An agent-based customer channel migration behavior simulation method and related device

Abstract

This invention relates to the field of computer data processing technology, and in particular to a method and related apparatus for simulating customer channel migration behavior based on intelligent agents. The method includes: simulation initialization, in which a channel switching resistance coefficient is introduced; service allocation and revenue release; channel satisfaction calculation and customer satisfaction state classification; channel utility calculation and selection; dynamic investment amount adjustment; and iterative judgment until the end of the simulation. This invention simulates the "static friction" effect of state transitions by introducing a channel switching resistance coefficient, avoiding large-scale invalid migrations and computational oscillations in the system under small utility differences, improving the convergence speed of the simulation engine, and solving the technical problem of frequent oscillations in nonlinear systems.

Description

Technical Field

[0001] This invention relates to the field of computer data processing technology, and in particular to a method and related apparatus for simulating customer channel migration behavior based on intelligent agents. Background Technology

[0002] With the explosive growth of financial data, simulating the evolution of individual behavior in complex market environments using simulation technology has become a key research direction. In the multi-channel service system of banks, how to construct high-precision agent simulation models to simulate customer migration behavior has significant scientific and applied value for understanding the evolutionary laws of complex systems.

[0003] Existing simulation techniques have the following limitations when dealing with such large-scale intelligent agent interactions: 1. Simulated decision-making logic is overly idealized, lacking characterization of individual "decision friction": Existing simulation methods (such as utility maximization models based on Markowitz theory) typically assume that agents are perfectly rational, meaning that agents will migrate as long as the utility of the target channel is slightly higher than that of the current channel. However, in the real physical world and digital systems, individual behavior exhibits significant "path dependence" and "migration costs." Existing models fail to introduce effective constraint operators to quantify this migration resistance, resulting in low accuracy of simulation results when simulating "behavioral lag" and "system stickiness" in real environments, with simulation evolution curves often deviating significantly from real data.

[0004] 2. The simulation feedback mechanism is too simplistic to achieve recursive simulation of "nonlinear psychological fluctuations": Existing technologies typically use instantaneous satisfaction as the decision variable, neglecting the cumulative effect and recursive evolution of the agent's psychological state over time. In complex system simulations, individual decisions are often influenced by the weighted effect of historical experience (i.e., the habituation effect). Existing linear decay models or static probability models struggle to accurately simulate this complex psychological feedback loop, leading to oscillations or non-convergence issues in long-term simulations.

[0005] 3. Lack of multi-objective decision-making dimensions makes it difficult to simulate the fine-grained allocation of funds across multiple nodes: Existing simulation schemes often simplify channel migration into a discrete "0-1" selection problem, meaning that the agent can only choose a single node at any given time. However, in real multi-channel coupled systems, funds often exhibit a cross-node asset portfolio distribution. Current technologies lack computational models for dynamically allocating weights to funds across different nodes, failing to accurately reproduce the continuous evolution of the "fund flow" within the system, resulting in insufficient granularity in the simulation models.

[0006] 4. Lack of real-time response mechanism for dynamic intervention factors: Existing simulation models often struggle to introduce external dynamic intervention variables such as "marketing hedging" in real time during simulation operation after initialization. Due to the lack of corresponding hedging operators and compensation mechanisms, the model cannot quantitatively assess the offsetting effect of external input signals on individual migration thresholds, limiting the application effectiveness of the simulation system in strategy stress testing scenarios.

[0007] Therefore, how to construct a simulation model that integrates migration resistance determination, recursive satisfaction calculation, and dynamic asset portfolio allocation to improve the accuracy, robustness, and system fit of complex individual migration behavior simulation is a technical challenge that urgently needs to be solved in the field of computer simulation. Summary of the Invention

[0008] The purpose of this invention is to provide a simulation method and related apparatus for customer channel migration behavior based on intelligent agents, aiming to solve at least one of the technical problems existing in the simulation of individual migration decisions, such as system oscillation and non-convergence, lack of physical characteristics of state transitions, and coarse granularity of asset allocation. This invention constructs a first-order recursive smoothing filter (corresponding to a satisfaction decay model) and an asymmetric state transition resistance operator (corresponding to step resistance). ) and the dynamic weight allocation matrix based on utility ratio (corresponding to This enables accurate simulation of individual state evolution in complex topological environments.

[0009] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a customer channel migration behavior simulation method based on an intelligent agent, comprising: Perform initial configuration of the simulation environment and the state of the client intelligent agent; The service level is assigned to each customer agent for this period simulation. After the current period expires, the actual rate of return of the product is released by each channel. The return deviation is calculated based on the actual rate of return. For each customer agent, calculate the current period's real-time satisfaction based on the current service level, revenue deviation, and investment amount; calculate the customer's average satisfaction score with the channel based on the current period's real-time satisfaction score; and determine the customer's satisfaction status based on the customer's average satisfaction score with the channel. Based on the expected rate of return, channel fee rate, and current real-time satisfaction rate in the initial configuration of the simulation environment, calculate the utility value of each channel for each customer agent; based on the customer satisfaction status of each customer agent, the utility value of each channel, and the channel switching resistance coefficient in the initial configuration of the simulation environment. The decision results are obtained; the decision results include the channel selection for the next simulation and the allocation ratio of investment funds for each channel; Each customer agent simulates and obtains the next period's investment amount based on the current period's customer satisfaction status and decision results; if the next period's investment amount drops to zero or is lower than the preset minimum investment threshold, the corresponding customer agent will be removed from the simulation environment. The simulation is iterated until the termination condition is met, and the simulation results of customer migration evolution data under each channel strategy are obtained.

[0010] A further improvement of this invention is that the initial configuration step of the simulation environment and customer agent state specifically includes: constructing a heterogeneous customer agent cluster, generating initial channel selection, setting channels, setting time and space, setting decision triggering parameters, and initializing the marketing incentive pool; wherein, the decision triggering parameters include the channel switching resistance coefficient. The channel switching resistance coefficient Used to quantify the system friction faced by a client intelligent agent when changing connection nodes; Building a heterogeneous customer intelligent agent cluster includes: generating Each customer agent is randomly and evenly assigned risk attitudes to customers. The initial channel selection can be generated as follows: use only channel 1 or channel 0, or select to use both channels and randomly assign customer channels. Setting up channels includes: for each channel, setting the channel fee rate, service level, and the attributes of the financial products offered on that channel; Setting time and space includes: setting the social radius and setting the maximum simulation time T; Setting decision trigger parameters includes: setting channel utility threshold and setting satisfaction habit factor. and setting the channel switching resistance coefficient Channel switching resistance coefficient This is used to quantify the operational costs and psychological inertia faced by customers when adjusting their asset allocation ratios. Different levels of customer satisfaction correspond to different resistance coefficients. Initializing the marketing incentive pool includes: setting the external intervention signals that can be invoked by each channel and their corresponding utility compensation values. .

[0011] A further improvement of this invention is that: in the step of calculating the real-time satisfaction level for each customer agent based on the service level, revenue deviation, and investment amount for the current period, during simulation time... After the customer intelligent agent completes the channel interaction, the real-time satisfaction rate for the current period is calculated based on the multi-dimensional input components. :

[0012] in, For the amount of investment, For service level weighting, For profit deviation; All of these are feature weights, and their sum equals 1; This is the attenuation factor.

[0013] A further improvement of this invention is that: in the step of calculating the average customer satisfaction value for the channel based on the current period's real-time satisfaction level, the average satisfaction value is:

[0014] in, This represents the average customer satisfaction with the channel over simulation time t. This represents the average customer satisfaction with the channel at simulation time t-1; Satisfaction habit factor; In the step of determining customer satisfaction status based on the average customer satisfaction score for the channel, the calculated... With two preset thresholds Compare and determine customer satisfaction levels: like The customer satisfaction status is: very satisfied; like Customer satisfaction status: moderately satisfied; like The customer satisfaction status is: Unsatisfied.

