Premium pricing method and device based on random deep network model, and medium

Abstract

The invention relates to the technical field of intelligent decision making, and provides an insurance premium pricing method, device and equipment based on a random deep network model, and a computer readable storage medium. On the basis of the random depth mechanism, part of network layers can be randomly skipped in the training process, the calculation amount of forward and back propagation during model training is reduced, and therefore the calculation complexity during model training is reduced; and only a part of network layers are used in each training, so that the training time is remarkably shortened, the convergence of the model is ensured, the model is prevented from being over-fitted, and the robustness of the model to abnormal data and noise data is enhanced. In addition, a pricing model is obtained based on random depth mechanism training. On the premise of ensuring the model performance, the model training efficiency and the pricing accuracy can be further improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent decision-making technology, and in particular to a premium pricing method, device, computer equipment and computer-readable storage medium based on a random deep network model. Background Art

[0002] Insurance products have always relied on traditional actuarial models and data analysis techniques to conduct risk assessment and premium calculations. However, with the rapid growth of insurance product data and the increasing complexity of data structures, the computational complexity of traditional actuarial models has increased dramatically, resulting in huge computing time and resource consumption. In addition, insurance product data is high-dimensional and usually includes a large number of data variables and data features, which makes it easy for traditional models to overfit or underfit when processing insurance product data. As a result, the current traditional models have low pricing efficiency and accuracy for insurance products.

[0003] Therefore, how to improve the pricing efficiency and accuracy of the current premium pricing model has become a technical problem that needs to be solved urgently. Summary of the invention

[0004] The main purpose of the present invention is to provide a premium pricing method, device, equipment and computer-readable storage medium based on a random deep network model, aiming to improve the pricing efficiency and accuracy of the current premium pricing model.

[0005] To achieve the above object, the present invention provides a premium pricing method based on a random deep network model, the premium pricing method comprising the following steps:

[0006] Initialize the weight parameters and bias parameters of each network layer in the random deep network model, and generate the survival probability parameters of each network layer based on preset rules;

[0007] Based on the survival probability parameters of each network layer, a target network layer and an optimized network layer are determined in each network layer, wherein the target network layer is a network layer to be trained, and the optimized network layer is a network layer for which training needs to be skipped;

[0008] Based on a preset risk assessment sample set, an initial weight parameter, and an initial bias parameter, the optimization network layer is skipped, and the target network layer is trained until a preset convergence condition is reached, thereby obtaining a trained random deep network model as a target random deep network model;

[0009] Based on the target random deep network model, a risk assessment is performed on the target user to obtain the risk level of the target user, and based on a preset pricing strategy and the risk level, the premium of the target user is calculated.

[0010] In addition, to achieve the above-mentioned purpose, the present invention also provides a premium pricing device based on a random deep network model, and the premium pricing device based on a random deep network model comprises:

[0011] The random parameter generation module is used to initialize the weight parameters and bias parameters of each network layer in the random deep network model, and generate the survival probability parameters of each network layer based on preset rules;

[0012] A network layer classification module, used to determine a target network layer and an optimized network layer in each network layer based on the survival probability parameters of each network layer, wherein the target network layer is a network layer to be trained, and the optimized network layer is a network layer for which training needs to be skipped;

[0013] A random model training module is used to skip the optimization network layer based on a preset risk assessment sample set, an initial weight parameter, and an initial bias parameter, and train the target network layer until a preset convergence condition is reached, thereby obtaining a trained random deep network model as a target random deep network model;

[0014] The user premium pricing module is used to perform risk assessment on the target user based on the target random deep network model, obtain the risk level of the target user, and calculate the premium of the target user based on a preset pricing strategy and the risk level.

[0015] In addition, to achieve the above-mentioned purpose, the present invention also provides a computer device, which includes a processor, a memory, and a premium pricing program based on a random deep network model stored in the memory and executable by the processor, wherein when the premium pricing program based on the random deep network model is executed by the processor, the steps of the premium pricing method based on the random deep network model as described above are implemented.

[0016] In addition, to achieve the above-mentioned purpose, the present invention also provides a computer-readable storage medium, on which a premium pricing program based on a random deep network model is stored, wherein when the premium pricing program based on the random deep network model is executed by a processor, the steps of the premium pricing method based on the random deep network model as described above are implemented.

[0017] The present application provides a premium pricing method based on a random deep network model. The method initializes weight parameters and bias parameters of each network layer in the random deep network model, and generates survival probability parameters of each network layer based on preset rules; based on the survival probability parameters of each network layer, determines a target network layer and an optimized network layer in each network layer, wherein the target network layer is a network layer to be trained, and the optimized network layer is a network layer for which training needs to be skipped; based on a preset risk assessment sample set, initial weight parameters, and initial bias parameters, skips the optimized network layer, and trains the target network layer until a preset convergence condition is reached, thereby obtaining a trained random deep network model as a target random deep network model; based on the target random deep network model, performs risk assessment on a target user to obtain the risk level of the target user, and calculates the premium of the target user based on a preset pricing strategy and the risk level. Through the above method, the present application provides a random depth mechanism based on the survival probability corresponding to each network layer. Based on the random depth mechanism, some network layers can be randomly skipped during the training process, reducing the amount of forward and backward propagation calculations during model training, thereby reducing the computational complexity of model training; and only part of the network layer is used in each training, so that the training time is significantly shortened while ensuring the convergence of the model, preventing the model from overfitting, and enhancing the robustness of the model to abnormal data and noise data. In addition, the pricing model obtained by training based on the random depth mechanism. Under the premise of ensuring model performance, the model training efficiency and pricing accuracy can be further improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a flow chart of the first embodiment of the premium pricing method based on the random deep network model of the present invention;

