CN115641088A

CN115641088A - Quantum computing-fused approval strategy combination obtaining method, device and medium

Info

Publication number: CN115641088A
Application number: CN202211378544.2A
Authority: CN
Inventors: 赵战营; 马自谦; 范桢
Original assignee: Shanghai Pudong Development Bank Co Ltd
Current assignee: Shanghai Pudong Development Bank Co Ltd
Priority date: 2022-11-04
Filing date: 2022-11-04
Publication date: 2023-01-24

Abstract

The invention relates to an examination and approval strategy combination obtaining method, equipment and a medium fusing quantum computing, wherein the method comprises the following steps: obtaining the approval information of the customers, and segmenting the customer group by adopting an A card scoring model and a random forest model to obtain a rule set comprising a plurality of rule groups; constructing a quantum Bayesian network according to a rule set through a structure learning process and a training process in sequence; and generating guest group combined data by adopting a hyper-heuristic algorithm strategy, taking the guest group combined data as the input of a quantum Bayesian network, taking a preset target function as an optimization target, and obtaining the optimal examination and approval strategy combination after multiple iterations. Compared with the prior art, the method uses the quantum Bayesian network to measure the overall approval rate and the overdue rate, so that the optimized target function is made dominant and the advantages of the quantum Bayesian network are retained. And performing combined optimization solution by using a hyper-heuristic optimization algorithm, so that the combined solution efficiency is improved. And the random forest algorithm is used, so that the passenger group classification can be refined.

Description

Quantum computing-fused approval strategy combination obtaining method, device and medium

Technical Field

The invention relates to the field of quantum finance, in particular to a method, equipment and medium for obtaining an approval strategy combination by fusing quantum computing.

Background

The scheme established by the existing credit card approval strategy combination can be divided into three steps: (1) dividing guest groups and establishing corresponding admission thresholds for different guest groups; (2) under the corresponding admission threshold rule, the historical data is used for measuring and calculating the overall approval rate and overdue rate of the guest group; (3) and outputting the best examination and approval strategy combination in all search results through grid search. The specific contents of each step in the prior art scheme are as follows:

(1) Dividing the guest groups and making corresponding admission thresholds for different guest groups

Firstly, scoring and sequencing all customers by using an A card model, and screening head and tail customer groups of the A card; then, according to the personal basic information characteristics of the customer, such as age, sex, academic calendar and the like, and credit information of the customer, a customer group division rule is made, 5-10 customer groups are divided by using the rules, and each customer group is set with different screening standards, namely an admission threshold.

(2) Measuring and calculating the approval rate and overdue rate of the whole guest group

When the client groups are divided according to the rules, one client accords with a plurality of rules, so that the clients in different client groups have overlapping parts, the calculation of overdue rate and approval rate needs to traverse the data of all the clients for statistics, and repeated calculation caused by the overlapping parts is eliminated.

(3) Outputting the best examination and approval strategy combination in all search results through grid search

Continuously adjusting a passenger group dividing mode and an admission threshold through a grid searching mode to form different examination and approval strategy combinations; but different approval strategy combinations need to respectively measure and calculate the approval rate and overdue rate of the whole corresponding passenger groups; and finally, comparing the overall approval rate and overdue rate of the corresponding guest groups of all approval strategy combinations through grid search, outputting the optimal combination, and taking the combination as a credit card approval strategy combination.

Chinese patent application No. CN202111447176.8 discloses a business approval process configuration method, apparatus, computer device and storage medium, the method comprising: acquiring dimension information corresponding to different preset dimensions respectively, wherein the preset dimensions comprise at least one of a source channel dimension, a product channel dimension, an account attribute dimension and a resource setting dimension; combining the dimension information based on different preset dimensions to obtain a plurality of service scenes; for each service scene, determining a plurality of target service nodes matched with the corresponding service scene; for each service scene, respectively configuring a node time sequence among a plurality of target service nodes corresponding to the corresponding service scene; and combining the target service nodes according to the node time sequence configured with the corresponding service scene to obtain service approval processes corresponding to the service scenes respectively. The method can improve the working efficiency of accessing the new node into the workflow. However, this patent does not disclose an acquisition method for a combination of credit card approval policies.