[0015] A further improvement of this invention lies in: calculating the utility value of each channel for each customer agent based on the expected rate of return, channel fee rate, and current real-time satisfaction rate in the initial configuration of the simulation environment; and calculating the utility value of each channel based on the customer satisfaction status of each customer agent, the utility value of each channel, and the channel switching resistance coefficient in the initial configuration of the simulation environment. The steps to obtain the decision result specifically include: The basic utility value for each channel is calculated using the following function:

[0016] in, Indicates the expected return gain. This represents the rate loss operator. To represent average satisfaction, a, b, and c are the weighting coefficients for revenue, fee rate, and satisfaction, respectively; obtain the marketing incentive compensation factors issued to customers through each channel. The basic utility value is adjusted to obtain the adjusted utility value. :

[0017] Calculate the adjusted utility difference between alternative channels and currently held channels. When the difference in the modified utility value Greater than the channel switching resistance coefficient At that time, the fund allocation coefficient is dynamically calculated:

[0018] in, This indicates that the customer's intelligent agent will be deployed to the channel in the next simulation cycle. Resource weight; N is the total number of channels; If the fund allocation coefficient of a certain channel If the funds fall below a preset threshold, the funds in that channel will be cleared, and the system will revert to a single-channel model.

[0019] A further improvement of this invention is that: the step of simulating and obtaining the next period's investment amount based on the current period's customer satisfaction status and decision results; and removing the corresponding customer agent from the simulation environment if the next period's investment amount drops to zero or is lower than a preset minimum investment threshold, specifically includes: First, the customer agent for the next simulation cycle is determined based on the current customer satisfaction status and decision-making results. Total investment amount For customers who are very satisfied, the total investment amount in the next period will fluctuate upwards according to the normal distribution based on the initial distribution; for customers who are moderately or unsatisfied, the total investment amount in the next period will be reduced proportionally according to their satisfaction score. The client intelligent agent in the next simulation cycle Total investment amount According to the capital allocation coefficient Alternatively, the existing proportions can be allocated to various selected channels to achieve dynamic adjustments to the asset portfolio.

[0020] Secondly, the present invention provides a customer channel migration behavior simulation device based on an intelligent agent, comprising: The initialization module is used to perform the initial configuration of the simulation environment and the state of the client intelligent agent; The service allocation and revenue release module is used to allocate service levels to each customer agent for the current period simulation. After the current period expires, the actual rate of return of the product is released by each channel. The revenue deviation is calculated based on the actual rate of return. The channel satisfaction calculation and customer satisfaction status classification module is used to calculate the current period's real-time satisfaction for each customer agent based on the current period's service level, revenue deviation, and investment amount; calculate the customer's average satisfaction score with the channel based on the current period's real-time satisfaction score; and determine the customer's satisfaction status based on the customer's average satisfaction score with the channel. The channel utility calculation and selection module is used to calculate the utility value of each channel for each customer agent based on the expected rate of return, channel fee rate, and current real-time satisfaction rate in the initial configuration of the simulation environment; and to calculate the utility value of each channel based on the customer satisfaction status of each customer agent, the utility value of each channel, and the channel switching resistance coefficient in the initial configuration of the simulation environment. The decision results are obtained; the decision results include the channel selection for the next simulation and the allocation ratio of investment funds for each channel; The dynamic investment amount adjustment module is used by each customer agent to simulate and obtain the investment amount for the next period based on the current customer satisfaction status and decision results. If the investment amount for the next period drops to zero or is lower than the preset minimum investment threshold, the corresponding customer agent will be removed from the simulation environment. The loop determination module is used for iterative loop simulation until the termination condition is met, and obtains the simulation results of customer migration evolution data under each channel strategy.

[0021] A further improvement of this invention is that: in the step of calculating the real-time satisfaction level for each customer agent based on the service level, revenue deviation, and investment amount for the current period, during simulation time... After the customer intelligent agent completes the channel interaction, the real-time satisfaction rate for the current period is calculated based on the multi-dimensional input components. :

[0022] in, For the amount of investment, For service level weighting, For profit deviation; All of these are feature weights, and their sum equals 1; This is the attenuation factor.

[0023] Compared with the prior art, the present invention has the following beneficial effects: This invention provides a simulation method for customer channel migration behavior based on intelligent agents, comprising: initial configuration of the simulation environment and customer intelligent agent states; assigning service levels to each customer intelligent agent for the current period simulation, and calculating the actual rate of return of each channel's product releases after the current period expires; calculating the revenue deviation based on the actual rate of return; calculating the current period's immediate satisfaction for each customer intelligent agent based on the current period's service level, revenue deviation, and investment amount, and calculating the average customer satisfaction value for the channel based on the current period's immediate satisfaction value; determining the customer satisfaction status based on the average customer satisfaction value for the channel; calculating the utility value of each channel for each customer intelligent agent based on the expected rate of return, channel fee rate, and current period's immediate satisfaction value in the initial configuration of the simulation environment; and calculating the channel switching resistance coefficient based on the customer satisfaction status of each customer intelligent agent, the utility value of each channel, and the channel switching resistance coefficient in the initial configuration of the simulation environment. The simulation yields decision results, including the channel selection for the next simulation period and the allocation ratio of investment funds for each channel. Each customer agent simulates and obtains the next period's investment amount based on the current customer satisfaction status and decision results. If the next period's investment amount drops to zero or falls below a preset minimum investment threshold, the corresponding customer agent is removed from the simulation environment. The simulation is iterated until a termination condition is met, yielding simulation results of customer migration evolution data under each channel strategy. This invention introduces a channel switching resistance coefficient. The "static friction" effect of state transitions was simulated, which avoided large-scale invalid migrations and computational oscillations in the system under small differences in utility, improved the convergence speed of the simulation engine, and solved the technical problem of frequent oscillations in nonlinear systems.

[0024] Furthermore, by introducing a first-order recursive smoothing filter to calculate the average satisfaction, this invention effectively suppresses data noise caused by instantaneous gain fluctuations, making the agent's state evolution more consistent with real physical inertial characteristics and improving the numerical stability of the simulation system.

[0025] Furthermore, this invention utilizes a piecewise nonlinear utility function to quantify the coupling relationship between an agent's risk preference and loss sensitivity, solving the problem that traditional models struggle to characterize the heterogeneity of individual decisions and achieving accurate mapping of high-dimensional features to the decision space.

[0026] Furthermore, this invention realizes the complete life cycle simulation of the agent from "state acquisition - decision execution - asset scaling - failure exit" by using a normal distribution random incremental model and minimum investment threshold determination. It provides a high-confidence stress test environment for complex financial market environments and constructs a closed-loop survival evolution verification mechanism.

[0027] Furthermore, this invention calculates the modified utility value of each channel based on expected rate of return, channel fee rate, and immediate satisfaction; wherein the expected return utility is calculated using a piecewise nonlinear function and utilizing a risk attitude coefficient. and loss sensitivity To characterize an individual's asymmetric sensitivity to gains and losses.

[0028] Furthermore, this invention matches differentiated channel switching resistance coefficients based on customer satisfaction levels (very satisfied, moderately satisfied, dissatisfied). For moderate or unsatisfactory situations: calculate the utility difference between the alternative and the current channel. Only when Asset reallocation is triggered at certain times. For a highly satisfactory state: a dual-threshold judgment logic determines whether to activate the dual-channel parallel investment mode.