[0019] Figure 2 It is a flow chart of the second embodiment of the premium pricing method based on the random deep network model of the present invention;

[0020] Figure 3 It is a flow chart of the third embodiment of the premium pricing method based on the random deep network model of the present invention;

[0021] Figure 4 It is a schematic diagram of a premium pricing device based on a random deep network model of the present invention;

[0022] Figure 5 The figure is a schematic diagram of the hardware structure of a computer device involved in an embodiment of the present invention.

[0023] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0024] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0025] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0026] It should be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0027] It should be further understood that the term “and / or” used in the specification and appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0028] The premium pricing method based on the random deep network model involved in the embodiment of the present invention is mainly applied to computer equipment, which can be a PC, a portable computer, a mobile terminal and other equipment with display and processing functions.

[0029] In conjunction with the accompanying drawings, some embodiments of the present application are described in detail below. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0030] Reference Figure 1 , Figure 1 This is a flow chart of the first embodiment of the premium pricing method based on the random deep network model of the present invention.

[0031] like Figure 1 As shown, an embodiment of the present invention provides a premium pricing method based on a random deep network model, and the premium pricing method based on a random deep network model includes steps S10 to S40.

[0032] Step S10, initializing the weight parameters and bias parameters of each network layer in the random deep network model, and generating the survival probability parameters of each network layer based on preset rules;

[0033] At present, with the rapid growth of insurance product data and the increasing complexity of data structure, traditional actuarial models:

[0034] 1. High computational complexity: With the increase of insurance data, the computational complexity of traditional actuarial models has increased dramatically, resulting in huge computational time and resource consumption;

[0035] 2. High data dimension: Insurance data usually includes a large number of variables and features. Traditional models perform poorly when processing high-dimensional data, which can easily lead to overfitting or underfitting.

[0036] 3. Poor model robustness: Traditional models are unstable when dealing with abnormal data and noisy data, which affects the accuracy of risk assessment;

[0037] 4. Long training time: Although deep neural networks perform well in many tasks, their training time is very long, especially when the amount of data is huge, the training efficiency is low.

[0038] In this embodiment, in order to solve the above problems, a random depth technology based on a deep neural network is provided in this embodiment. By introducing a random depth mechanism in a deep neural network, the problems of high computational complexity, high data dimension, poor model robustness and long training time in the insurance field are solved. Specifically, based on the ResNet architecture, a deep neural network including a random depth mechanism is designed. By introducing the survival probability p in each layer of the network l , and decide whether this layer of network is skipped in the current training process based on the survival probability.

[0039] Specifically, the network parameters and the survival probability of each layer are initialized first. Then, in each forward propagation process, it is randomly decided whether to skip the layer according to the survival probability of each layer. For the network layers participating in the training, the parameters are updated by forward and backward propagation as usual. For the network layers not participating in the training, the parameters are not updated, thereby realizing multiple iterative training of the model to ensure that the model converges and achieves the optimal performance.

[0040] Exemplarily, the step S10 specifically includes:

[0041] Initializing the weight parameters and bias parameters of each network layer based on the He initialization method or the Xavier initialization method to obtain the initial weight parameters and the initial bias parameters;

[0042] Based on a linear decreasing strategy, the survival probability parameters of each network layer are generated, wherein the level of the network layer is inversely proportional to the corresponding survival probability parameter value.

[0043] In this embodiment, by introducing a random depth mechanism in the training process, efficient training of a deep neural network is achieved, specifically including:

[0044] 1. Initialize the network parameters, that is, initialize the weight parameters and bias parameters in each layer of the random deep neural network. Specifically, the model parameters of the random deep neural network can be initialized by using methods such as He initialization or Xavier initialization to ensure the stability and effectiveness of the training.

[0045] Second, initialize the survival probability, that is, assign an initial survival probability p to each layer of the network l .

[0046] Specifically, a linear decreasing strategy can be adopted, that is, the level of the network layer is inversely proportional to the corresponding survival probability parameter value. By adopting a linear decreasing survival probability design, the effective depth of the network during the training process is guaranteed, while avoiding overly complex calculations.

[0047] That is, the survival probability is higher near the input layer, while the survival probability is lower near the output layer. For example, the survival probability can be set as:

[0048]

[0049] Where L is the total number of network layers, p L It is the survival probability of the output layer in the previous round. The initial value can be taken in the first round. The value is generally low, such as less than 50%.

