc.), thereby formulating a more reasonable premium. In addition, customizing auto insurance premiums based on individual driving behavior can make premiums more closely aligned with actual driving risks, thereby improving user satisfaction.

[0105] It should be noted that for other corresponding descriptions of the functional units involved in the vehicle insurance premium determination device provided in the embodiment of the present application, reference can be made to Figure 1 and Figure 2 The corresponding description in will not be repeated here.

[0106] In an exemplary embodiment, see Figure 4 A device is also provided, comprising a communication bus, a processor, a memory, and a communication interface. The device may also include an input / output interface and a display device, wherein the various functional units can communicate with each other via the bus. The memory stores a computer program, and the processor is configured to execute the program stored in the memory and perform the method for determining automobile insurance premiums in the above-described embodiment.

[0107] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method for determining automobile insurance premiums.

[0108] Through the description of the above implementation methods, those skilled in the art can clearly understand that the present application can be implemented through hardware or by means of software plus a necessary general hardware platform. Based on this understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.), including a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each implementation scenario of the present application.

[0109] Those skilled in the art will understand that the accompanying drawings are only schematic diagrams of a preferred implementation scenario, and the modules or processes in the accompanying drawings are not necessarily required to implement the present application.

[0110] Those skilled in the art will appreciate that the modules in the devices in the implementation scenario can be distributed in the devices of the implementation scenario according to the implementation scenario description, or can be modified accordingly and located in one or more devices different from the implementation scenario. The modules in the above implementation scenario can be combined into one module or further split into multiple submodules.

[0111] The above application serial numbers are for description only and do not represent the advantages or disadvantages of the implementation scenarios.

[0112] The above disclosure only describes several specific implementation scenarios of the present application. However, the present application is not limited thereto, and any changes that can be conceived by those skilled in the art should fall within the scope of protection of the present application.

Claims

1. A method for determining automobile insurance premiums, characterized in that: include: Acquiring historical claims data, and reading historical driving behavior data from the historical claims data; constructing a sample generation model, inputting the historical driving behavior data into the sample generation model, so that the sample generation model captures nonlinear driving characteristics in the historical driving behavior data based on a generative adversarial network, and generates simulated driving behavior data according to the nonlinear driving characteristics; Building an initial determination model based on a deep reinforcement learning framework, and using the historical driving behavior data and the simulated driving behavior data to train the initial determination model to obtain a target auto insurance premium determination model; When the driving behavior data to be predicted is received, the driving behavior data to be predicted is input into the target vehicle insurance premium determination model to perform vehicle insurance premium prediction to obtain a vehicle insurance premium prediction result; The step of constructing a sample generation model and inputting the historical driving behavior data into the sample generation model so that the sample generation model captures nonlinear driving characteristics in the historical driving behavior data based on a generative adversarial network and generates simulated driving behavior data based on the nonlinear driving characteristics includes: Constructing the sample generation model based on a generative adversarial network, and initializing model parameters of the sample generation model, the model parameters including generator parameters of a generator and discriminator parameters of a discriminator, wherein the generator and the discriminator are respectively constructed based on deep neural networks; Continuously extracting a preset number of historical driving behavior data from the historical driving behavior data and merging them with random noise to generate data to be simulated; Inputting the data to be simulated into the sample generation model, using a generator and a discriminator of the sample generation model to capture nonlinear driving features in the historical driving behavior data, and generating simulated driving behavior data based on the nonlinear driving features; The method of constructing an initial determination model based on a deep reinforcement learning framework and training the initial determination model using the historical driving behavior data and the simulated driving behavior data to obtain a target auto insurance premium determination model includes: Aggregating the historical driving behavior data and the simulated driving behavior data into a target data set, and continuously extracting a preset number of target data from the target data set to generate a target training set, wherein the age characteristics in the target data meet a preset age condition; constructing an initial determination model based on a deep reinforcement learning framework, inputting target data from the target training set into the initial determination model, so that the initial determination model determines a driving state corresponding to the target data based on a Q network, and selects a specific auto insurance premium pricing strategy for the driving state in a preset action space, wherein the driving state is used to indicate driving characteristics in the target data, and the driving characteristics include age characteristics, driving record characteristics, vehicle type characteristics, and geographic location characteristics; obtaining feedback information associated with the specified auto insurance premium pricing strategy from the target data based on the initial determination model, so that the initial determination model calculates a reward value corresponding to the feedback information based on the Q network, and updates a weight of the specified auto insurance premium pricing strategy corresponding to the driving state according to the reward value; Target training data is continuously acquired to train the Q network, weights of different auto insurance premium pricing strategies under each driving state are updated, and a target auto insurance premium pricing strategy corresponding to each driving state is determined based on the weights to obtain the target auto insurance premium determination model.

2. The method according to claim 1, characterized in that The step of inputting the data to be simulated into the sample generation model, using a generator and a discriminator of the sample generation model to capture nonlinear driving features in the historical driving behavior data, and generating simulated driving behavior data based on the nonlinear driving features includes: Inputting the data to be simulated into the generator, receiving the data to be simulated through an input layer of the generator, passing the data to be simulated to multiple hidden layers of the generator, capturing the nonlinear driving characteristics through the multiple hidden layers, and generating simulation data according to the nonlinear driving characteristics, and outputting the simulation data as the simulated driving behavior data through an output layer of the generator, and passing the simulated driving behavior data to the discriminator; The discriminator receives the data to be simulated and the simulated driving behavior data, labels the simulated driving behavior data and the data to be simulated, and obtains a labeling result. The labeling result is used to calculate the generator loss of the generator and the discriminator loss of the discriminator, so that the generator updates the model parameters of the generator based on the generator loss, and the discriminator updates the model parameters of the discriminator based on the discriminator loss, wherein the discrimination label is a discrimination result indicating whether the data is true or false.