[0029] Furthermore, after triggering the redistribution, this invention calculates the fund allocation coefficient based on the relative weight of the corrected utility of each effective channel using a dynamic weight allocation matrix. ;like If the value is below the preset threshold, execute the single-channel regression logic.

[0030] Furthermore, the investment amount of this invention under a highly satisfactory state follows a normal distribution. The upward floating will achieve ecological increment; the medium to low satisfaction status will be reduced according to the proportion of satisfaction score.

[0031] Furthermore, this invention effectively improves the numerical stability of the simulation system in large-scale heterogeneous agent interaction scenarios by introducing a first-order recursive filter for satisfaction and a stepped resistance operator. This mechanism can filter out unsteady instantaneous fluctuation noise, ensuring good convergence of the simulation evolution trajectory during long-term operation, and solving the system collapse problem that traditional models are prone to when dealing with nonlinear feedback.

[0032] Furthermore, this invention constructs a system that includes a risk attitude coefficient. Loss sensitivity and channel switching resistance coefficient The segmented utility model achieves decoupled control between the agent's subjective preferences and the objective incentives of the environment. This enables the system to perform accurate parameter sensitivity analysis for single variables (such as rate changes or revenue fluctuations), providing a high-confidence simulation environment for boundary condition testing of complex strategies before physical deployment.

[0033] Furthermore, this invention, based on mechanistic modeling rather than a purely data-driven statistical model, significantly reduces the simulation system's reliance on all historical transaction data. By pre-setting behavioral benchmarks for heterogeneous agents, the system possesses 'cold start' simulation capabilities without historical data support, enabling it to predict potential customer migration trajectories for new financial products or entirely new sales channels that have not yet been launched, thus expanding the application boundaries of simulation technology. Attached Figure Description

[0034] The accompanying drawings, which form part of this specification, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a flowchart illustrating a customer channel migration behavior simulation method based on an intelligent agent according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating a customer channel migration behavior simulation method based on an intelligent agent, according to another embodiment of the present invention. Figure 3This is a schematic diagram of the structure of a customer channel migration behavior simulation device based on an intelligent agent according to an embodiment of the present invention; Figure 4 This is a structural block diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0035] The present invention will now be described in detail with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other.

[0036] The following detailed descriptions are exemplary and intended to provide further detailed explanation of the invention. Unless otherwise specified, all technical terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0037] 1. Technical Terminology and System Definitions: This invention provides a customer channel migration behavior simulation system based on multi-agent interaction. Its core is to construct a complex evolutionary network composed of a large number of heterogeneous agents to simulate the nonlinear state evolution of micro-individuals under multidimensional environmental stimuli.

[0038] Heterogeneous intelligent agents: The simulated subjects in this system, each possessing independent risk preference parameters. And the decision weight space. The agent not only interacts with the simulation environment, but also exchanges information with other agents in the neighborhood through the social radius.

[0039] Channel environment operators: system parameters that simulate physical channels, including expected revenue operators. Rate loss operator and service level parameters .

[0040] Non-principal-protected closed-end product simulation model: In this system, wealth management products are abstracted as having a specific volatility ( ) and expected value ( A time series generator. The product is in a locked steady state during the lock-up period, and the actual return rate signal is released only at the cycle nodes to trigger the agent's state transition determination.

[0041] 2. Overview of the Core Simulation Logic This invention focuses on simulating the "topological distribution evolution" of a large number of individuals in a constrained financial environment. Its technical logic is not simply commercial probability statistics, but is based on the following physical simulation principles: Micro-emergent mechanisms: Higher-level system attributes (such as channel share and system churn rate) originate from the interactions of lower-level subsystems (heterogeneous agents). By simulating N agents following specific nonlinear utility rules, complex macro-evolutionary phenomena can be observed in the system.

[0042] Resistance model for state transitions: The system introduces "switching resistance". This simulates psychological inertia and migration costs, analogous to static friction in physical systems. This allows the simulation model to overcome numerical instability, where "small utility perturbations cause severe system oscillations."

[0043] Nonlinear utility synthesis: The system utilizes piecewise functions Characterizing the asymmetric sensitivity of intelligent agents to gains and losses This maps multidimensional environmental characteristics into a single decision utility value. .

[0044] 3. Simulation process overview: Assume there is Customer agents with specific risk preference feature vectors are distributed within a predefined social network topology. During each simulation cycle, the system executes the following control loop: Input phase: Each channel environment releases a parameter package containing service level, expected revenue, and fee rate.

[0045] Processing phase: The agent updates its average satisfaction level based on the actual revenue feedback for this period. And based on this, the channel-adjusted utility value is calculated.

[0046] Decision-making phase: The agent triggers different decision paths (migration, maintenance, or dual-channel parallel) based on its satisfaction state (very satisfied, moderately satisfied, dissatisfied) and performs the reallocation of asset weights.

[0047] Output phase: The system performs a survival check based on the remaining asset size of the agent. If the investment amount is lower than the threshold, the agent is deemed to have failed and exits the environment.

[0048] Through the closed-loop simulation described above, this invention can provide financial institutions with a low-cost, high-precision system evolution testing platform before actual parameter adjustments.

[0049] Please see Figure 1 As shown, this embodiment of the invention provides a customer channel migration behavior simulation method based on an intelligent agent, including the following steps: Step 1: Simulation Initialization. Perform initial configuration of the system environment and agent states, specifically including setting the initial time. .

[0050] (1) Construct a heterogeneous customer intelligent agent cluster: generate Each client agent is assigned a heterogeneous risk attitude coefficient. The heterogeneous risk attitude coefficient is used to define the agent's nonlinear response sensitivity to system feedback signals, including risk-seeking, risk-averse, and risk-neutral types.

[0051] (2) Generate initial channel selection: use only channel 1 or channel 0, or select to use two channels and randomly assign customer channels.

[0052] (3) Set up channels: For each channel, set its basic attribute parameters, including: channel fee rate (corresponding to system access loss), service level (corresponding to environmental adjustment factor), and the attributes of the financial products deployed on the channel (including the expected rate of return of the financial products and the product risk category).

[0053] (4) Setting time and space: Set maximum simulation time .

[0054] Setting the social radius: This distributes customers evenly across different areas. The social radius defines the boundary within which each customer can interact with customers within their social radius. A larger social radius value means that each customer interacts with more customers, simulating information exchange in a topological environment.

[0055] (5) Set decision trigger parameters: Channel utility threshold: A criterion used to determine whether a customer will continue to use a particular channel.

[0056] Set satisfaction habit factors : Range of values This is used to define the rate at which an agent's memory of historical states decays. The larger the value, the stronger the system inertia.

[0057] Set the channel switching resistance coefficient This is used to quantify the system friction (including operational costs and psychological inertia) faced by an agent when changing connection nodes (channels). For agents in different satisfaction states, a stepped resistance coefficient is set (e.g., for dissatisfied states). Moderately satisfied ).

[0058] (6) Initialize the marketing incentive pool: Set the external intervention signals (coupons) that can be called by each channel and their corresponding utility compensation values. .

[0059] Step 2: Service allocation and revenue release: Start the simulation iteration process.

[0060] Service allocation: Based on the attribute settings of the customer agents in step 1, assign a corresponding service level to each agent. This service level will serve as an environmental impact factor in subsequent satisfaction calculations.

[0061] Feedback signal release: After the simulation runs expire, each channel will release the actual rate of return of the wealth management products.

[0062] Calculate the return deviation: The system automatically calculates the difference between the actual rate of return and the expected rate of return set in step 1, which is taken as the return deviation. This difference, along with the aforementioned service level and the investment amount in the customer's intelligent agent, constitutes the set of input variables for calculating real-time satisfaction in the next stage.