[0050] Step S20, based on the survival probability parameters of each network layer, determining a target network layer and an optimized network layer in each network layer, wherein the target network layer is a network layer to be trained, and the optimized network layer is a network layer for which training needs to be skipped;

[0051] In this embodiment, when training each network layer of the random deep network model, it is first determined whether to skip certain network layers based on the survival probability.

[0052] Specifically, a preset probability threshold can be set. When the survival probability parameter is not less than the preset probability threshold, the network layer can be recorded as the target network layer to train the network layer. When the survival probability parameter is less than the preset probability threshold, the network layer can be recorded as the optimized network layer to skip the network layer.

[0053] In a specific embodiment, a corresponding random number may be generated for each network layer, and the survival probability parameter of each network layer may be compared with the corresponding random number. When the survival probability parameter is not less than the random number, the network layer may be recorded as the target network layer; when the survival probability parameter is less than the random number, the network layer may be recorded as the optimized network layer.

[0054] Step S30, based on the preset risk assessment sample set, the initial weight parameter and the initial bias parameter, skipping the optimization network layer, and training the target network layer until the preset convergence condition is reached, to obtain the trained random deep network model as the target random deep network model;

[0055] In this embodiment, before training the random deep network model based on the preset risk assessment sample set, the original data corresponding to the preset risk assessment sample set can be preprocessed in combination with the specific needs of the insurance field, and the preprocessing includes data cleaning, feature selection and feature engineering, thereby obtaining a training sample set suitable for random deep network model training.

[0056] The initial random depth network model is a random depth network model with initial weight parameters and initial bias parameters as the current model parameters. In each forward propagation process, the insurer information in the preset risk assessment sample set is input into each target network layer in the initial random depth network model, that is, the output results of each target network layer skip the next layer of the optimization network layer and directly pass it to the next layer of the target network layer. After completing the training of all target network layers in the random depth network model, it is determined whether the current random depth network model has reached the convergence condition. If not, the target network layer in the random depth network model is iteratively trained again, and the optimization network layer is still skipped until the random depth network model reaches the preset convergence condition.

[0057] Among them, the preset convergence condition can be that the loss function of the random deep network model converges (that is, the loss function value is less than a preset function threshold), or the number of training times of the random deep network model reaches a preset number of training rounds.

[0058] Step S40: Based on the target random deep network model, a risk assessment is performed on the target user to obtain the risk level of the target user, and based on a preset pricing strategy and the risk level, the premium of the target user is calculated.

[0059] In this embodiment, the user information of the target user is input into the target random deep network model, so as to use the trained target random deep network model to perform risk assessment on the target user and output the risk level of the target user (i.e. the insured person). Then, based on the risk level of the target user and in combination with the premium pricing strategy of the corresponding insurance company, the reasonable premium of the target user is calculated.

[0060] Furthermore, based on the preset risk assessment sample set, the initial weight parameters and the initial bias parameters, the optimization network layer is skipped, and the target network layer is trained until the preset convergence condition is reached, and the trained random deep network model is obtained as the target random deep network model, and further includes:

[0061] Based on a preset test sample set and a preset test method, the target random depth network model is tested, and when the test accuracy is less than a preset accuracy threshold, the target random depth network model is updated. The preset test method includes a full depth test method and / or a random depth test method.

[0062] In this embodiment, after completing the training of the random depth network model, the model is further tested to ensure the stability and accuracy of the model prediction results. The test method includes a full depth test method (i.e., no network layer is skipped during the test) and a random depth test method (i.e., some network layers are skipped based on the survival probability). Specifically:

[0063] A. The full-depth test uses a full-depth network for forward propagation based on the test sample set, that is, no network layer in the target random depth network model is skipped during the test to obtain the test accuracy of the target random depth network model.

[0064] B. Random depth testing uses a random depth mechanism to perform forward propagation based on a test sample set. That is, during the test process, some network layers are skipped with the same or different survival probabilities to simulate the actual situation during training and obtain the test accuracy.

[0065] When the test accuracy is less than the preset accuracy threshold, it means that the prediction stability and accuracy of the target random deep network model are high; when the test accuracy is greater than the preset accuracy threshold, it means that the prediction stability and accuracy of the target random deep network model are low. The target random deep network model is updated until the test accuracy is less than the preset accuracy threshold to ensure the stability and accuracy of the prediction results of the random deep network model.

[0066] The embedding layer and convolution layer in the random deep training algorithm are used to reduce the dimensionality and extract features of the high-dimensional data corresponding to the sample set, thereby effectively reducing the computational complexity and training time of the deep neural network. While improving data processing efficiency and prediction accuracy, it further improves the robustness and performance of the random deep network model, making it particularly suitable for processing large-scale, high-dimensional financial and insurance data.

