CN115238789A

CN115238789A - Financial industry special data prediction method and system based on improved GRU

Info

Publication number: CN115238789A
Application number: CN202210867883.0A
Authority: CN
Inventors: 吕智涵; 王娜娜; 田亚峻; 马潇萌; 孙运传; 孟毅; 娄冉冉; 刘筱成
Original assignee: Xinhua Fusion Media Technology Development Beijing Co ltd
Current assignee: Xinhua Fusion Media Technology Development Beijing Co ltd
Priority date: 2022-07-22
Filing date: 2022-07-22
Publication date: 2022-10-25

Abstract

The invention discloses a financial industry alternative data prediction method and a system based on improved GRU, comprising the following steps: the system comprises an input feature acquisition module, a hyper-parameter determination module, a weight distribution module and a comparison module, wherein the input feature of a prediction model needs to be determined firstly. Secondly, determining each hyperparameter of the model by a Bayesian optimization algorithm. Subsequently, at the hidden layer of the model, the features are assigned different weights by an attention mechanism. And training by mass data to respectively obtain sensor data and a periodic prediction model. And comparing the prediction result of the model with the observed value by using the test set, and judging whether the end condition is reached according to a Bayesian optimization method. The invention has the advantages that: the method realizes accurate financial industry alternative data prediction, provides reliable alternative data change trend guidance, and provides powerful support for financial industry data analysis based on artificial intelligence algorithm.

Description

Financial industry special data prediction method and system based on improved GRU

Technical Field

The invention relates to the technical field of artificial intelligence data mining, in particular to a financial industry special data prediction method and system based on improved GRU.

Background

New technical innovation including key core technologies such as the Internet of things, block chains and artificial intelligence becomes the focus of international competition, and meanwhile financial science and technology development mode innovation is promoted. The internet of things is one of the most widely applied underlying technologies of financial science and technology, is deeply integrated with the vertical industry and is continuously applied to various businesses of the financial industry, and the internet of things technology continuously breaks the barrier of the financial industry and creates new value. The common tool for data analysis of artificial intelligence technology is widely applied to data analysis in the financial industry, in particular to the analysis of other types of data. In the process of fusing artificial intelligence technology into the analysis of the financial decision system, the highest earning rate at the lowest risk level is taken as the target in the constructed investment portfolio. Meanwhile, on the premise of an artificial intelligence technology, an intelligent agent consisting of a learning system, a decision-making system and intelligent execution predicts the financial development trend through data monitoring and module analysis.

At present, the finance of the Internet of things is widely applied to multiple fields such as financial credit, financial leasing and insurance. In the field of financial credit, the state of a collateral is monitored in real time by the finance of the Internet of things through the technology of the Internet of things, and the wind control accuracy is improved; in the finance lease field, the thing networking finance is through internet of things to the subject real time monitoring of financing lease, ensures that equipment safety, not by the use, simultaneously through the state monitoring analysis enterprise production and operation information to production facility, ensures the normal repayment ability of enterprise to guarantee the wind accuse accuracy of financing lease mechanism. In the field of insurance, the finance of the internet of things promotes the development of UBI car insurance business, the driving habits of drivers are known through the internet of vehicles technology, the driving habits and data such as vehicle information and surrounding environment are integrated, a human, vehicle and road (environment) multidimensional model pricing system is established, and the system helps insurance companies to reasonably price, accurately obtain passengers, reduce the odds paid rate and the like. At present, many enterprises develop the deployment of the financial Internet of things.