[0063] Step 3: Channel satisfaction calculation and customer satisfaction status classification: For each customer agent, perform a state update based on a recursive smoothing mechanism.

[0064] This step employs a two-layer computational architecture of "current real-time satisfaction sampling + historical satisfaction recursive smoothing" to accurately simulate the evolution of the agent's psychological state.

[0065] Step 3.1, calculate the current period's immediate satisfaction level. ; During simulation time After the customer's intelligent agent completes the channel interaction, the real-time satisfaction level for the current period is calculated based on the multi-dimensional input components:

[0066] (Investment Amount Component): Reflects the positive contribution of the client's asset size and available resources to satisfaction.

[0067] (Service Level Component): Reflects the overall service quality provided by the channel, including environmental factors such as business capabilities, response speed, and facility level.

[0068] (Yield Deviation Component): Represents the difference between the actual yield to maturity and the expected yield, quantifying the impact of the achievement of expected yield on individual evaluations.

[0069] (Feature weights): Represent the degree of attention paid by the client's intelligent agent to different influencing factors, and satisfy the condition that the sum of the weights equals 1.

[0070] (Decrease Factor): Simulates the diminishing marginal effect of customer satisfaction as time goes by and the number of repeat purchases increases.

[0071] Step 3.2: Calculate the average customer satisfaction score for the channel; To simulate the cumulative effect of real customers' impressions and consumption habits, this invention does not simply use an arithmetic average, but instead introduces a habit factor ( The average customer satisfaction with the channel is updated using a weighted moving average algorithm at simulation time t.

[0072] The updated current average satisfaction score represents the customer agent's steady-state impression of the channel; The historical average satisfaction value of the previous simulation cycle; : Represents the current immediate satisfaction level, calculated from step 3.1; Satisfaction habit factor; range of values This factor determines the stickiness of the system's memory of historical states. The larger the value, the more the agent is driven by historical impressions, and the smoother the system evolution. The smaller the value, the more sensitive the agent is to current fluctuations.

[0073] This recursive algorithm is physically equivalent to a digital low-pass filter. It can effectively filter out instantaneous noise signals caused by fluctuations in single-time revenue or service randomness, ensure the numerical convergence of the simulation system in long-term evolution, and prevent the agent from frequently generating abnormal state switching actions due to instantaneous interference.

[0074] Step 3.3: Customer Status Classification; Calculated With two preset thresholds ( By comparing these figures, we can determine the customer's satisfaction level (State) to guide investment decisions in the next phase. like → The status is then determined as: Very satisfied; like → The status is: moderately satisfied; like → The status is then determined as: unsatisfactory.

[0075] Step 4: Channel utility calculation and asset restructuring determination based on resistance constraints; This step quantifies and simulates the agent's next-period decision state under the game of external incentives and internal resistance by constructing a cascaded logic of "utility synthesis - resistance determination - weight allocation". Specifically, this step includes: Step 4.1, calculate the channel correction utility value of the introduced compensation operator.

[0076] The system first synthesizes basic utility by combining the inherent characteristics of each channel with the subjective feedback of the agent, and then superimposes external intervention signals. The calculation logic is as follows: (1) Basic utility Synthesis:

[0077] Expected return gain component Nonlinear modeling: Introducing the bank deposit rate as a risk-free rate reference point to calculate excess return signals. .Will Substitute into the following piecewise investment utility function

[0078]

[0079] Risk attitude operator The range of values ​​is . Represents risk-taking. Represents risk neutrality. This indicates risk aversion.

[0080] Loss Sensitivity Operator Used to simulate an individual's stress sensitivity to loss. A stepped sensitivity is set to meet... .

[0081] Set a fixed nonlinear order This setting aims to ensure that all heterogeneous agents in the simulation system follow a uniform second-order nonlinear response pattern when facing negative fluctuations, thereby eliminating systematic biases caused by differences in function patterns and ensuring the numerical convergence and comparability of simulation results over long periods.

[0082] (2) Rate loss operator After customers discuss the next channel fee rate, each customer's AI agent will use its social radius to obtain information about the channel fee rate from other customers in its vicinity and calculate the average. The AI ​​agent will then compare the target channel's fee rate with this average to derive a utility component reflecting cost losses.

[0083] (3) Average satisfaction component : This component represents the agent's historical accumulated satisfaction with the channel.

[0084] If the customer's intelligent agent has used this channel in the current period, then The average satisfaction level calculated in step 3 is used directly. .

[0085] If it has not been used before, refer to the average satisfaction value of neighboring agents within its social radius who have used the channel.

[0086] (4) : These are the weighting coefficients for revenue, fee rate, and satisfaction, respectively, satisfying... .

[0087] (5) Introduce external intervention compensation operator : To simulate the dynamic adjustment of channel attractiveness by external strategies, the system introduces a marketing incentive compensation factor based on the aforementioned basic utility. This factor serves as a compensation input to the external system, used to correct the final utility value of the channel, thus obtaining the corrected utility value. :

[0088] The simulation simulates the energy compensation sent by an external system (such as a bank's marketing system) to the customer's intelligent agent in an attempt to counteract the inherent resistance within the system.

[0089] Step 4.2: Perform state transition determination based on resistance threshold.

[0090] The system introduces a channel switching resistance coefficient. (Quantifying operational costs and psychological inertia). The value of is dynamically constrained by the satisfaction state determined in step 3, reflecting the decision stickiness of individuals at different satisfaction levels: (1) Step resistance matching logic: Unsatisfactory state: Matching the lowest resistance threshold The client's intelligent agent is in an unstable state and is highly susceptible to external compensation signals. The driver generates migration behavior.

[0091] Moderately satisfactory state: Matching a moderate resistance threshold Customer intelligent agents exhibit path dependence, migrating only when the utility of alternative channels is significantly higher than that of the current channel.

[0092] Very Satisfactory State: Matching the maximum resistance threshold (Or defined as steady-state locking). Client agents in this state do not trigger unidirectional migration decisions by default. Their energy threshold is only used to trigger the "dual-channel parallel" logic, that is, the asset structure will only be changed when the utility of both channels is extremely high, otherwise the status quo will be maintained.

[0093] (2) Decision logic and transition criteria: For customer agents in a state of "dissatisfaction" or "moderate satisfaction", calculate the utility difference. .

[0094] like The customer's intelligent agent determines that external incentives are insufficient to overcome the psychological inertia of the current state, and locks into the status quo.

[0095] like : Determine the driving force penetration resistance threshold and trigger the asset reallocation process.

[0096] Step 4.3, Calculation of weights for dynamic asset portfolio.

[0097] Only when At this time, the asset redistribution mechanism is triggered. The system no longer executes discrete "either / or" choices, but instead calculates the fund allocation coefficient based on the relative proportion of the adjusted utility value of each channel. :

[0098] in, This indicates that the customer's intelligent agent will be deployed to the channel in the next simulation cycle. Resource weight (funding ratio); where N is the total number of channels.

[0099] Single-channel regression logic: If the calculated value of a certain channel... If the value falls below a preset small threshold (e.g., 0.05), to simulate abandonment due to excessive management costs, the system will forcibly shut down the channel. With zeroing out, the customer intelligent agent reverts to a single-channel model.

[0100] Step 5: Dynamic adjustment of investment amount and survival determination based on satisfaction feedback; Step 5, as the final step of the simulation cycle, performs nonlinear scaling adjustment on the total asset pool based on the customer agent's satisfaction state (State) and executes the final survival check.