In summary, the existing method for acquiring the combination of the examination and approval policies has the following disadvantages:

(1) The number of the divided guest groups is limited, the rule granularity is too coarse, and the number of the divided guest groups is not more than 10 generally, so that a fine examination and approval strategy is difficult to realize, risk customers cannot be accurately identified, and the overall examination and approval rate is low and the risk rate is high.

(2) The time for evaluating the overall approval rate and the overdue rate of the guest group combination is long (the calculation speed of the target function is low), each approval strategy combination needs to carry out measurement and calculation of the overall approval rate and the overdue rate of the guest group combination once, namely, the target function is measured and calculated once, the historical data condition of a corresponding individual needs to be traversed once each calculation, the number of times the same as that of the target group needs to be traversed once each calculation, and the evaluation time is long.

(3) The traversal range is huge, the time required by the existing scheme is too long, if 150 high-risk and low-risk rules (namely 150 optional customer groups) are specified, and whether the rules (the customer groups) are used or not is selected from the rules, 2150 possibilities exist, the possibilities are huge, and in actual calculation, the optimal solution cannot be obtained by using a traversal search mode.

Disclosure of Invention

The invention aims to overcome the defects in the prior art and provide an approval strategy combination obtaining method, equipment and medium which can measure and calculate the approval rate and the overdue rate of the whole passenger group without traversing the historical data of all persons in the passenger group and can perform fine classification on the passenger group and integrate quantum computing.

The purpose of the invention can be realized by the following technical scheme:

one aspect of the invention provides a quantum computing-fused approval strategy combination obtaining method, which comprises the following steps:

obtaining client approval information, and segmenting a client group by adopting an A card scoring model and a random forest model to obtain a rule set comprising a plurality of rule groups;

according to the rule set, sequentially performing a structure learning process and a training process to construct a quantum Bayesian network;

and generating guest group combined data by adopting a hyper-heuristic algorithm strategy, taking the guest group combined data as the input of a quantum Bayesian network, taking a preset target function as an optimization target, and obtaining an optimal examination and approval strategy combination after multiple iterations.

As a preferred technical solution, the obtaining of the rule set includes the following steps:

acquiring the approval information including approval results, overdue or not and a client identification field, and marking the clients with risks higher than a threshold value and lower than the threshold value as approval refusal and approval passing according to a preset client risk evaluation rule, wherein the client risk evaluation rule is determined by a card A scoring model;

taking the approval information as a sample set, training a random forest model, and extracting a rule set from a large number of generated decision trees;

and dividing the rule set into a high-risk rule group and a low-risk rule group, and marking the corresponding guest groups as refused examination and approval and passing examination and approval respectively.

As a preferred technical solution, the structure learning process includes the following steps:

judging whether the rule groups are started as nodes of a classic Bayesian network or not by using independence test, and if the rule groups are independent, not setting edges among the nodes;

establishing and/or removing and/or adjusting edges between nodes according to credit card strategy approval experience of a service expert;

respectively mapping the node states to corresponding superposition states according to the state number of each node of the classic Bayesian network;

the process of mapping the probabilities to the stacked state amplitudes is adjusted to a rotating and controlled rotating gate, completing the structure learning process.

As a preferred technical solution, the training process includes:

exhaustively exhausting the opening state combination of the multiple rule groups to serve as a training set, and taking the training set as the input of a classical Bayesian network to obtain a probability distribution table of each node;

and manufacturing a quantum circuit for performing addition, subtraction, multiplication and division operations on the probability amplitude of the quantum state and the probability amplitude, and assigning values to the quantum probability amplitude of the uncertain nodes according to the auxiliary state probability amplitude to obtain the quantum Bayesian network.