In practical applications, many financial companies on the market start to deploy the financial internet of things. The Sunning finance is based on the supervision of the warehousing vehicles of the Internet of things, an automobile garage fusion platform is established, the remote monitoring requirements of the vehicles are met through a mobile phone program, the correspondence between the loan flow direction and trade opponents, between a pledge list and stock real objects, between a loan amount and pledge value is realized, and the transparency of information and the credit risk management and control capability are enhanced. The Jiangsu bank utilizes the internet of things and the block chain technology to provide the online internet-of-things movable property pledge financing business. By collecting the conditions of warehouses and passenger flow in real time, collecting production energy consumption, acquiring the pledge information of enterprises, mastering the operating capacity and the debt paying capacity, one-to-one correspondence between the materials and the loans is realized, the problems of property risks and information transfer of the past material pooling are avoided, the time and the capital cost of financial services are greatly saved, and the turnover requirement of production and operation capital is timely solved. The safety bank completes the nationwide layout of the internet of things warehouse and establishes a deep cooperation relationship with hundreds of large warehouses and logistics parks; and a new paradigm of port Internet of things financial cooperation is created, and the port Internet of things financial cooperation is cooperated with bulk cargos such as mineral products and grains developed in more than 10 large ports.

The alternative data is new data different from the traditional exchange disclosure and company bulletin disclosure, and is valuable information for investors to make investment decision. The method has the advantages that the method is widely applied to the fields of credit, anti-fraud, abnormal transaction detection and the like at present through the analysis of the financial alternative data by artificial intelligence, and provides solid technical support for the fraud risk analysis and early warning monitoring of the financial industry. At present, the financial security is ensured by using artificial intelligence at home and abroad, and a great deal of research is also carried out on the aspect of financial decision.

With the continuous popularization of artificial intelligence in the financial field, relevant standards for standardizing the use of heterogeneous data generated by the internet of things, the application specification of artificial intelligence in finance and the like in the financial industry are lacking at present, and the current deployment of artificial intelligence is standardized. However, the national standards related to financial security, which only include the GB/T36618-2018 information security technology and financial information service security specifications, do not describe artificial intelligence and evaluation criteria for alternative data in the financial field although providing the related contents of financial information security. In addition, the IEEE standard definition committee (IEEE SA) is currently performing financial security related work, mainly discussing the impact of artificial intelligence on the financial industry. At present, the IEEE standard definition committee is making two standards, namely a service standard of a credible data and artificial intelligence system and an influence of the artificial intelligence system on the risk of the financial industry.

Disclosure of Invention

Aiming at the defects of the prior art, the invention provides a financial industry alternative data prediction method and system based on an improved GRU.

In order to realize the purpose, the technical scheme adopted by the invention is as follows:

a financial industry special data prediction method based on improved GRU comprises the following steps:

first the input features of the prediction model need to be determined.

Secondly, determining each hyperparameter of the model by a Bayesian optimization algorithm.

Subsequently, at the hidden layer of the model, the features are assigned different weights by a focus mechanism.

And training a large amount of data to respectively obtain sensor data and a periodic prediction model.

And comparing the prediction result of the model with the observed value by using the test set, and judging whether the end condition is reached according to a Bayesian optimization method.

If yes, the model is used for predicting the change trend of the financial alternative data; if not, continuing parameter optimization.

Further, during the training process, mean Square Error (MSE) is selected as the loss function, as shown in formula (1), where n is the number of samples, y is _i Is an observed value, x _i Is a predicted value. The updating of the weight parameters is done by an adaptive momentum estimation (Adam) algorithm. The Adam optimizer combines the advantages of the RMSProp and AdaGrad algorithms that are good at dealing with sparse gradients and non-stationary targets, and can obtain good results at a fast speed. In order to prevent overfitting in the model training process, early Stopping is adopted, and with the increase of training rounds, if the test error on the verification set rises, the training is stopped.

Furthermore, the financial industry special data prediction method needs to use a front-and-back average value filling mode to process the missing value, namely, the average value of the attribute value of the missing value at the previous moment and the attribute value of the missing value at the next moment is used as the filling value of the missing moment. The fill value is calculated as shown in equation (19). When a plurality of continuous values are missing, the average value of two non-null values adjacent to each other before and after the missing is used for filling.