[0067] The present embodiment provides a premium pricing method based on a random deep network model, the method initializes weight parameters and bias parameters of each network layer in the random deep network model, and generates survival probability parameters of each network layer based on preset rules; based on the survival probability parameters of each network layer, determines a target network layer and an optimized network layer in each network layer, the target network layer is a network layer to be trained, and the optimized network layer is a network layer for which training needs to be skipped; based on a preset risk assessment sample set, initial weight parameters, and initial bias parameters, skips the optimized network layer, and trains the target network layer until a preset convergence condition is reached, thereby obtaining a trained random deep network model as a target random deep network model; based on the target random deep network model, performs risk assessment on a target user to obtain a risk level of the target user, and calculates the premium of the target user based on a preset pricing strategy and the risk level. Through the above method, the present application provides a random depth mechanism based on the survival probability corresponding to each network layer. Based on the random depth mechanism, some network layers can be randomly skipped during the training process, reducing the amount of forward and backward propagation calculations during model training, thereby reducing the computational complexity of model training; and only part of the network layer is used in each training, so that the training time is significantly shortened while ensuring the convergence of the model, preventing the model from overfitting, and enhancing the robustness of the model to abnormal data and noise data. In addition, the pricing model obtained by training based on the random depth mechanism. Under the premise of ensuring model performance, the model training efficiency and pricing accuracy can be further improved.

[0068] Reference Figure 2 , Figure 2 This is a flow chart of the second embodiment of the premium pricing method based on the random deep network model of the present invention.

[0069] Based on the above Figure 1 In the embodiment shown, in this embodiment, the step S20 specifically includes:

[0070] Step S21, obtaining any value within a preset value range as a random parameter of a network layer;

[0071] Step S22, comparing the survival probability parameter of each network layer with its corresponding random parameter;

[0072] Step S23, based on the comparison results of each network layer, determine the network layer whose survival probability parameter is not less than the random parameter as the target network layer, and determine the network layer whose survival probability parameter is less than the random parameter as the optimized network layer.

[0073] Wherein, the step S30 specifically includes:

[0074] The first level network layer of the random deep network model with the initial weight parameters and the initial bias parameters is obtained as the current network layer, and the insurer information sample in the risk assessment sample set is input into the current network layer to obtain a first output result;

[0075] Obtaining the next network layer of the current network layer, and determining whether the next network layer of the current network layer is the target network layer;

[0076] When the next network layer of the current network layer is the target network layer, inputting the first output result into the next network layer of the current network layer to obtain a second output result;

[0077] The current network layer is updated to the next network layer of the current network layer, and the execution is returned to: the next network layer of the current network layer is obtained, and it is determined whether the next network layer of the current network layer is the target network layer, until the training of all network layers in the random depth network model is completed, and the current risk level corresponding to the insurer information sample is output;

[0078] Based on the risk level samples corresponding to the insurer information samples in the risk assessment sample set and the current risk level, the random deep network model is iteratively trained until a preset convergence condition is reached to obtain the target random deep network model.

[0079] Further, after obtaining the next network layer of the current network layer and determining whether the next network layer of the current network layer is the target network layer, the method further includes:

[0080] When the next network layer of the current network layer is the optimized network layer, the current network layer is updated to the next network layer of the current network layer, and the execution is returned to: obtain the next network layer of the current network layer, and determine the next network layer of the current network layer to skip the training of the optimized network layer.

[0081] Furthermore, the iterative training of the random deep network model based on the risk level sample corresponding to the insurer information sample in the risk assessment sample set and the current risk level until a preset convergence condition is reached to obtain the target random deep network model specifically includes:

[0082] Based on a preset loss function, calculate the risk level sample corresponding to the insurer information sample and the loss function value of the current risk level;

[0083] Based on a preset learning rate and the loss function value, a current gradient corresponding to the target network layer is calculated, and the initial weight parameter and the initial bias parameter are updated based on the current gradient;

[0084] Based on the updated weight parameters and bias parameters, the random deep network model is iteratively trained until the loss function value is less than the preset function value or the number of model training rounds reaches the preset number of training rounds, thereby obtaining the target random deep network model.

[0085] For example, by introducing a random depth mechanism during the training process, efficient training of a deep neural network is achieved, that is, the network parameters and the survival probability of each layer are initialized, and then each network layer of the initialized random depth network model is forward propagated, and in each forward propagation process, a random decision is made whether to skip the layer based on the survival probability of each layer. Then, backpropagation is performed on the network layers involved in the training to update the parameters of the network layers involved in the training. Finally, multiple iterative training is performed to ensure that the model converges and achieves optimal performance. The specific process includes:

[0086] 1. Initialize network parameters: Initialize weight parameters and bias parameters at each layer of the deep neural network, using common methods such as He initialization or Xavier initialization to ensure the stability and effectiveness of training.

[0087] 2. Initialize survival probability: assign an initial survival probability p to each layer of the network l Generally, a linear decreasing strategy is adopted, that is, the survival probability of the layers close to the input layer is higher, while the survival probability of the layers close to the output layer is lower.

[0088] 3. Perform forward propagation on each network layer of the initialized random deep network model. During each forward propagation process, the survival probability is used to determine whether to skip certain network layers. The specific judgment process is as follows:

[0089] First, the insurer information sample in the risk assessment sample set is input into the first-level network layer.