[0101] Step 5.1: Construct a dynamic investment amount generation model based on satisfaction status; The system reads the customer satisfaction status determined in step 3 and the decision results produced in step 4, and calculates the customer agent for the next simulation cycle according to the following logic. Total investment amount : Customer satisfaction status is now "very satisfied" (incremental injection mechanism): This manifests as ecological asset retention and incremental investment. The next period's investment amount for the intelligent agent will fluctuate randomly upwards based on the current amount, following a normal distribution. The calculation formula is as follows:

[0102] Gain perturbation factor : Follows a normal distribution .in The preset average capital increase rate (simulating the average level of additional investment by customers due to increased trust). Volatility (simulating market stochasticity).

[0103] This model simulates the natural asset growth trend of highly loyal customers under a positive feedback loop. The introduction of a normal distribution is to restore the random fluctuation characteristics of customer investment behavior in the real market.

[0104] Customer satisfaction is at a moderate or unsatisfactory level (stock depletion mechanism): This manifests as asset withdrawal or a wait-and-see attitude. The next investment amount will be based on the average satisfaction score. Reduce proportionally:

[0105] Retention coefficient : is a monotonically increasing function of satisfaction (e.g. The lower the satisfaction level, the lower the retention rate. The smaller the value, the greater the reduction in funds.

[0106] This mechanism quantifies the "erosion effect" of negative experiences on asset size, realizing a dynamic mapping from psychological dissatisfaction to capital outflow.

[0107] Step 5.2: Execute based on allocation coefficient Asset allocation; The total investment amount pool for the next period has been determined. Then, the system calls the fund allocation coefficient calculated in step 4. Complete the final asset allocation update:

[0108] in, To deploy customer intelligent agents to specific channels in the next cycle The absolute amount.

[0109] This invention combines the abstract concept of "total amount fluctuation" with the specific concept of "channel selection weight," thereby enabling the tracking of fund flows at the micro level.

[0110] Step 5.3, Customer Churn Determination and Exit Mechanism (Survival Verification); After the amount update is completed, the system performs boundary condition checks: Judgment logic: If the calculated total investment amount for the next period... Drop to zero, or below the preset minimum threshold for investment. (Right now ).

[0111] Perform the following actions: The agent has been determined to have "completely disappeared." It has been removed from the financial market simulation environment.

[0112] Clear the Agent state information of the agent and mark the customer's Lifetime Value (CLV) as terminated.

[0113] Step 6: Simulation cycle iteration and termination condition determination This step is responsible for controlling the timing progression of the entire multi-agent system: Timing update: If the current simulation time step is... The preset maximum iteration period has not been reached. (Right now ), then let The system will update the agent's status in step 5 (including investment amounts from various channels). and weight Using this as the initial value, return to step 2 to start a new round of calculations.

[0114] Termination and convergence criteria: If the system reaches the maximum simulation time step... If all active agents in the simulation environment trigger the churn exit mechanism, the current simulation experiment is considered to have ended.

[0115] Results Export: The system stops iterating and summarizes all the data produced in each cycle to generate the final customer channel migration and evolution trajectory report.

[0116] This invention provides a simulation method for customer channel migration behavior based on intelligent agents, and also a simulation method for the evolution of complex systems based on multi-agent interaction. By modeling the behavior of large-scale heterogeneous intelligent agents under nonlinear utility-driven conditions, the system can output an evolutionary trajectory report reflecting state transitions in complex topological environments, specifically including: System steady-state distribution dynamic trajectory: Real-time plotting the evolution of the agent set over time step, showing the distribution evolution of the agent set in environments with different parameter configurations (such as channels of different rate operators and revenue operators), used to analyze the topological stability of the system under different external stimuli.

[0117] Agent survival state and failure probability distribution: Based on the "resource threshold depletion" definition, the agent failure model (Survival Model) quantitatively outputs the scale evolution speed and distribution characteristics of the agent exiting the simulation environment under different parameter perturbations.

[0118] Parameter sensitivity cluster identification: Through multi-dimensional vector space analysis, identify clusters of intelligent agents that are extremely sensitive to environmental parameters (such as high cost sensitivity coefficient and in a negative satisfaction steady state), and accurately lock the critical threshold that triggers state mutation.

[0119] Based on the simulation results of the above-mentioned system evolution trajectory, this invention achieves optimization and steady-state control of the environment configuration of complex systems: System environment parameter configuration optimization (Sensitivity Analysis): Before actually adjusting environment operators (such as physical channel rates and service level parameters), parameter sensitivity testing is performed using a simulation engine. For example: by adjusting operators... The weights are used to determine whether the observed agent is shifted from the existing distribution or whether its system survival time has been extended.

[0120] System robustness and stress testing: Under simulated extreme disturbance signals (such as the collapse of the satisfaction feedback system caused by drastic fluctuations in the expected return operator), whether the system will experience "mass state collapse (large-scale migration of agents)" is evaluated to assess the dynamic steady-state maintenance capability of the simulation system.

[0121] Heterogeneity mapping configuration recommendations: Through simulation, we discover the response priorities of agents with specific attribute vectors to different environmental operators (e.g., the priority of responding to "service level" is higher than that of "rate loss"), thereby providing nonlinear optimization path recommendations for resource scheduling in complex systems.

[0122] The core technological contribution of this invention lies in the "coupling optimization research of heterogeneous intelligent agents and multidimensional environmental space". Through simulation results feedback system, it demonstrates under what parameter configurations an intelligent agent with a specific risk attitude feature vector and sensitivity weight exhibits a higher degree of stable survival.

[0123] Its technical effects are manifested in the following ways: by utilizing the matching rules between the simulated agent behavior patterns and environmental parameters (Agent-Environment Fit), the distribution structure of agents in the system is optimized by dynamically adjusting the environmental parameter mapping, thereby improving the operating efficiency and individual survival rate of the entire simulation system and preventing the system from collapsing due to large-scale agent failures.

[0124] Please see Figure 2 As shown, this embodiment of the invention provides a customer channel migration behavior simulation method based on an intelligent agent, including: S100. Perform initial configuration of the simulation environment and the client agent's state: Initialize n client agents and assign a heterogeneous risk attitude coefficient to each agent. The heterogeneous risk attitude coefficient is used to define the agent's nonlinear response sensitivity to system feedback signals, including risk-seeking, risk-averse, and risk-neutral types.

[0125] Generate initial channel selection: use only channel 1 or channel 0, or select to use both channels, and randomly assign customer channels.

[0126] Set up channels: For each channel, set its basic attribute parameters, including: channel fee rate (corresponding to system access cost), service level (corresponding to environmental adjustment factor), and the attributes of the financial products offered on this channel (including the expected rate of return and risk category of the financial products).

[0127] Setting time and space: Set maximum simulation time .

[0128] Setting the social radius: This distributes customers evenly across different areas. The social radius defines the boundary within which each customer can interact with customers within their social radius. A larger social radius value means that each customer interacts with more customers, simulating information exchange in a topological environment.

[0129] Set decision trigger parameters: Channel utility threshold: A criterion used to determine whether a customer will continue to use a particular channel.

[0130] Set satisfaction habit factors : Range of values This is used to define the rate at which an agent's memory of historical states decays. The larger the value, the stronger the system inertia.

[0131] Set the channel switching resistance coefficient This is used to quantify the system friction (including operational costs and psychological inertia) faced by an agent when changing connection nodes (channels). For agents in different satisfaction states, a stepped resistance coefficient is set (e.g., for dissatisfied states). Moderately satisfied ).

[0132] Initialize the marketing incentive pool: Set the external intervention signals (coupons) that can be invoked by each channel and their corresponding utility compensation values. .