Furthermore, various insurance premiums of the insurance beneficiaries in the last year and the income level of the insurance beneficiaries are selected as input characteristics of the prediction model.

Further, feature normalization: zero-mean normalization processing of the feature data was used. The normalized data mean is 0 and standard deviation is 1, and obey the standard normal distribution. The calculation formula is shown in (3), wherein n is the sample size, X ^* Is the processed data, X is the raw data,

is the mean of the raw data, δ is the standard deviation of the raw data. Standard of referenceThe difference is calculated as shown in equation (4).

Further, the hyperparameter is optimized based on a Bayesian optimization algorithm, and the optimization iteration number is 30. And (3) the value range and the final value of the hyper-parameter to be optimized, wherein time step is the time step, units is the number of neurons, dense is the number of nodes of a full connection layer, n identifiers tree number, and max depth is the maximum depth of the tree. In addition, the learning rate of the neural network GRU is 0.001, and the training rounds are 100. The GRU activation function is tanh.

The invention also discloses a financial industry alternative data prediction system based on the improved GRU, which comprises the following steps:

the input characteristic acquisition module is used for determining the input characteristics of the prediction model;

the hyper-parameter determining module is used for determining each hyper-parameter of the model;

the weight distribution module distributes different weights to the features through an attention mechanism;

the comparison module is used for comparing the prediction result of the model with the observation value by using the test set, judging whether the ending condition is reached according to a Bayesian optimization method, and if so, predicting the change trend of the financial alternative data by using the model; if not, continuing parameter optimization.

The invention also discloses a prediction device for predicting financial industry alternative data, which comprises a memory, a processor and a computer program which is stored in the memory and can run on the processor, wherein the processor realizes the financial industry alternative data prediction method based on the improved GRU when executing the computer program.

The invention also discloses a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to realize the financial industry special data prediction method based on the improved GRU.

Compared with the prior art, the invention has the advantages that:

more reliable alternative data trend guidance can be provided for financial practitioners in a specific scene. Previous studies have shown that GRU can simplify the neural network structure and improve its learning efficiency compared to LSTM because of the elimination of contributing small gates and their corresponding weights. The model is added with a Bayesian optimization algorithm to optimize the hyper-parameters of the model on the basis of the original GRU, and in addition, an attention mechanism is added to distribute different weights to the features in the training process of the model so as to achieve a more accurate prediction effect. The GRU based on the Bayesian optimization and attention mechanism can realize accurate financial industry alternative data prediction, and provides powerful support for financial industry data analysis based on an artificial intelligence algorithm.

Drawings

FIG. 1 is a schematic structural diagram of an LSTM network unit;

FIG. 2 is a schematic diagram of a GRU network unit structure;

FIG. 3 is a diagram illustrating a comparison between a random search and a web search;

FIG. 4 is a comparison graph of predicted data for an embodiment of the present invention;

FIG. 5 is a comparison graph of alternative data prediction according to an embodiment of the present invention;

FIG. 6 is a comparison graph of alternative data change predictions according to an embodiment of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be further described in detail below with reference to the accompanying drawings by way of examples.

Gated cyclic unit (GRU) networks are developed based on RNN. RNN has memory, parameter sharing and picture-flexibility. By establishing associations for neurons between layers in the network, RNNs solve the problem of mutual independence of front and back inputs in traditional neural networks. Therefore, RNN has certain advantages in learning the non-linear characteristics of a sequence, making it more suitable for dealing with time problems. RNN is widely applied in the fields of natural language processing, time series prediction and the like. To address the long-term dependency problem, neural networks based on RNN improvement continue to emerge, including LSTM and GRU.