[0090] Then, each network layer except the first-level network layer in the random deep network model is processed layer by layer, and each network layer of the initialized random deep network model is forward propagated:

[0091] For each network layer l, generate a random number r between [0,1] l , as random parameters of each level of network layer;

[0092] The random parameters r of each network layer are l The corresponding survival probability parameter p l To compare:

[0093] 1. If r l ≤p l , then this layer participates in the calculation and performs normal forward propagation:

[0094] x l+1 =f l (x l ,W l ,b l )

[0095] Among them, x l is the input of the lth layer, W l and b l are the weight and bias of the lth layer, respectively, and f l is the activation function of this layer.

[0096] 2. If r l >p l , then skip this layer and pass the current input directly to the next layer:

[0097] x l+1 =x l

[0098] In the above manner, this embodiment generates corresponding random parameters for each level of network layer to ensure the randomness of skipping the layer, and then compares the random parameters of the layer with its corresponding survival probability parameters, and determines whether to skip the layer during training based on the comparison results. Thus, the randomness of the optimized network layer is ensured based on the random parameters, thereby avoiding the fixed probability parameter threshold leading to a fixed optimized network layer, and further improving the accuracy of the prediction results of the random deep network model.

[0099] After completing a round of forward propagation of the model, back propagation is performed to update the parameters of the network layers involved in the forward propagation. Specifically:

[0100] 1. Calculate the risk level sample (i.e. target value y) corresponding to the insurer information sample true ) and the loss function of the current risk level (network output y):

[0101] According to the network output y and target value y true Calculate the loss function L(y,y true ), using cross entropy loss or mean squared error loss.

[0102] 2. Back propagation layer by layer:

[0103] Starting from the output layer, the gradient is calculated forward layer by layer.

[0104] For each layer l, if the layer participates in the forward propagation (i.e., r l ≤p l ), the gradient is calculated normally and the parameters are updated:

[0105] Among them, η is the preset learning rate.

[0106] If the layer is skipped (i.e. r l >p l ), no parameter update is performed and the gradient is directly passed to the previous layer:

[0107] In the above manner, this embodiment will iterate multiple times on the entire training set based on the above forward propagation and back propagation steps until the loss function converges or reaches a preset number of training rounds. During each iteration, the survival probability p of each network layer in this round is l Keep it unchanged to ensure the stability and effect of model training.

[0108] In summary, compared with the traditional model, this application:

[0109] 1. Introducing random depth mechanism: Introducing random depth mechanism in deep neural network reduces computational complexity and improves model training efficiency and robustness by randomly skipping some network layers.

[0110] 2. Design of survival probability: The linearly decreasing survival probability design is adopted to ensure the effective depth of the network during the training process while avoiding overly complex calculations.

[0111] 3. Adapt to high-dimensional data: By introducing embedding layers and convolutional layers, the features of high-dimensional insurance data can be effectively processed and extracted, improving the performance of the model.

[0112] 4. Multi-level risk assessment: The risk assessment method based on random deep neural network provides more detailed and multi-level risk assessment results, thereby providing insurance companies with more accurate premium pricing.

[0113] This solves problems in the insurance field such as high computational complexity, high data dimension, poor model robustness and long training time.

[0114] Reference Figure 3 , Figure 3 This is a flow chart of the third embodiment of the premium pricing method based on the random deep network model of the present invention.

[0115] Based on the above Figure 1 In the embodiment shown, in this embodiment, after step S30, the following steps are further included:

[0116] Step S50, based on a preset test sample set and a preset test method, the target random depth network model is tested, and when the test accuracy is less than a preset accuracy threshold, the target random depth network model is updated, and the preset test method includes a full depth test method and / or a random depth test method.

[0117] In this embodiment, after completing the training of the random depth network model, the model is further tested to ensure the stability and accuracy of the model prediction results. The test method includes a full depth test method (i.e., no network layer is skipped during the test) and a random depth test method (i.e., some network layers are skipped based on the survival probability). Specifically:

[0118] A. The full-depth test uses a full-depth network for forward propagation based on the test sample set, that is, no network layer in the target random depth network model is skipped during the test to obtain the test accuracy of the target random depth network model.

[0119] B. Random depth testing uses a random depth mechanism to perform forward propagation based on a test sample set. That is, during the test process, some network layers are skipped with the same or different survival probabilities to simulate the actual situation during training and obtain the test accuracy.

[0120] When the test accuracy is less than the preset accuracy threshold, it means that the prediction stability and accuracy of the target random deep network model are high; when the test accuracy is greater than the preset accuracy threshold, it means that the prediction stability and accuracy of the target random deep network model are low. The target random deep network model is updated until the test accuracy is less than the preset accuracy threshold to ensure the stability and accuracy of the prediction results of the random deep network model.

[0121] The embedding layer and convolution layer in the random deep training algorithm are used to reduce the dimensionality and extract features of the high-dimensional data corresponding to the sample set, thereby effectively reducing the computational complexity and training time of the deep neural network. While improving data processing efficiency and prediction accuracy, it further improves the robustness and performance of the random deep network model, making it particularly suitable for processing large-scale, high-dimensional financial and insurance data.