[0133] S200: Assign service levels to each customer agent and conduct simulation for this period. After the current period expires, release the actual rate of return of the product through each channel. Calculate the profit deviation based on the actual rate of return. Service allocation: Based on the attribute settings of the customer agents in step S100, assign a corresponding service level to each agent. This service level serves as an environmental impact factor in subsequent satisfaction calculations.

[0134] Feedback signal release: After the simulation runs expire, each channel will release the actual rate of return of the wealth management products.

[0135] Calculate the return deviation: The system automatically calculates the difference between the actual rate of return and the expected rate of return set in step S100, which is taken as the return deviation. This difference, along with the aforementioned service level and the investment amount in the customer's intelligent agent, constitutes the set of input variables for calculating real-time satisfaction in the next stage.

[0136] S300: For each customer agent, calculate the current period's real-time satisfaction based on the current period's service level, revenue deviation, and investment amount; calculate the customer's average satisfaction score with the channel based on the current period's real-time satisfaction score; and determine the customer's satisfaction status based on the customer's average satisfaction score with the channel. During simulation time After the customer's intelligent agent completes the channel interaction, the real-time satisfaction level for the current period is calculated based on the multi-dimensional input components:

[0137] (Investment Amount Component): Reflects the positive contribution of the client's asset size and available resources to satisfaction.

[0138] (Service Level Component): Reflects the overall service quality provided by the channel, including environmental factors such as business capabilities, response speed, and facility level.

[0139] (Yield Deviation Component): Represents the difference between the actual yield to maturity and the expected yield, quantifying the impact of the achievement of expected yield on individual evaluations.

[0140] (Feature weights): Represent the degree of attention paid by the client's intelligent agent to different influencing factors, and satisfy the condition that the sum of the weights equals 1.

[0141] (Decrease Factor): Simulates the diminishing marginal effect of customer satisfaction as time goes by and the number of repeat purchases increases.

[0142] Calculating the average customer satisfaction score for the channel: To simulate the cumulative effect of real customer impressions and consumption habits, this invention does not simply use an arithmetic average, but instead introduces a habit factor ( The average customer satisfaction with the channel is updated using a weighted moving average algorithm at simulation time t.

[0143] The updated current average satisfaction score represents the customer agent's steady-state impression of the channel; The historical average satisfaction value of the previous simulation cycle; : Represents the current immediate satisfaction level, calculated from step 3.1; Satisfaction habit factor; range of values This factor determines the stickiness of the system's memory of historical states. The larger the value, the more the agent is driven by historical impressions, and the smoother the system evolution. The smaller the value, the more sensitive the agent is to current fluctuations.

[0144] This recursive algorithm is physically equivalent to a digital low-pass filter. It can effectively filter out instantaneous noise signals caused by fluctuations in single-time revenue or service randomness, ensure the numerical convergence of the simulation system in long-term evolution, and prevent the agent from frequently generating abnormal state switching actions due to instantaneous interference.

[0145] Customer status classification: Calculated With two preset thresholds ( By comparing these figures, we can determine the customer's satisfaction level (State) to guide investment decisions in the next phase. like → The status is then determined as: Very satisfied; like → The status is: moderately satisfied; like → The status is then determined as: unsatisfactory.

[0146] S400: Based on the expected rate of return, channel fee rate, and current real-time satisfaction rate in the initial configuration of the simulation environment, calculate the utility value of each channel for each customer agent; based on the customer satisfaction status of each customer agent, the utility value of each channel, and the channel switching resistance coefficient in the initial configuration of the simulation environment. The decision results are obtained; the decision results include the channel selection for the next simulation and the allocation ratio of investment funds for each channel; (1) Basic utility Synthesis:

[0147] Expected return gain component Nonlinear modeling: Introducing the bank deposit rate as a risk-free rate reference point to calculate excess return signals. .Will Substitute into the following piecewise investment utility function

[0148]

[0149] Risk attitude operator The range of values ​​is . Represents risk-taking. Represents risk neutrality. This indicates risk aversion.

[0150] Loss Sensitivity Operator Used to simulate an individual's stress sensitivity to loss. A stepped sensitivity is set to meet... .

[0151] Set a fixed nonlinear order This setting aims to ensure that all heterogeneous agents in the simulation system follow a uniform second-order nonlinear response pattern when facing negative fluctuations, thereby eliminating systematic biases caused by differences in function patterns and ensuring the numerical convergence and comparability of simulation results over long periods.

[0152] (2) Rate loss operator After customers discuss the next channel fee rate, each customer's AI agent will use its social radius to obtain information about the channel fee rate from other customers in its vicinity and calculate the average. The AI ​​agent will then compare the target channel's fee rate with this average to derive a utility component reflecting cost losses.

[0153] (3) Average satisfaction component : This component represents the agent's historical accumulated satisfaction with the channel.

[0154] If the customer's intelligent agent has used this channel in the current period, then The average satisfaction level calculated in step 3 is used directly. .

[0155] If it has not been used before, refer to the average satisfaction value of neighboring agents within its social radius who have used the channel.

[0156] (4) : These are the weighting coefficients for revenue, fee rate, and satisfaction, respectively, satisfying... .

[0157] (5) Introduce external intervention compensation operator : To simulate the dynamic adjustment of channel attractiveness by external strategies, the system introduces a marketing incentive compensation factor based on the aforementioned basic utility. This factor serves as a compensation input to the external system, used to correct the final utility value of the channel, thus obtaining the corrected utility value. :

[0158] The simulation simulates the energy compensation sent by an external system (such as a bank's marketing system) to the customer's intelligent agent in an attempt to counteract the inherent resistance within the system.

[0159] Perform state transition determination based on resistance threshold; introduce channel switching resistance coefficient. (Quantifying operational costs and psychological inertia). The value of is dynamically constrained by the satisfaction state determined in step 3, reflecting the decision stickiness of individuals at different satisfaction levels: (1) Step resistance matching logic: Unsatisfactory state: Matching the lowest resistance threshold The client's intelligent agent is in an unstable state and is highly susceptible to external compensation signals. The driver generates migration behavior.

[0160] Moderately satisfactory state: Matching a moderate resistance threshold Customer intelligent agents exhibit path dependence, migrating only when the utility of alternative channels is significantly higher than that of the current channel.

[0161] Very Satisfactory State: Matching the maximum resistance threshold (Or defined as steady-state locking). Client agents in this state do not trigger unidirectional migration decisions by default. Their energy threshold is only used to trigger the "dual-channel parallel" logic, that is, the asset structure will only be changed when the utility of both channels is extremely high, otherwise the status quo will be maintained.

[0162] (2) Decision logic and transition criteria: For customer agents in a state of "dissatisfaction" or "moderate satisfaction", calculate the utility difference. .

[0163] like The customer's intelligent agent determines that external incentives are insufficient to overcome the psychological inertia of the current state, and locks into the status quo.

[0164] like : Determine the driving force penetration resistance threshold and trigger the asset reallocation process.

[0165] Weighting of dynamic asset portfolios: only when At this time, the asset redistribution mechanism is triggered. The system no longer executes discrete "either / or" choices, but instead calculates the fund allocation coefficient based on the relative proportion of the adjusted utility value of each channel. :

[0166] in, This indicates that the customer's intelligent agent will be deployed to the channel in the next simulation cycle. Resource weight (funding ratio); where N is the total number of channels.

[0167] Single-channel regression logic: If the calculated value of a certain channel... If the value falls below a preset small threshold (e.g., 0.05), to simulate abandonment due to excessive management costs, the system will forcibly shut down the channel. With zeroing out, the customer intelligent agent reverts to a single-channel model.