LSTM controls the transfer of information in the network through three gate devices (forgetting gate, input gate, output gate). Each gate contains a sigma (sigmoid function) and a dot product operation. The sigma outputs a number between 0 and 1 indicating how much information can pass, 0 indicating that no information is allowed to pass, and 1 indicating that any information is allowed to pass, the calculation formula is shown in equation (1). The three gates create a self-loop over the internal state of the LSTM cell relative to the recursive computation that RNN creates over the system state. The input gate determines the input of the current time step and the updating of the internal state by the system state of the previous time step; the forgetting gate determines the updating of the internal state of the previous time step to the internal state of the current time step; the output gate determines the update of the internal state to the system state. The structure of the LSTM is shown in fig. 1.

The GRU unit includes only an update gate and a reset gate. The update gate is similar to the forget gate and the output gate of LSTM and is used to control the extent to which the state information at the previous time is brought into the current state, with a larger value of the update gate indicating more state information is brought in at the previous time. The reset gate is similar to the input gate of LSTM and determines how to combine new input information with previous memory, the smaller the reset gate, the less information of the previous state is written. The structure of the GRU is shown in fig. 2. Updating r in a door _t And z in the reset gate _t Obtained from formula (2) and formula (3), respectively. Where U and W are weight parameters.

R _t ＝σ(W _r x _t +U _r h _t-1 )#(2)

Z _t ＝σ(W _z x _t +U _z h _t-1 )#(3)

Current hidden state h _t From equation (4), wherein the candidate set

The calculation process of (2) is shown in equation (5). tan h is a hyperbolic tangent function, and the expression is shown in equation (6).

Bayesian optimization

The super-parameter tuning of the model is one of the important factors influencing the final prediction effect. At present, the commonly used parameter tuning methods in research are grid search, random search and bayesian optimization. And (3) the grid search tries the effect of each parameter value combination in the test set in a traversal mode according to the given candidate list values of all the parameters, and finally finds out the parameter value combination with the best effect. The trellis search is very time consuming because all combinations of parameter candidates need to be traversed. The random search is similar to the grid search, but unlike the grid search, which traverses all parameter value combinations, it randomly selects a fixed number of parameter value combinations within a given parameter value range to achieve the purpose of finding out the optimal parameter value or the approximate value of the optimal parameter value. The random search has a faster search speed, but the resulting parameter values may not be optimal parameter values. The difference between the random search and the grid search is shown in fig. 3.

The optimizing strategy of the Bayesian parameter optimizing method is to use Gaussian process to the parameter value combination selected in sampling mode to obtain the posterior distribution of the given target function, and then to select the posterior distribution according to the posterior distribution of the previous parameter value combinationUntil the posterior distribution matches the true distribution. For search space X _n Optimum solution x for Bayesian optimization _best Can be expressed by equation (7), where f is the objective function. Compared with grid search and random search, the Bayesian optimization method has the advantages of less cycle times, higher speed and more stable performance, and can make a decision on the next selection by using historical parameter value combinations in a mode of continuously updating a priori through a Gaussian process.

The bayesian-optimized gaussian process consists of a mean and a covariance function, as shown in equation (8), where μ is the mean and k (x, x') is the covariance function. For dataset D = { (x) ₁ ,f(x ₁ )),(x ₂ ,f(x ₂ )),…,(x _t ,f(x _t ) ) } gaussian distribution as shown in equation (9).

f(x)～gp(μ,k(x,x′))#(8)

For new sample x _t+1 The Gaussian distribution is shown in equation (10). f. of _t+1 The posterior probability distribution of (2) is shown in equation (13).

k＝[((x _t+1 ,x ₁ ),(x _t+1 ,x ₂ )…(x _t+1 ,x _t ))]#(12)

P(f _t+1 |D,x _t+1 )＝gp(u(x _t+1 ),δ ² (x _t+1 ))#(13)

u(x _t+1 )＝kK ^-1 f _1∶t #(14)

δ ² (x _t+1 )＝k(x _t+1 ,x _t+1 )-kK ^-1 k ^T #(15)

The process of bayesian optimization is as follows:

1) And randomly initializing and selecting a group of parameter value combinations in a search space, and calculating the value of the target optimization function.