[0122] See also Figure 4 , Figure 4 The present invention provides a schematic diagram of the functional modules of a premium pricing device based on a random deep network model.

[0123] like Figure 4 As shown, the premium pricing device 400 based on the random deep network model includes:

[0124] The random parameter generation module 410 is used to initialize the weight parameters and bias parameters of each network layer in the random deep network model, and generate the survival probability parameters of each network layer based on preset rules;

[0125] A network layer classification module 420, for determining a target network layer and an optimized network layer in each network layer based on the survival probability parameters of each network layer, wherein the target network layer is a network layer to be trained, and the optimized network layer is a network layer for which training needs to be skipped;

[0126] A random model training module 430 is used to skip the optimization network layer and train the target network layer based on a preset risk assessment sample set, an initial weight parameter, and an initial bias parameter until a preset convergence condition is reached, thereby obtaining a trained random deep network model as a target random deep network model;

[0127] The user premium pricing module 440 is used to perform risk assessment on the target user based on the target random deep network model, obtain the risk level of the target user, and calculate the premium of the target user based on a preset pricing strategy and the risk level.

[0128] Furthermore, the network layer classification module 420 specifically includes:

[0129] A random parameter generating unit, used to obtain any value within a preset value range as a random parameter of a network layer;

[0130] A network parameter comparison unit, used to compare the survival probability parameter of each network layer with its corresponding random parameter;

[0131] The network layer classification unit is used to determine, based on the comparison results of each network layer, the network layer whose survival probability parameter is not less than the random parameter as the target network layer, and determine the network layer whose survival probability parameter is less than the random parameter as the optimized network layer.

[0132] Furthermore, the random model training module 430 specifically includes:

[0133] A first network training unit is used to obtain the first level network layer of the random deep network model of the initial weight parameters and the initial bias parameters as the current network layer, and input the insurer information sample in the risk assessment sample set into the current network layer to obtain a first output result;

[0134] A first network determination unit, configured to obtain a next network layer of the current network layer, and determine whether the next network layer of the current network layer is the target network layer;

[0135] an output result delivery unit, configured to input the first output result into the next network layer of the current network layer to obtain a second output result when the next network layer of the current network layer is the target network layer;

[0136] The current network updating unit is used to update the current network layer to the next network layer of the current network layer, and return to execute: obtaining the next network layer of the current network layer, and judging whether the next network layer of the current network layer is the target network layer, until the training of all network layers in the random depth network model is completed, and the current risk level corresponding to the insurer information sample is output;

[0137] The target model generating unit is used to iteratively train the random deep network model based on the risk level samples corresponding to the insurer information samples in the risk assessment sample set and the current risk level until a preset convergence condition is reached to obtain the target random deep network model.

[0138] Furthermore, the random model training module 430 specifically includes:

[0139] The optimized network skipping unit is used to update the current network layer to the next network layer of the current network layer when the next network layer of the current network layer is the optimized network layer, and return to execute: obtain the next network layer of the current network layer, and determine the next network layer of the current network layer to skip the training of the optimized network layer.

[0140] Furthermore, the target model generating unit is also used for:

[0141] Based on a preset loss function, calculate the risk level sample corresponding to the insurer information sample and the loss function value of the current risk level;

[0142] Based on a preset learning rate and the loss function value, a current gradient corresponding to the target network layer is calculated, and the initial weight parameter and the initial bias parameter are updated based on the current gradient;

[0143] Based on the updated weight parameters and bias parameters, the random deep network model is iteratively trained until the loss function value is less than the preset function value or the number of model training rounds reaches the preset number of training rounds, thereby obtaining the target random deep network model.

[0144] Furthermore, the random parameter generation module 410 specifically includes:

[0145] A parameter initialization unit, used to initialize the weight parameters and bias parameters of each network layer based on the He initialization method or the Xavier initialization method to obtain the initial weight parameters and the initial bias parameters;

[0146] The survival probability generating unit is used to generate the survival probability parameters of each network layer based on a linear decreasing strategy, wherein the level of the network layer is inversely proportional to the corresponding survival probability parameter value.

[0147] Furthermore, the premium pricing device 400 based on the random deep network model further includes:

[0148] The target model testing module is used to test the target random depth network model based on a preset test sample set and a preset test method, and to update the target random depth network model when the test accuracy is less than a preset accuracy threshold. The preset test method includes a full-depth test method and / or a random depth test method.

[0149] It should be noted that those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described device and each module can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0150] The above-mentioned device can be implemented in the form of a computer program. Figure 5 Runs on the computer device shown.

[0151] See also Figure 5 , Figure 5 1 is a schematic block diagram of a computer device provided in an embodiment of the present application. The computer device may be a server.

[0152] See also Figure 5 The computer device includes a processor, a memory and a network interface connected through a system bus, wherein the memory may include a non-volatile storage medium and an internal memory.

[0153] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions, and when the program instructions are executed, the processor can execute any premium pricing method based on a random deep network model.

[0154] The processor is used to provide computing and control capabilities and support the operation of the entire computer equipment.

[0155] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor, the processor can execute any premium pricing method based on the random deep network model.