As a preferred technical solution, the generation process of the guest group combination data includes the following steps:

and converting the combined data into quantum data by using a variational embedding method, normalizing the data to a [0, pi/2 ] interval by using classical data as a parameter of a quantum line, performing quantum writing as a partial parameter of the quantum line, and encoding data by using a fixed variational line to obtain the guest group combined data.

As a preferred technical scheme, in the hyper-heuristic algorithm, the bottom-layer heuristic algorithm comprises a tabu search algorithm and a simulated annealing algorithm, and the high-layer strategy comprises an identification and calculation bottleneck strategy and a switching bottom-layer heuristic algorithm.

As a preferred technical scheme, the objective function is the difference between the approval rate and the overdue rate of the general customer group.

As a preferred technical solution, the quantum bayesian network includes a high-risk rule sub-network and a low-risk rule sub-network, an output of the high-risk rule sub-network is used as an input of the low-risk rule sub-network, and an output of the low-risk rule sub-network is an approval rate of the entire customer group under the current customer group combination.

In another aspect of the present invention, an electronic device is provided, which includes one or more processors, a memory, and one or more programs stored in the memory, where the one or more programs include instructions for performing the method for combined approval strategy acquisition fused with quantum computation.

In another aspect of the present invention, a computer-readable storage medium is provided, which includes one or more programs for execution by one or more processors of an electronic device, the one or more programs including instructions for performing the approval policy combination obtaining method of fused quantum computation described above.

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

(1) The method comprises the steps of obtaining a rule set comprising a large number of rule groups by adopting a random forest algorithm, refining customer group classification can be carried out, training a random forest model by using historical approval data and historical expected data, disassembling a plurality of decision trees in the random forest to obtain the rule set, and classifying the customer groups by using the rule set, so that the number of the customer groups is greatly increased, and the long tail effect of a plurality of small customer groups is repeatedly utilized.

(2) The method has the advantages that the approval rate and the overdue rate of the whole customer group are measured and calculated by using the quantum Bayesian network, so that the optimized objective function is dominant, the advantages of the quantum Bayesian network in the aspects of calculation speed, storage, energy consumption and topological structure are retained, the resource consumed by the quantum Bayesian network is far smaller than that of a classical computer, the quantum Bayesian network has obvious exponential acceleration, and for the classical Bayesian network, only gates with probability changes need to be modified when the probability distribution of nodes in the classical Bayesian network is changed, and the structural relation among the quantum gates does not need to be changed when the node relation is not changed;

(3) Aiming at the difficulty that the probability of rule combination is high when a hyperheuristic optimization algorithm is adopted to solve the credit card approval strategy combination optimization problem, the invention does not need to traverse all rule combination possibilities, carries out optimization calculation by adopting the hyperheuristic optimization algorithm, and can quickly and efficiently find out satisfactory rule combination.

Drawings

FIG. 1 is a schematic flow chart of an approval policy combination obtaining method of fusion quantum computing in an embodiment;

FIG. 2 is a diagram illustrating a quantum Bayesian network according to an embodiment;

fig. 3 is an electronic circuit diagram of a quantum bayesian network in an embodiment.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be obtained by a person skilled in the art without any inventive step based on the embodiments of the present invention, shall fall within the scope of protection of the present invention.

Example 1

Fig. 1 is a schematic flow chart of the examination and approval policy combination obtaining method fusing quantum computing according to this embodiment. The service scene used by the method is a credit card incoming approval strategy combination scene, different strategy combinations are optimized and iterated by measuring and calculating approval rates and overdue rates of different strategy combinations, and a higher (approval rate-overdue rate) strategy combination is obtained. The method comprises the following steps:

and S01, segmenting the guest groups through A card scoring and a random forest algorithm.