[0168] S500: Each customer agent simulates and obtains the next period's investment amount based on the current period's customer satisfaction status and decision results; if the next period's investment amount drops to zero or is lower than the preset minimum investment threshold, the corresponding customer agent is removed from the simulation environment. Total asset pool adjustment: Very satisfied with the intelligent agent: Next phase investment amount ,in Simulate ecological increments.

[0169] Medium / Unsatisfactory Agent: Next period investment amount based on The proportion has been reduced.

[0170] Churn determination: If If the agent fails or is lost, the agent's information will be cleared.

[0171] S600, iterative simulation, until the termination condition is met, to obtain the simulation results of customer migration evolution data under each channel strategy; If the maximum simulation time has not been reached and there are still surviving agents, update the time step back to S200; otherwise, output the system evolution trajectory report.

[0172] Please see Figure 3 As shown, this embodiment of the invention provides a customer channel migration behavior simulation device based on an intelligent agent, comprising: The initialization module is used to perform the initial configuration of the simulation environment and the state of the client intelligent agent; The service allocation and revenue release module is used to allocate service levels to each customer agent for the current period simulation. After the current period expires, the actual rate of return of the product is released by each channel. The revenue deviation is calculated based on the actual rate of return. The channel satisfaction calculation and customer satisfaction status classification module is used to calculate the current period's real-time satisfaction for each customer agent based on the current period's service level, revenue deviation, and investment amount; calculate the customer's average satisfaction score with the channel based on the current period's real-time satisfaction score; and determine the customer's satisfaction status based on the customer's average satisfaction score with the channel. The channel utility calculation and selection module is used to calculate the utility value of each channel for each customer agent based on the expected rate of return, channel fee rate, and current real-time satisfaction rate in the initial configuration of the simulation environment; and to calculate the utility value of each channel based on the customer satisfaction status of each customer agent, the utility value of each channel, and the channel switching resistance coefficient in the initial configuration of the simulation environment. The decision results are obtained; the decision results include the channel selection for the next simulation and the allocation ratio of investment funds for each channel; The dynamic investment amount adjustment module is used by each customer agent to simulate and obtain the investment amount for the next period based on the current customer satisfaction status and decision results. If the investment amount for the next period drops to zero or is lower than the preset minimum investment threshold, the corresponding customer agent will be removed from the simulation environment. The loop determination module is used for iterative loop simulation until the termination condition is met, and obtains the simulation results of customer migration evolution data under each channel strategy.

[0173] This invention provides a method and related apparatus for simulating customer channel migration behavior based on intelligent agents. By constructing a multidimensional heterogeneous intelligent agent model and a nonlinear dynamic evolution mechanism, it achieves high-precision simulation of individual migration behavior in complex topological environments. Its core technological contributions and beneficial effects are reflected in: 1. This invention solves the technical problem of inaccurate modeling of the "lag" in individual decision-making in simulation systems: It breaks through the idealized assumption of "utility-driven instant migration" in traditional simulation technology and innovatively introduces the channel switching resistance coefficient. Marketing incentive hedging operator The collaborative decision-making mechanism effectively replicates the "decision friction" prevalent in physical systems by quantifying the psychological inertia and external intervention compensation of intelligent agents during node migration, significantly improving the fitting accuracy and robustness of the simulation evolution curve in simulating real complex systems.

[0174] 2. Enhanced the "long-term memory" and smoothness characteristics of individual psychological state evolution during simulation: This invention introduces a habit factor-based approach. Recursive satisfaction calculation model: This mechanism uses a weighted recursive approach that combines the agent's current psychological state with its historical state, overcoming the simulation data oscillations and non-convergence caused by instantaneous parameter disturbances in traditional models. This enables the simulation system to have higher numerical stability in large-scale, long-term evolution predictions.

[0175] 3. Achieved a granular leap from "discrete state" to "continuous weight allocation" in simulation results: This invention changes the traditional "either / or" discrete migration mode and constructs an asset portfolio allocation algorithm based on modified utility ratios. This algorithm allows the agent to dynamically adjust the resource allocation ratio based on the continuous distribution of node utility. This has enabled a leap from "individual flow simulation" to "resource scale evolution simulation", greatly improving the simulation system's ability to depict the dynamic allocation process of resources across multiple channels.

[0176] 4. Enhanced accuracy of simulation model in evaluating external dynamic intervention signals: By constructing a closed-loop simulation link that includes multi-dimensional factors such as investment amount, service level, and return difference, this invention establishes a complete response mechanism of "strategy input - resistance game - behavioral feedback." This enables the simulation system to quantitatively measure the impact of external interventions of different intensities on the overall evolutionary equilibrium point of the system, solving the technical pain point of existing technologies that struggle to accurately stress test dynamic intervention strategies when historical data accumulation is lacking.

[0177] Please see Figure 4 As shown, this embodiment of the invention provides an electronic device 100 for implementing a customer channel migration behavior simulation method based on an intelligent agent; the electronic device 100 includes a memory 101, at least one processor 102, a computer program 103 stored in the memory 101 and executable on the at least one processor 102, and at least one communication bus 104.

[0178] The memory 101 can be used to store the computer program 103. The processor 102 implements the steps of the agent-based customer channel migration behavior simulation method described in the embodiment by running or executing the computer program stored in the memory 101 and calling the data stored in the memory 101. The memory 101 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the electronic device 100 (such as audio data), etc. In addition, the memory 101 may include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other non-volatile solid-state storage device.

[0179] The at least one processor 102 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The processor 102 may be a microprocessor or any conventional processor. The processor 102 is the control center of the electronic device 100, connecting various parts of the electronic device 100 via various interfaces and lines.

[0180] The memory 101 in the electronic device 100 stores multiple instructions to implement a customer channel migration behavior simulation method based on an intelligent agent, and the processor 102 can execute the multiple instructions to achieve the following: Perform initial configuration of the simulation environment and the state of the client intelligent agent; The service level is assigned to each customer agent for this period simulation. After the current period expires, the actual rate of return of the product is released by each channel. The return deviation is calculated based on the actual rate of return. For each customer agent, calculate the current period's real-time satisfaction based on the current service level, revenue deviation, and investment amount; calculate the customer's average satisfaction score with the channel based on the current period's real-time satisfaction score; and determine the customer's satisfaction status based on the customer's average satisfaction score with the channel. Based on the expected rate of return, channel fee rate, and current real-time satisfaction rate in the initial configuration of the simulation environment, calculate the utility value of each channel for each customer agent; based on the customer satisfaction status of each customer agent, the utility value of each channel, and the channel switching resistance coefficient in the initial configuration of the simulation environment. The decision results are obtained; the decision results include the channel selection for the next simulation and the allocation ratio of investment funds for each channel; Each customer agent simulates and obtains the next period's investment amount based on the current period's customer satisfaction status and decision results; if the next period's investment amount drops to zero or is lower than the preset minimum investment threshold, the corresponding customer agent will be removed from the simulation environment. The simulation is iterated until the termination condition is met, and the simulation results of customer migration evolution data under each channel strategy are obtained.

[0181] If the modules / units integrated in the electronic device 100 are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, and a read-only memory (ROM).