2) And continuing to randomly select a hyper-parameter combination, calculating an objective function value, and if the value is better than the optimal value acquired historically, saving the point.

3) And (5) repeating the step (2) until the set iteration times are reached.

Attention mechanism

Attention mechanisms stem from the study of human vision. In cognitive science, due to the bottleneck of information processing, a human being selectively focuses on a part of information that the human being wants to see, while ignoring other visible information. This mechanism is called the attention mechanism. Attention mechanisms are now widely used in the field of artificial intelligence, including image recognition, natural language processing, speech recognition, and the like. In neural networks, the attention mechanism is the focus on the assignment of input weights. Note that the force mechanism can assign weights to the importance levels of the elements, with high weights to de-focus important information, and low weights to ignore irrelevant information. And it can also adjust the weight continuously, make it choose important information under different situations too, therefore have higher scalability and robustness. In the timing prediction problem, attention mechanism can prevent important features from being ignored due to the increase of time step. The weight assignment method can be expressed by the formulas (16), (17), wherein h _t For hidden layer state vectors in the neural network at time t, e _t To pay attention to the probability distribution value, a _t To score attention, u _a And W _a As an attention weight vector, b _a Is the attention bias vector.

e _t ＝u _a tanh(W _a h _t +b _a )#(17)

Improved GRU financial alternative data prediction model

1. Model structure

First, input features of the prediction model need to be determined. Secondly, determining each hyperparameter of the model by a Bayesian optimization algorithm. Subsequently, at the hidden layer of the model, the features are assigned different weights by an attention mechanism. And training a large amount of data to respectively obtain sensor data and a periodic prediction model. And comparing the prediction result of the model with the observed value by using a test set, and judging whether the end condition is reached according to a Bayesian optimization method. If yes, predicting the change trend of the financial type-II data by using the model; if not, the parameter optimization is continued.

In the training process of the model, the Mean Square Error (MSE) is selected as a loss function in the invention, as shown in a formula (18), wherein n is the number of samples, y _i Is an observed value, x _i Is a predicted value. The updating of the weight parameters is done by an adaptive momentum estimation (Adam) algorithm. The Adam optimizer combines the advantages of the RMSProp and AdaGrad algorithms that are good at dealing with sparse gradients and non-stationary targets, and can obtain good results at a fast speed. In order to prevent overfitting in the model training process, early Stopping is adopted, and as the training round increases, if the test error on the verification set increases, the training is stopped.

2. Financial industry alternative data

The present invention takes insurance and medicaid service Center (CMS) prepared and published insurance financial plan data to implement an example prediction of alternative data. Data was derived from the health and dental program database in the U.S. health insurance market (https:// www.kagger.com/datasets/hhs/health-interior-marktp lace).

3 data preprocessing

3.1. Missing value padding

The missing value is processed by using a front-back average value filling mode, namely the average value of the attribute value of the missing value at the previous moment and the attribute value of the missing value at the next moment is used as the filling value of the missing moment. The fill value is calculated as shown in equation (19). When a plurality of continuous values are missing, the average value of two non-null values adjacent to each other before and after the missing is used for filling.

3.2. Feature selection

Historical studies indicate that historical insurance premium and financial income are important factors affecting the insurance benefits of insurance beneficiaries. Based on past research, in order to train a model with high prediction precision under the condition of not consuming a large amount of computing resources, the invention selects various insurance premiums of insurance beneficiaries in the last year and income levels of the insurance beneficiaries as the characteristics of the prediction model.

3.3. Feature standardization

In a model with a plurality of features, the difference of feature metric units can cause the difference of calculation results, the feature with large scale can play a decisive role, and the feature with small scale can be neglected. In order to eliminate the influence of unit and scale differences among different features, zero-mean standardization is adopted for processing feature data. The method can accelerate the speed of solving the optimal solution by gradient descent. The normalized data mean is 0 and standard deviation is 1, following a standard normal distribution. The calculation formula is shown as (20), wherein n is the sample size, X ^* Is the processed data, X is the raw data,

is the mean of the raw data, δ is the standard deviation of the raw data. The standard deviation is calculated as shown in equation (21).