3. The method according to claim 2, characterized in that The method of capturing the nonlinear driving characteristics through the multiple hidden layers, generating simulation data according to the nonlinear driving characteristics, outputting the simulation data as the simulated driving behavior data through the output layer of the generator, and transmitting the simulated driving behavior data to the discriminator includes: Receiving the data to be simulated transmitted by the input layer through a first hidden layer, so that the first hidden layer captures the nonlinear characteristics in the data to be simulated based on a rectified linear unit (ReLU) function, and linearly transforms the data to be simulated using a weight matrix and a bias, introducing the nonlinear characteristics into the linear transformation result, and obtaining simulated data corresponding to the first hidden layer; Receiving the analog data transmitted by the first hidden layer through the second hidden layer, so that the second hidden layer captures the nonlinear characteristics in the analog data based on the rectified linear unit (ReLU) function, and linearly transforming the analog data using a weight matrix and a bias, introducing the nonlinear characteristics into the linear transformation result, and obtaining analog data corresponding to the second hidden layer; receiving the analog data transmitted by the second hidden layer through the third hidden layer, determining the analog data corresponding to the third hidden layer, until reaching the last hidden layer, and obtaining analog data corresponding to each hidden layer, wherein the second hidden layer is the next layer after the first hidden layer; The output layer of the generator receives the simulated data transmitted by the last hidden layer, maps the simulated data into the simulated driving behavior data and transmits it to the discriminator.

4. The method according to claim 1, wherein The step of obtaining feedback information associated with the specified auto insurance premium pricing strategy from the target data based on the initial determination model, so that the initial determination model calculates a reward value corresponding to the feedback information based on the Q network, and updates a weight of the specified auto insurance premium pricing strategy corresponding to the driving state according to the reward value, includes: Determining a reward function based on the initial determination model and obtaining feedback information associated with the specified auto insurance premium pricing strategy from the target data, so that the initial determination model uses the Q network, determines a reward type associated with the feedback information through the reward function, and obtains a reward value indicated by the reward type; Based on the initial determination model, the weight of each motor vehicle insurance premium pricing strategy corresponding to the driving state is determined, the weight corresponding to the specified motor vehicle insurance premium pricing strategy is read, and the weight corresponding to the specified motor vehicle insurance premium pricing strategy is updated using the sum of the weight and the reward value.

5. The method according to claim 1, wherein When the driving behavior data to be predicted is received, the driving behavior data to be predicted is input into the target vehicle insurance premium determination model to perform vehicle insurance premium prediction to obtain a vehicle insurance premium prediction result, including: When receiving the driving behavior data to be predicted, inputting the driving behavior data to be predicted into the target vehicle insurance premium determination model; Based on the target auto insurance premium determination model, the driving state corresponding to the driving behavior data to be predicted and the weights of different auto insurance premium pricing strategies corresponding to the driving state are determined, and the target auto insurance premium pricing strategy with a higher weight value than other auto insurance premium pricing strategies is used as the auto insurance premium prediction result.

6. A device for determining automobile insurance premium, characterized in that: include: An acquisition module, configured to acquire historical claims data and read historical driving behavior data from the historical claims data; a generation module, configured to construct a sample generation model, input the historical driving behavior data into the sample generation model, so that the sample generation model captures nonlinear driving characteristics in the historical driving behavior data based on a generative adversarial network, and generates simulated driving behavior data based on the nonlinear driving characteristics; A construction module is used to construct an initial determination model based on a deep reinforcement learning framework, and use the historical driving behavior data and the simulated driving behavior data to train the initial determination model to obtain a target auto insurance premium determination model; a determination module, configured to, upon receiving the driving behavior data to be predicted, input the driving behavior data to be predicted into the target vehicle insurance premium determination model to perform vehicle insurance premium prediction and obtain a vehicle insurance premium prediction result; Wherein, the generating module is further used for: Constructing the sample generation model based on a generative adversarial network, and initializing model parameters of the sample generation model, the model parameters including generator parameters of a generator and discriminator parameters of a discriminator, wherein the generator and the discriminator are respectively constructed based on deep neural networks; Continuously extracting a preset number of historical driving behavior data from the historical driving behavior data and merging them with random noise to generate data to be simulated; Inputting the data to be simulated into the sample generation model, using a generator and a discriminator of the sample generation model to capture nonlinear driving features in the historical driving behavior data, and generating simulated driving behavior data based on the nonlinear driving features; Wherein, the building block is further used for: Aggregating the historical driving behavior data and the simulated driving behavior data into a target data set, and continuously extracting a preset number of target data from the target data set to generate a target training set, wherein the age characteristics in the target data meet a preset age condition; constructing an initial determination model based on a deep reinforcement learning framework, inputting target data from the target training set into the initial determination model, so that the initial determination model determines a driving state corresponding to the target data based on a Q network, and selects a specific auto insurance premium pricing strategy for the driving state in a preset action space, wherein the driving state is used to indicate driving characteristics in the target data, and the driving characteristics include age characteristics, driving record characteristics, vehicle type characteristics, and geographic location characteristics; obtaining feedback information associated with the specified auto insurance premium pricing strategy from the target data based on the initial determination model, so that the initial determination model calculates a reward value corresponding to the feedback information based on the Q network, and updates a weight of the specified auto insurance premium pricing strategy corresponding to the driving state according to the reward value; Target training data is continuously acquired to train the Q network, weights of different auto insurance premium pricing strategies under each driving state are updated, and a target auto insurance premium pricing strategy corresponding to each driving state is determined based on the weights to obtain the target auto insurance premium determination model.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 5 are implemented.

8. A readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images