And S02, constructing a quantum Bayesian network for evaluating the overall approval rate and overdue rate of the guest group combination. The method mainly comprises the following steps:

and step S021, learning the quantum Bayesian network structure.

And S022, training a quantum Bayesian network.

And S03, optimizing by using a hyper-heuristic algorithm fused with quantum computing, selecting a more optimal customer group combination, and finally obtaining an approximate optimal solution of the combination through iteration. The method mainly comprises the following steps:

and S031, generating the guest group combination intelligently.

And step S032, calculating key indexes by using a quantum Bayesian network.

Step S033, comparing the iteration results, and selecting a better guest group combination.

Wherein, step S01 mainly includes two aspects:

(1) Determining a client risk scale through an A card scoring model, using the risk scale as a threshold for risk admission, and directly marking the head and tail high-risk and low-risk groups as approved or rejected groups;

(2) Segmenting the guest groups by using a random forest algorithm:

a) Performing model training by using a random forest algorithm and taking historical approval data and overdue data as sample sets, so that a large number of decision trees generated by the random forest model extract rule sets in the decision trees, wherein each rule set is used for screening out a guest group conforming to a corresponding rule;

b) Analyzing each rule set so as to classify each rule set into a high-risk rule set and a low-risk rule set;

c) And aiming at the guest groups corresponding to the high-risk rule sets and the low-risk rule sets, using a guest group admission judgment method, namely, if the guest group corresponds to the low-risk rule set, all the guest groups pass the approval, and if the guest group corresponds to the high-risk rule set, all the guest groups refuse the approval.

d) Here, it should be noted that: in formulating rules in the risk rule set (which may be understood as when the customer groups are grouped), rule thresholds (thresholds) have been considered, e.g., "annual income >20 ten thousand", which rules have included a threshold of "20 ten thousand", which index threshold has been considered in the customer group rules.

e) The business scenario credit card approval policy combination in this embodiment is whether to enable a policy combination of each rule set, that is, whether to approve or reject a guest group combination corresponding to the rule set.

In step S02, the business scenario corresponding to the scheme in this embodiment acquired in step S01 is a credit card approval policy combination, and the quantum bayesian network has a function of rapidly measuring and calculating the approval rate and risk level of a specific guest group combination in the scenario. Each guest group corresponds to a rule set, and the combination of guest groups is a rule set combination. Therefore, the approval rate and the overdue rate of the general customer base are evaluated by adopting the quantum Bayesian network, and the node in the Bayesian network is to determine whether the customer base is enabled or not, namely, whether the rule set is enabled or not.

The Bayesian network is a probability graph model, is one of the most effective analysis models for the current uncertainty and probabilistic problems, can well represent a complex random system containing various condition control factors, and performs computational analysis and decision.

However, the computational complexity and data size of the bayesian network per se increase with the node number of the model and the total state number index level of the nodes, so that not only efficient problem calculation and solution cannot be realized, but also a large-scale multi-node model cannot be supported.

The quantum Bayesian network realized by introducing quantum computing not only can perfectly inherit the existing functions of the Bayesian network, but also can give full play to the advantages of quantum computing, greatly reduce the computational complexity, support a larger-scale model, expand more functions and develop more application fields on the basis.

A classical Bayesian network model for the business scenario is generated using classical computational assistance, and then a quantum Bayesian network is developed using quantum computing techniques.

For step S021, the essence of the quantum bayesian network structure learning is to check the independence relationship between nodes, i.e., the independence relationship between rule sets. Generally speaking, the relationship between nodes is: parent-child relationships (the previous node determines the next node), sibling relationships (two nodes are controlled by the same node), and unrelated relationships (no direct influence relationship between two nodes). The structure learning (i.e., the process of quantum bayesian structure learning) using the training data set includes:

and step S0211, detecting the direct independence relation of each node through independence test, and if the nodes are independent, not setting edges among the nodes.