4. Model hyper-parameters

The invention optimizes the hyperparameter of GRU financial alternative data power model and the hyperparameter of other three comparison algorithms based on Bayesian optimization algorithm, and the optimization iteration number is 30. Wherein, time step is the time step, units is the number of neurons, dense is the number of nodes of the full connection layer, n identifiers tree number, and max depth is the maximum depth of the tree. In addition, the learning rates of the neural networks GRU, LSTM, and MLP are all 0.001, and the training rounds are all 100. The activation function of GRU and LSTM is tanh as shown in equation (4), and the activation function of MLP is a linear rectification function (ReLU) as shown in equation (22).

ReLU(x)＝max(0,x)#(22)

5. Model evaluation index

In order to estimate the prediction accuracy of the model in an all-round way, the invention selects MSE, root Mean Square Error (RMSE), mean Absolute Error (MAE), mean Absolute Percentage Error (MAPE), pearson correlation coefficient (R) and decision coefficient (R) ² ) As an evaluation index of the model. Through the evaluation indexes, the performance of the prediction model in the test set can be clearly seen, including the difference between the observed value and the predicted value and the correlation degree between the observed value and the predicted value. They are expressed by the formulas (18), (23), (24), (25), (26), (27), respectively, where n is the number of samples, y _i Is an observed value, x _i Is a predicted value of the number of the frames,

is y _i Is determined by the average value of (a) of (b),

is x _i Average value of (a).

Prediction of results for alternative types of data

Table 4 shows the premium prediction results of the premium prediction models trained by the four algorithms respectively. The optimal results have been marked in bold. As can be seen from the results, since both LSTM and GRU are improved by RNN, the history information can be effectively learned. Their predictive effect is better than MLP and RF. All the evaluation indexes of the improved GRU based on the invention at two sites are optimal. The prediction accuracy of MLP is higher than that of RF, so that the prediction effect of the neural network in premium prediction is better than that of the traditional machine learning algorithm. Compared with the LSTM algorithm, in the premium prediction based on the Bayesian optimization and the attention mechanism, the MSE is reduced by about 8.3%, the RMSE is reduced by about 3.8%, the MAE is reduced by about 10.9%, the MAPE is reduced by about 12.4%, the R is improved by about 0.2%, and the R is improved by about 0.2% in comparison with the LSTM algorithm ² The improvement is about 0.5%.

TABLE 1 premium prediction results

In order to observe the premium prediction effect of the model more vividly, observation data of a period of time are respectively selected from two sites and compared with prediction data of four algorithms, and a comparison curve graph 4 is obtained. In the comparison graph, the fitting effect of the GRU algorithm based on the Bayesian optimization and attention mechanism is better than that of other algorithms. The LSTM and the GRU have similar performance, and the prediction effect is satisfactory. The prediction curves of the MLP and RF algorithms have many fluctuations, especially in RF, and the prediction effect is not as good as that of GRU and LSTM, which may be related to simpler model structure.

Table 5 summarizes the premium cycle forecast results for the four classes of algorithms on both sites, with the best results being shown in bold. The GRU algorithm based on Bayesian optimization and attention mechanism provided by the invention has the advantages that each evaluation index is optimal, LSTM and MLP are inferior, and RF is worst. Compared with the LSTM algorithm, the GRU based on the Bayesian optimization and attention mechanism has the advantages that the MSE is reduced by about 3.4%, the RMSE is reduced by about 1.8%, the MAE is reduced by about 0.6%, the MAPE is reduced by about 0.5%, the R is improved by about 0.2%, and the R is improved by about 0.2% in the premium period prediction of the NJI site ² The improvement is about 0.4%.