[0182] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0183] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0184] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0185] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0186] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A method for simulating customer channel migration behavior based on intelligent agents, characterized in that, include: Perform initial configuration of the simulation environment and the state of the client intelligent agent; The service level is assigned to each customer agent for this period simulation. After the current period expires, the actual rate of return of the product is released by each channel. The return deviation is calculated based on the actual rate of return. For each customer agent, calculate the current period's real-time satisfaction based on the current period's service level, revenue deviation, and investment amount, and calculate the customer's average satisfaction score with the channel based on the current period's real-time satisfaction. Customer satisfaction status is determined based on the average customer satisfaction score for the channel. Based on the expected rate of return, channel fee rate, and current real-time satisfaction rate in the initial configuration of the simulation environment, calculate the utility value of each channel for each customer agent; based on the customer satisfaction status of each customer agent, the utility value of each channel, and the channel switching resistance coefficient in the initial configuration of the simulation environment. The decision results are obtained; the decision results include the channel selection for the next simulation and the allocation ratio of investment funds for each channel; Each customer agent simulates and obtains the next period's investment amount based on the current period's customer satisfaction status and decision results. If the investment amount in the next period drops to zero or is lower than the preset minimum investment threshold, the corresponding customer agent will be removed from the simulation environment. The simulation is iterated until the termination condition is met, and the simulation results of customer migration evolution data under each channel strategy are obtained.

2. The customer channel migration behavior simulation method based on intelligent agents according to claim 1, characterized in that, The initial configuration steps for the simulation environment and customer agent state specifically include: constructing a heterogeneous customer agent cluster, generating initial channel selection, setting channels, setting time and space, setting decision triggering parameters, and initializing the marketing incentive pool; wherein, the decision triggering parameters include the channel switching resistance coefficient. The channel switching resistance coefficient Used to quantify the system friction faced by a client intelligent agent when changing connection nodes; Building a heterogeneous customer intelligent agent cluster includes: generating Each customer agent is randomly and evenly assigned risk attitudes to customers. The initial channel selection can be generated as follows: use only channel 1 or channel 0, or select to use both channels and randomly assign customer channels. Setting up channels includes: for each channel, setting the channel fee rate, service level, and the attributes of the financial products offered on that channel; Setting time and space includes: setting the social radius and setting the maximum simulation time T; Setting decision trigger parameters includes: setting channel utility threshold and setting satisfaction habit factor. and setting the channel switching resistance coefficient Channel switching resistance coefficient This is used to quantify the operational costs and psychological inertia faced by customers when adjusting their asset allocation ratios. Different levels of customer satisfaction correspond to different resistance coefficients. Initializing the marketing incentive pool includes: setting the external intervention signals that can be invoked by each channel and their corresponding utility compensation values. .

3. The customer channel migration behavior simulation method based on intelligent agents according to claim 1, characterized in that, In the step of calculating the real-time satisfaction for each customer agent based on the service level, revenue deviation, and investment amount, during simulation time... After the customer intelligent agent completes the channel interaction, the real-time satisfaction rate for the current period is calculated based on the multi-dimensional input components. : in, For the amount of investment, For service level weighting, For profit deviation; All of these are feature weights, and their sum equals 1; This is the attenuation factor.

4. The agent-based customer channel migration behavior simulation method according to claim 3, characterized in that, In the step of calculating the average customer satisfaction score for the channel based on the current period's immediate satisfaction score, the average satisfaction score is: in, This represents the average customer satisfaction level with the channel over simulation time t. This represents the average customer satisfaction with the channel at simulation time t-1; Satisfaction habit factor; In the step of determining customer satisfaction status based on the average customer satisfaction score for the channel, the calculated... With two preset thresholds Compare and determine customer satisfaction levels: like The customer satisfaction status is: very satisfied; like Customer satisfaction status: moderately satisfied; like The customer satisfaction status is: Unsatisfied.

5. The agent-based customer channel migration behavior simulation method according to claim 3, characterized in that, Based on the expected rate of return, channel fee rate, and current real-time satisfaction rate in the initial configuration of the simulation environment, the utility value of each channel for each customer agent is calculated; based on the customer satisfaction status of each customer agent, the utility value of each channel, and the channel switching resistance coefficient in the initial configuration of the simulation environment, the utility value of each channel is calculated; The steps to obtain the decision result specifically include: The basic utility value for each channel is calculated using the following function: in, Indicates the expected return gain. This represents the rate loss operator. To represent average satisfaction, a, b, and c are the weighting coefficients for revenue, fee rate, and satisfaction, respectively; obtain the marketing incentive compensation factors issued to customers through each channel. The basic utility value is adjusted to obtain the adjusted utility value. : Calculate the adjusted utility difference between alternative channels and currently held channels. When the difference in the modified utility value Greater than the channel switching resistance coefficient At that time, the fund allocation coefficient is dynamically calculated: in, This indicates that the customer's intelligent agent will be deployed to the channel in the next simulation cycle. Resource weight; N is the total number of channels; If the fund allocation coefficient of a certain channel If the funds fall below a preset threshold, the funds in that channel will be cleared, and the system will revert to a single-channel model.

6. The customer channel migration behavior simulation method based on intelligent agents according to claim 5, characterized in that, Each customer agent simulates and obtains the investment amount for the next period based on the current customer satisfaction status and decision results. The step of removing the corresponding customer agent from the simulation environment if the next investment amount drops to zero or falls below the preset minimum investment threshold specifically includes: First, the customer agent for the next simulation cycle is determined based on the current customer satisfaction status and decision-making results. Total investment amount For customers who are very satisfied, the total investment amount in the next period will fluctuate upwards according to the normal distribution based on the initial distribution; for customers who are moderately or unsatisfied, the total investment amount in the next period will be reduced proportionally according to their satisfaction score. The client intelligent agent in the next simulation cycle Total investment amount According to the capital allocation coefficient Alternatively, the existing proportions can be allocated to various selected channels to achieve dynamic adjustments to the asset portfolio.

7. A customer channel migration behavior simulation device based on intelligent agents, characterized in that, include: The initialization module is used to perform the initial configuration of the simulation environment and the state of the client intelligent agent; The service allocation and revenue release module is used to allocate service levels to each customer agent for the current period simulation. After the current period expires, the actual rate of return of the product is released by each channel. The revenue deviation is calculated based on the actual rate of return. The channel satisfaction calculation and customer satisfaction status classification module is used to calculate the current period's real-time satisfaction for each customer agent based on the current period's service level, revenue deviation, and investment amount, and to calculate the customer's average satisfaction value with the channel based on the current period's real-time satisfaction. Customer satisfaction status is determined based on the average customer satisfaction score for the channel. The channel utility calculation and selection module is used to calculate the utility value of each channel for each customer agent based on the expected rate of return, channel fee rate, and current real-time satisfaction rate in the initial configuration of the simulation environment; and to calculate the utility value of each channel based on the customer satisfaction status of each customer agent, the utility value of each channel, and the channel switching resistance coefficient in the initial configuration of the simulation environment. The decision results are obtained; the decision results include the channel selection for the next simulation and the allocation ratio of investment funds for each channel; The dynamic investment amount adjustment module is used by each customer agent to simulate and obtain the investment amount for the next period based on the current customer satisfaction status and decision results. If the investment amount for the next period drops to zero or is lower than the preset minimum investment threshold, the corresponding customer agent will be removed from the simulation environment. The loop determination module is used for iterative loop simulation until the termination condition is met, and obtains the simulation results of customer migration evolution data under each channel strategy.

8. The customer channel migration behavior simulation device based on intelligent agents according to claim 7, characterized in that, In the step of calculating the real-time satisfaction for each customer agent based on the service level, revenue deviation, and investment amount, during simulation time... After the customer intelligent agent completes the channel interaction, the real-time satisfaction rate for the current period is calculated based on the multi-dimensional input components. : in, For the amount of investment, For service level weighting, For profit deviation; All of these are feature weights, and their sum equals 1; This is the attenuation factor.

9. An electronic device, characterized in that, It includes a processor and a memory, the processor being used to execute a computer program stored in the memory to implement an agent-based customer channel migration behavior simulation method as described in any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one instruction, which, when executed by a processor, implements an agent-based customer channel migration behavior simulation method as described in any one of claims 1 to 6.

Images