And S0212, establishing, eliminating and adjusting edges between nodes according to the credit card strategy approval experience of the service expert, and confirming the network structure.

And S0213, respectively mapping the node states to corresponding superposition states based on the state number of each node of the classical Bayesian network.

Step S0214, the process of mapping the probability to the superposition state amplitude is adjusted to a rotating and controlled revolving gate, and this process may need to add additional auxiliary bits, and meanwhile, this process can be split into multiple CCNOT gates and revolving gates.

Step S0215, the judgment of the node state number and the father node number, the construction of the corresponding superposition state and the estimation of the full-network quantum bit number can be realized by programming, and further, the automatic quantum version conversion of the classic Bayesian network is realized.

An example of a quantum bayesian network comprising 8 ruleset nodes is shown in fig. 2, and the output result is the approval rate of the whole guest group under the combination of the guest groups. The network contains 2 sub-networks, a high risk rules sub-network and a low risk rules sub-network, respectively. The output result of the high-risk rule network has the approval veto power, so that the output result of the high-risk rule network is used as the input of the low-risk network, two sub-networks are connected in series, and the total approval passing rate is finally calculated. And similarly, constructing the Bayesian network with the same structure for calculating the overall risk probability.

For step S022, the following is mainly included:

step S0221, a training set structure is adopted, the possibility of rule combination is sampled, for example, 2 ten thousand possibilities are sampled, the approval rate and overdue rate of the guest group corresponding to the rule in each possibility are counted to obtain a table 2 with 2 ten thousand pieces of data, and the Bayesian network is trained by using the data set.

Step S0222, under the condition that the Bayesian network structure is known, the training data in the step S0221 is used for learning the probability distribution table of each node.

Step S0223 develops a quantum circuit for performing addition, subtraction, multiplication, and division on the probability amplitude and the probability amplitude of the quantum state, and a quantum circuit for performing addition, subtraction, multiplication, and division on the probability amplitude and the probability amplitude of the quantum state. The quantum bayesian network electronics diagram is depicted in fig. 3.

And S0224, performing probability formula operation by combining the auxiliary state probability amplitude, and assigning values to the quantum probability amplitude of the uncertain node.

In the training step of the quantum Bayesian network, on one hand, a probability distribution table of each node is obtained through sample data, and on the other hand, the Bayesian network is developed and converted into the quantum Bayesian network through a circuit.

For the step S03, the basic idea of the step is to adopt a mixed quantum-classical algorithm as an algorithm framework for solving the optimization problem in the step S03, and the algorithm framework aims to solve the specific optimization problem by exerting the capability of the quantum computer as much as possible while relying on the power of the classical computer. The role of the classical computing part is mainly twofold: (1) the classical calculation is used for serially connecting service data and a quantum algorithm, recording an optimal result, determining an optimal path and identifying an optimal rule combination; (2) and the input and the output of the quantum algorithm and the classical operational planning algorithm are connected in series to form a closed loop of the classical algorithm and the quantum algorithm, and the output data of the quantum algorithm is utilized to optimize and solve the business scene problem in combination with the classical operational planning algorithm. In the scheme, a meta-heuristic optimization algorithm is selected as a classic operation algorithm in the algorithm framework, and a high-risk rule combination and a low-risk rule combination are solved by combining the output result of the quantum Bayesian network. The definition of the hyper-heuristic algorithm is as follows: the hyper-heuristic provides some High-Level Strategy (HLS) to obtain a new heuristic by manipulating or managing a set of Low-Level Heuristics (LLH). The method comprises the following steps:

step S031, the intelligent generation of the guest group combination includes:

step S0311, the heuristic algorithm strategy is formulated:

the bottom layer heuristic algorithm in the hyper-heuristic algorithm comprises the following steps:

(1) Tabu search algorithm: and a promotion area is set, so that the solving speed is promoted, but the local optimal solution is easy to fall into.