TABLE 2 premium prediction results

Fig. 5 shows a graph comparing predicted values and observed values for four algorithms. Since the GRU and the LSTM can make a decision in the future according to the change rule of the historical time series information, the GRU and the LSTM can better fit the observed value.

Table 6 shows the near premium change prediction performance of the four algorithms under different evaluation indexes, and the optimal result has been marked in bold. From the comparison of the algorithms, four algorithms showed satisfactory results in cost variation prediction, and R of each of them was more than 91%. The GRU based on Bayesian optimization and attention mechanism performs best in all evaluation indexes, the MAE is 0.5555 in NJI site ² The content was 91.27%. The prediction results of LSTM and MLP are similar and have no obvious difference. The results verify that the four algorithms of GRU, LSTM, MLP and RF are all high in premium change predictionAnd the superiority of the improved GRU proposed in this section in cost variation prediction.

TABLE 3 alternative data prediction results

Fig. 6 shows a comparison of the predicted values of alternative data changes versus observed values for the four algorithms in a 1 hour prediction. As can be seen from the figure, the predicted values of GRU and LSTM are closer to the observed value; the predicted values of MLP and RF are easy to fluctuate and have large errors.

The embodiment of the invention also discloses a financial industry alternative data prediction system based on the improved GRU, which comprises the following steps:

For specific limitations of the financial industry alternative data prediction system based on the improved GRU, reference may be made to the above limitations on the financial industry alternative data prediction method, and details are not described herein again. The various modules in the financial industry alternative data prediction system described above may be implemented in whole or in part by software, hardware, and combinations thereof. The modules can be embedded in a hardware form or independent from a processor in the computer device, and can also be stored in a memory in the computer device in a software form, so that the processor can call and execute operations corresponding to the modules.

The invention also discloses a prediction device for predicting the alternative data in the financial industry, and the computer device can be a server. The computer device includes a processor, a memory, a network interface, and a database connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The database of the computer equipment is used for storing data of the resistance equivalent model and the equivalent submodel, and storing equivalent resistance, working resistance and contact resistance obtained in the process of executing calculation. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to realize the financial industry alternative data prediction method.

In one embodiment, a computer-readable storage medium is provided, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, the processor implementing the following steps when executing the computer program: first the input features of the prediction model need to be determined. Secondly, determining each hyper-parameter of the model by a Bayesian optimization algorithm. Subsequently, at the hidden layer of the model, the features are assigned different weights by an attention mechanism. And training by mass data to respectively obtain sensor data and a periodic prediction model. And comparing the prediction result of the model with the observation value by using the test set, and judging whether the end condition is reached according to a Bayesian optimization method. If yes, the model is used for predicting the change trend of the financial alternative data; if not, continuing parameter optimization.

An embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored, where the computer program, when executed by a processor, implements the following steps: first, input features of the prediction model need to be determined. Secondly, determining each hyper-parameter of the model by a Bayesian optimization algorithm. Subsequently, at the hidden layer of the model, the features are assigned different weights by an attention mechanism. And training by mass data to respectively obtain sensor data and a periodic prediction model. And comparing the prediction result of the model with the observation value by using the test set, and judging whether the end condition is reached according to a Bayesian optimization method. If yes, the model is used for predicting the change trend of the financial alternative data; if not, continuing parameter optimization.

The above-described method according to the present invention can be implemented in hardware, firmware, or as software or computer code storable in a recording medium such as a CD ROM, a RAM, a floppy disk, a hard disk, or a magneto-optical disk, or as computer code originally stored in a remote recording medium or a non-transitory machine-readable medium and to be stored in a local recording medium downloaded through a network, so that the method described herein can be stored in such software processing on a recording medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware such as an ASIC or FPGA. It will be appreciated that the computer, processor, microprocessor controller or programmable hardware includes memory components (e.g., RAM, ROM, flash memory, etc.) that can store or receive software or computer code that, when accessed and executed by the computer, processor or hardware, implements the processing methods described herein. Further, when a general-purpose computer accesses code for implementing the processes shown herein, execution of the code transforms the general-purpose computer into a special-purpose computer for performing the processes shown herein.