[0156] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art will understand that Figure 5The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0157] It should be understood that the processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0158] In one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:

[0159] Initialize the weight parameters and bias parameters of each network layer in the random deep network model, and generate the survival probability parameters of each network layer based on preset rules;

[0160] Based on the survival probability parameters of each network layer, a target network layer and an optimized network layer are determined in each network layer, wherein the target network layer is a network layer to be trained, and the optimized network layer is a network layer for which training needs to be skipped;

[0161] Based on a preset risk assessment sample set, an initial weight parameter, and an initial bias parameter, the optimization network layer is skipped, and the target network layer is trained until a preset convergence condition is reached, thereby obtaining a trained random deep network model as a target random deep network model;

[0162] Based on the target random deep network model, a risk assessment is performed on the target user to obtain the risk level of the target user, and based on a preset pricing strategy and the risk level, the premium of the target user is calculated.

[0163] In one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:

[0164] Obtain any value within a preset value range as a random parameter of a network layer;

[0165] Compare the survival probability parameter of each network layer with its corresponding random parameter;

[0166] Based on the comparison results of each network layer, the network layer whose survival probability parameter is not less than the random parameter is determined as the target network layer, and the network layer whose survival probability parameter is less than the random parameter is determined as the optimized network layer.

[0167] In one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:

[0168] The first level network layer of the random deep network model with the initial weight parameters and the initial bias parameters is obtained as the current network layer, and the insurer information sample in the risk assessment sample set is input into the current network layer to obtain a first output result;

[0169] Obtaining the next network layer of the current network layer, and determining whether the next network layer of the current network layer is the target network layer;

[0170] When the next network layer of the current network layer is the target network layer, inputting the first output result into the next network layer of the current network layer to obtain a second output result;

[0171] The current network layer is updated to the next network layer of the current network layer, and the execution is returned to: the next network layer of the current network layer is obtained, and it is determined whether the next network layer of the current network layer is the target network layer, until the training of all network layers in the random depth network model is completed, and the current risk level corresponding to the insurer information sample is output;

[0172] Based on the risk level samples corresponding to the insurer information samples in the risk assessment sample set and the current risk level, the random deep network model is iteratively trained until a preset convergence condition is reached to obtain the target random deep network model.

[0173] In one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:

[0174] When the next network layer of the current network layer is the optimized network layer, the current network layer is updated to the next network layer of the current network layer, and the execution is returned to: obtain the next network layer of the current network layer, and determine the next network layer of the current network layer to skip the training of the optimized network layer.

[0175] In one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:

[0176] Based on a preset loss function, calculate the risk level sample corresponding to the insurer information sample and the loss function value of the current risk level;

[0177] Based on a preset learning rate and the loss function value, a current gradient corresponding to the target network layer is calculated, and the initial weight parameter and the initial bias parameter are updated based on the current gradient;

[0178] Based on the updated weight parameters and bias parameters, the random deep network model is iteratively trained until the loss function value is less than the preset function value or the number of model training rounds reaches the preset number of training rounds, thereby obtaining the target random deep network model.

[0179] In one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:

[0180] Initializing the weight parameters and bias parameters of each network layer based on the He initialization method or the Xavier initialization method to obtain the initial weight parameters and the initial bias parameters;

[0181] Based on a linear decreasing strategy, the survival probability parameters of each network layer are generated, wherein the level of the network layer is inversely proportional to the corresponding survival probability parameter value.

[0182] In one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:

[0183] Based on a preset test sample set and a preset test method, the target random depth network model is tested, and when the test accuracy is less than a preset accuracy threshold, the target random depth network model is updated. The preset test method includes a full depth test method and / or a random depth test method.

[0184] A computer-readable storage medium is also provided in an embodiment of the present application, wherein the computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and the processor executes the program instructions to implement any of the premium pricing methods based on the random deep network model provided in the embodiments of the present application.

[0185] The computer-readable storage medium may be an internal storage unit of the computer device described in the above embodiment, such as a hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the computer device.

[0186] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A premium pricing method based on a random deep network model, characterized in that: The premium pricing method comprises the following steps: Initialize the weight parameters and bias parameters of each network layer in the random deep network model, and generate the survival probability parameters of each network layer based on preset rules; Based on the survival probability parameters of each network layer, a target network layer and an optimized network layer are determined in each network layer, wherein the target network layer is a network layer to be trained, and the optimized network layer is a network layer for which training needs to be skipped; Based on a preset risk assessment sample set, an initial weight parameter, and an initial bias parameter, the optimization network layer is skipped, and the target network layer is trained until a preset convergence condition is reached, thereby obtaining a trained random deep network model as a target random deep network model; Based on the target random deep network model, a risk assessment is performed on the target user to obtain the risk level of the target user, and based on a preset pricing strategy and the risk level, the premium of the target user is calculated.