(2) And (3) simulating an annealing algorithm: the computing mechanism is suitable for jumping out of the local optimal solution, but the computing speed is low;

the high-level strategy in the hyperheuristic algorithm is as follows:

(1) Identifying a computing bottleneck: and identifying whether the current computation bottleneck is low in computation speed or falls into a local optimal solution according to the state of computation iteration.

(2) Switching a bottom layer heuristic algorithm: and when the approximation speed of the target function is pursued, switching to a tabu search algorithm, and when the iteration falls into a local optimal solution, switching to a simulated annealing algorithm.

Step S0312, quantum data conversion, using variational embedding method to convert the combined data into quantum data, using classical data as the parameter of quantum line, normalizing the data to [0, pi/2 ] interval, then using the normalized data as partial parameter of quantum line to carry out quantum writing, and using fixed variational line to encode data.

Step S032, calculating a key index (target function) by using a quantum Bayesian network. The specific objective function is determined through the business objective, and the form of the objective function is adjusted through the equivalent change in the design mathematics, so that the function can be efficiently calculated on a quantum computer through measurement. In the scheme, the quantum Bayesian network obtained in the step S02 is used as a target function for measurement and calculation, and the specific function is of a general customer group (approval rate-overdue rate);

step S033, comparing the target functions corresponding to the different iteration results, and retaining the better guest group combination.

By step S03, the approximately optimal combination of the credit card approval policies is finally obtained.

The examination and approval strategy combination obtaining method fusing quantum computing in the embodiment has the following advantages:

1. advantage of using quantum Bayesian network to measure and calculate approval rate and overdue rate of whole customer group

(1) Making optimized target function explicit

Aiming at the problem of low overdue rate testing speed of the overall passenger group approval rate, the method and the system have the advantages that the Bayesian network model is constructed for measurement and calculation, so that the optimized objective function is dominant.

The trained Bayesian network can calculate the approval rate and the overdue rate of the target crowd, so that the step of traversing all people's historical data is omitted, and the calculation speed can be rapidly increased.

(2) Self-advantages of quantum Bayesian networks

(1) Advantage of calculating speed

A quantum bayesian network is a lossless algorithm input to a quantum computer, with no specific computation in between. The Bayesian network multiply-divide in a classical computer requires only one quantum gate for resolution in the corresponding quantum line. The very large number of bits required by the bottom layer of the computer only needs to correspond to the angle of the quantum gate at the quantum computer level. And the quantum Bayesian network has obvious exponential acceleration when being applied to the processes of global variable detection and direction derivation. For a chain network, the quantum Bayesian network has the acceleration advantage of exponential calculation when calculating the global nodes. For the convergent network, the quantum bayesian network has the advantage of exponential storage when calculating the global or last node.

(2) Storage and power consumption advantages of quantum computers

The amplitude encoding work of a quantum computer can encode 2n quantum states on n qubits, i.e. states in which 2n qubits can be stored, the amount of information stored being doubled for each additional qubit. This makes the quantum bayesian network consume far less resources on quantum computers than on classical computers.

(3) Topological advantage

When the classical Bayesian network changes the probability distribution of the nodes therein, only the gates with probability changes need to be modified, and the structural relationship among the sub-gates does not need to be changed when the node relationship does not change. In the classical computer, the bayesian network needs to be reconstructed and the calculation started again each time the probability distribution of the nodes is changed. The topological structure of the quantum computer naturally fits the Bayesian network, so that a new result can be obtained more quickly when nodes are transformed.

In summary, according to the scheme in this embodiment, when the approval rate and the overdue rate of the entire guest group are measured, the overall speed can be increased by more than 10 times.

2. Advantages of hyper-heuristic optimization algorithm in solving credit card approval strategy combination optimization problem

Aiming at the difficulty that the rule combination possibility is high, the scheme can not traverse all rule combination possibilities, and a hyperheuristic optimization algorithm is adopted to carry out optimization calculation: the hyper-heuristic optimization algorithm is a heuristic optimization algorithm with a top-level strategy, and can identify optimization calculation bottlenecks in different stages, so that different targeted optimization algorithms are called, and finally, a satisfactory rule combination is quickly and efficiently found.