It will be appreciated by those of ordinary skill in the art that the examples described herein are intended to assist the reader in understanding the manner in which the invention is practiced, and it is to be understood that the scope of the invention is not limited to such specifically recited statements and examples. Those skilled in the art, having the benefit of this disclosure, may effect numerous modifications thereto and changes may be made without departing from the scope of the invention in its aspects.

Claims

1. A financial industry special data prediction method based on improved GRU is characterized by comprising the following steps:

firstly, determining input characteristics of a prediction model;

secondly, determining each hyper-parameter of the model through a Bayesian optimization algorithm;

then, in a hidden layer of the model, different weights are assigned to the features through an attention mechanism;

obtaining sensor data and a period prediction model respectively after a large amount of data training;

comparing the prediction result of the model with the observed value by using a test set, and judging whether an ending condition is reached according to a Bayesian optimization method;

if yes, predicting the change trend of the financial type-II data by using the model; if not, the parameter optimization is continued.

2. The method of claim 1, wherein the method comprises the following steps: in the training process, the Mean Square Error (MSE) is selected as the loss function, as shown in equation (1):

wherein n is the number of samples, y _i Is an observed value, x _i Is a predicted value; updating of the weight parameters is done by an adaptive momentum estimation (Adam) algorithm; the Adam optimizer combines the advantages of RMSProp and AdaGrad algorithms which are good at processing sparse gradients and non-stationary targets, and can obtain good results at a high speed; in order to prevent overfitting in the model training process, early Stopping is adopted, and with the increase of training rounds, if the test error on the verification set rises, the training is stopped.

3. The method of claim 1, wherein the method comprises the steps of: in the financial industry alternative data prediction method, a missing value is processed in a mode of filling a front average value and a rear average value, namely, the average value of an attribute value of the missing value at the previous moment and the attribute value of the missing value at the subsequent moment is used as a filling value of the missing moment; the fill value is calculated as shown in equation (2);

when a plurality of continuous values are missing, the average value of two non-null values adjacent to each other before and after the missing value is used for filling.

4. The method of claim 1, wherein the method comprises the steps of: various insurance premiums of the insurance beneficiaries in the last year and the income level of the insurance beneficiaries are selected as input characteristics of the prediction model.

5. The method of claim 1, wherein the method comprises the following steps: and (3) feature standardization: zero-mean standardized processing of characteristic data is adopted; the mean value of the normalized data is 0, the standard deviation is 1, and the normalized data obeys standard normal distribution; the calculation formula is shown in (3):

wherein n is the sample size, X ^* Is the processed data, X is the raw data,

is the mean of the raw data, δ is the raw data standard deviation;

the standard deviation is calculated as shown in equation (4):

6. the method of claim 1, wherein the method comprises the following steps: optimizing the hyperparameter based on a Bayesian optimization algorithm, wherein the number of optimization iterations is 30; the value range and the final value of the hyper-parameter to be optimized are obtained, wherein time step is time step, units is the number of neurons, dense is the number of nodes of a full connection layer, n estimators trees and max depth is the maximum depth of the trees; in addition, the learning rate of the neural network GRU is 0.001, and the training rounds are 100; the activation function of GRU is tanh.

7. A financial industry alternative data prediction system based on an improved GRU, comprising:

the hyper-parameter determining module is used for determining various hyper-parameters of the model;

8. A prediction device for prediction of financial industry alternative data comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: the processor, when executing the computer program, implements the financial industry alternative data prediction method of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, the computer program characterized in that: the computer program when executed by a processor implements the financial industry alternative data prediction method of claims 1 to 6.