2. The premium pricing method based on the random deep network model according to claim 1, characterized in that: The determining of the target network layer and optimizing the network layer at each network layer based on the survival probability parameters of each network layer specifically includes: Obtain any value within a preset value range as a random parameter of a network layer; Compare the survival probability parameter of each network layer with its corresponding random parameter; Based on the comparison results of each network layer, the network layer whose survival probability parameter is not less than the random parameter is determined as the target network layer, and the network layer whose survival probability parameter is less than the random parameter is determined as the optimized network layer.

3. The premium pricing method based on the random deep network model as claimed in claim 2, characterized in that: Based on the preset risk assessment sample set, the initial weight parameters and the initial bias parameters, the optimization network layer is skipped, and the target network layer is trained until the preset convergence condition is reached to obtain a trained random deep network model, which specifically includes: The first level network layer of the random deep network model with the initial weight parameters and the initial bias parameters is obtained as the current network layer, and the insurer information sample in the risk assessment sample set is input into the current network layer to obtain a first output result; Obtaining the next network layer of the current network layer, and determining whether the next network layer of the current network layer is the target network layer; When the next network layer of the current network layer is the target network layer, inputting the first output result into the next network layer of the current network layer to obtain a second output result; The current network layer is updated to the next network layer of the current network layer, and the execution is returned to: the next network layer of the current network layer is obtained, and it is determined whether the next network layer of the current network layer is the target network layer, until the training of all network layers in the random depth network model is completed, and the current risk level corresponding to the insurer information sample is output; Based on the risk level samples corresponding to the insurer information samples in the risk assessment sample set and the current risk level, the random deep network model is iteratively trained until a preset convergence condition is reached to obtain the target random deep network model.

4. The premium pricing method based on the random deep network model as claimed in claim 3, characterized in that: After obtaining the next network layer of the current network layer and determining whether the next network layer of the current network layer is the target network layer, the method further includes: When the next network layer of the current network layer is the optimized network layer, the current network layer is updated to the next network layer of the current network layer, and the execution is returned to: obtain the next network layer of the current network layer, and determine the next network layer of the current network layer to skip the training of the optimized network layer.

5. The premium pricing method based on the random deep network model as claimed in claim 3, characterized in that: The iterative training of the random deep network model based on the risk level samples corresponding to the insurer information samples in the risk assessment sample set and the current risk level until a preset convergence condition is reached to obtain the target random deep network model specifically includes: Based on a preset loss function, calculate the risk level sample corresponding to the insurer information sample and the loss function value of the current risk level; Based on a preset learning rate and the loss function value, a current gradient corresponding to the target network layer is calculated, and based on the current gradient, the initial weight parameter and the initial bias parameter are updated; Based on the updated weight parameters and bias parameters, the random deep network model is iteratively trained until the loss function value is less than the preset function value or the number of model training rounds reaches the preset number of training rounds, so as to obtain the target random deep network model.

6. The premium pricing method based on the random deep network model according to claim 1, characterized in that: Initializing the weight parameters and bias parameters of each network layer in the random deep network model and generating the survival probability parameters of each network layer based on preset rules specifically includes: Initializing the weight parameters and bias parameters of each network layer based on the He initialization method or the Xavier initialization method to obtain the initial weight parameters and the initial bias parameters; Based on a linear decreasing strategy, the survival probability parameters of each network layer are generated, wherein the level of the network layer is inversely proportional to the corresponding survival probability parameter value.

7. The premium pricing method based on a random deep network model according to any one of claims 1 to 6, characterized in that: The method further comprises: skipping the optimization network layer based on the preset risk assessment sample set, the initial weight parameter and the initial bias parameter, and training the target network layer until a preset convergence condition is reached to obtain a trained random deep network model as the target random deep network model; Based on a preset test sample set and a preset test method, the target random depth network model is tested, and when the test accuracy is less than a preset accuracy threshold, the target random depth network model is updated. The preset test method includes a full depth test method and / or a random depth test method.

8. A premium pricing device based on a random deep network model, characterized in that: The premium pricing device based on the random deep network model includes: The random parameter generation module is used to initialize the weight parameters and bias parameters of each network layer in the random deep network model, and generate the survival probability parameters of each network layer based on preset rules; A network layer classification module, used to determine a target network layer and an optimized network layer in each network layer based on the survival probability parameters of each network layer, wherein the target network layer is a network layer to be trained, and the optimized network layer is a network layer for which training needs to be skipped; A random model training module is used to skip the optimization network layer based on a preset risk assessment sample set, an initial weight parameter, and an initial bias parameter, and train the target network layer until a preset convergence condition is reached, thereby obtaining a trained random deep network model as a target random deep network model; The user premium pricing module is used to perform risk assessment on the target user based on the target random deep network model, obtain the risk level of the target user, and calculate the premium of the target user based on a preset pricing strategy and the risk level.

9. A computer device, characterized in that: The computer device includes a processor, a memory, and a premium pricing program based on a random deep network model stored in the memory and executable by the processor, wherein when the premium pricing program based on the random deep network model is executed by the processor, the steps of the premium pricing method based on the random deep network model as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a premium pricing program based on a random deep network model, wherein when the premium pricing program based on a random deep network model is executed by a processor, the steps of the premium pricing method based on a random deep network model as described in any one of claims 1 to 7 are implemented.

Images