3. A random forest algorithm is used for obtaining a large number of rule sets so as to perform refined guest group classification

The method includes the steps that a random forest model is trained by using historical approval data and historical expected data, a rule set is obtained by disassembling multiple decision trees in the random forest, and the rule set is used for carrying out customer group classification, so that the number of customer groups is greatly increased, and the long tail effect of multiple small customer groups is repeatedly used.

Example 2

The present embodiment provides an electronic device, comprising one or more processors, memory, and one or more programs stored in the memory, the one or more programs including instructions for performing an approval policy combination retrieval method of fused quantum computing as described in embodiment 1.

Example 3

The present embodiments provide a computer-readable storage medium comprising one or more programs for execution by one or more processors of an electronic device, the one or more programs including instructions for performing the combined approval strategy acquisition method of fused quantum computing as described in embodiment 1.

While the invention has been described with reference to specific embodiments, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. An approval strategy combination obtaining method fused with quantum computing is characterized by comprising the following steps:

and generating guest group combined data by adopting a hyper-heuristic algorithm strategy, taking the guest group combined data as the input of a quantum Bayesian network, taking a preset objective function as an optimization objective, and obtaining the optimal examination and approval strategy combination after multiple iterations.

2. The method for obtaining the combination of approval strategies fused with quantum computing according to claim 1, wherein the obtaining of the rule set comprises the following steps:

acquiring the approval information including approval results, overdue and client identification fields, and marking the clients with higher risks than a threshold value and lower risks than the threshold value as refusal to approve and pass approval according to a preset client risk evaluation rule, wherein the client risk evaluation rule is determined by a card A scoring model;

3. The quantum computing-fused approval strategy combination obtaining method as claimed in claim 1, wherein the structure learning process comprises the following steps:

the process of mapping the probabilities to the superimposed state amplitudes is adjusted to a rotating and controlled rotating gate, completing the structure learning process.

4. The quantum computing-fused approval strategy combination obtaining method as claimed in claim 1, wherein the training process comprises:

exhaustively exhausting the opening state combination of the rule groups to serve as a training set, and taking the training set as the input of a classic Bayesian network to obtain a probability distribution table of each node;

5. The quantum computing fused approval strategy combination obtaining method as claimed in claim 1, wherein the generation process of the guest group combination data comprises the following steps:

and converting the combined data into quantum data by using a variational embedding method, normalizing the data to a [0, pi/2 ] interval by using the classical data as a parameter of a quantum line, writing the data into a quantum by using the classical data as a partial parameter of the quantum line, and acquiring the guest group combined data by using fixed variational line coded data.

6. The method as claimed in claim 1, wherein the heuristic algorithms at the bottom include tabu search algorithm and simulated annealing algorithm, and the heuristic algorithms at the top include identifying computational bottleneck policy and switching the heuristic algorithms at the bottom.

7. The method as claimed in claim 1, wherein the objective function is a difference between an approval rate and an overdue rate of the tenant group.

8. The method as claimed in claim 1, wherein the quantum Bayesian network comprises a high-risk rule subnetwork and a low-risk rule subnetwork, an output of the high-risk rule subnetwork is an input of the low-risk rule subnetwork, and an output of the low-risk rule subnetwork is an approval rate of the entire customer group under the current customer group.

9. An electronic device comprising one or more processors, memory, and one or more programs stored in the memory, the one or more programs including instructions for performing the approval policy combination retrieval method of fused quantum computation of any of claims 1-8.

10. A computer-readable storage medium comprising one or more programs for execution by one or more processors of an electronic device, the one or more programs including instructions for performing the approval policy combination retrieval method of fused quantum computation of any of claims 1-8.