A method and system for building an artificial intelligence-based inclusive finance business system

Abstract

The application provides a method and system for building an artificial intelligence-based inclusive financial business system, relating to the field of financial system construction. The method comprises converting financial rule data into a simulated modulation signal feature vector containing value strength, risk correlation degree and time decay rate through a three-dimensional feature conversion module, and combining a multi-stage state machine to perform market behavior capture, rule space positioning and strategy generation operations; the regulatory text is converted into a continuous control signal injection feature generation logic through a dynamic constraint feedback loop, forming a closed-loop adjustment mechanism for value gain calibration, risk decay compensation and time synchronization correction; a physical signal cross-validation module is used to verify the dual coupling of the main channel waveform features and the auxiliary channel time sequence features of the strategy; through the reference clock synchronization coordination feature conversion, state transition and constraint feedback time sequence deviation, the time sequence deviation of the operation rhythm of each module of the system is ensured to be lower than the preset tolerance.

Description

Technical Field

[0001] This invention relates to the field of financial system construction, specifically to a method and system for constructing an inclusive financial business system based on artificial intelligence. Background Technology

[0002] Inclusive finance, as a core direction of fintech innovation, is committed to lowering service barriers and improving the efficiency of financial resource allocation through technological means. The integration of artificial intelligence technology has promoted the rapid development of fields such as risk assessment and intelligent decision-making, but the complex and ever-changing financial market environment and the need for real-time supervision have placed higher demands on the dynamic adaptability of the system. How to achieve efficient rule parsing, accurate strategy generation, and reliable execution has become a difficult problem that the industry urgently needs to overcome.

[0003] In existing technologies, some systems process financial data based on traditional machine learning models and generate business strategies by matching preset conditions through rule engines; other solutions use natural language processing technology to extract keywords from regulatory texts, form static constraints, and embed them into the decision-making process; in addition, some studies use time series analysis algorithms to monitor market behavior fluctuations and manually adjust strategy parameters in combination with expert experience; the above methods mostly rely on centralized architecture and software-layer logic control, and maintain system operation by updating the rule base at fixed intervals.

[0004] However, traditional machine learning models have limited accuracy in modeling nonlinear relationships of multidimensional dynamic features, resulting in insufficient ability to capture implicit correlations between rules; static rule engines lack real-time feedback mechanisms, making it difficult to respond promptly to market fluctuations and policy adjustments; data processing and strategy generation are delayed under centralized architecture, and high-concurrency scenarios can easily lead to system performance bottlenecks; in addition, the asynchronicity of regulatory constraints and market changes may cause strategy lag or deviation, affecting the accuracy and compliance of business execution. Summary of the Invention

[0005] (a) Technical problems to be solved

[0006] To address the shortcomings of existing technologies, this invention provides a method and system for building an inclusive financial business system based on artificial intelligence. This addresses the following issues raised in the background: traditional machine learning models lack the accuracy to model the nonlinear relationships of multidimensional dynamic features, resulting in weak rule correlation capture capabilities; static rule engines, lacking real-time feedback mechanisms, struggle to respond promptly to market fluctuations and policy changes; centralized architectures suffer from significant data processing delays and performance bottlenecks in high-concurrency scenarios; and the asynchronous nature of regulatory constraints and market changes leads to strategy lags or execution deviations.

[0007] (II) Technical Solution

[0008] To achieve the above objectives, the present invention provides the following technical solution: a method and system for constructing an inclusive financial business system based on artificial intelligence, comprising the following steps:

[0009] S1 receives financial business rule data and generates a rule feature vector containing value intensity, risk correlation, and time-effect decay rate through a three-dimensional feature conversion module. The rule feature vector expresses the dynamic relationship of the three dimensions of value intensity, risk correlation, and time-effect decay rate through analog modulation signal.

[0010] S2 uses a multi-stage state machine to process the feature vector, sequentially executing market behavior capture, rule space positioning, association rule activation, topology adjustment, business strategy generation, and strategy persistence operations. Each stage is triggered by a hardware-level event to initiate a state transition, wherein the market behavior capture includes collecting user transaction behavior features.

[0011] S3 receives regulatory text data and constructs a dynamic constraint feedback loop. It converts the requirements in the regulatory text data into continuous control signals and injects the control signals into the feature vector generation logic of the three-dimensional feature conversion module, forming a closed-loop adjustment mechanism of value gain calibration, risk attenuation compensation, and timeliness synchronization correction.

[0012] S4 performs dual verification of the generation strategy through the physical signal cross-verification module. When the waveform characteristics of the main verification channel and the timing characteristics of the auxiliary verification channel meet the preset coupling conditions, the strategy execution is activated.

[0013] S5 maintains system timing synchronization by coordinating the operation rhythm of each module through a reference clock, ensuring that the timing deviations of feature transitions, state transitions, and constraint feedback are below a preset tolerance. The reference clock shares a clock source with the physical signal cross-validation module and synchronizes the operation rhythm of each module through a phase-locked loop circuit.

[0014] Preferably, the three-dimensional feature conversion module parses the text information in the financial business rule data using a natural language processing model, extracting economic value parameters, risk transmission parameters, and timeliness impact parameters corresponding to value intensity, risk correlation, and timeliness decay rate. The economic value parameters undergo logarithmic conversion via a hardware multiplier and logarithmic amplifier circuit. This includes inputting the normalized economic value parameters into the logarithmic amplifier to generate an amplitude modulation signal, the dynamic range of which is controlled by financial market volatility. When an increase in price series variance is detected, the signal amplitude gradient is compressed to suppress excessive volatility. Simultaneously, the analog-to-digital converter writes the amplitude signal into a specified address in the register. The formula for generating the amplitude modulation signal from the logarithmic conversion of the economic value parameters is: Where k1 is the dynamic adjustment coefficient, V max The maximum normalized value of the economic value parameter is determined when the price series variance σ is detected. 2When the dynamic range of the compressed signal increases, the adjustment coefficient k1 is updated as follows: Where α is the compression factor; the risk transmission parameter is inversely proportionally converted by a voltage-controlled oscillator. The input parameter is scaled proportionally to drive the oscillator to output a frequency modulation signal, the frequency of which is inversely proportional to the risk transmission speed, as shown in the following formula: Where k2 is the proportionality coefficient and ε is a very small constant to prevent division by zero; the deviation is calibrated in real time by a frequency counter to ensure that the signal frequency strictly matches the risk parameters; the time-effect parameters are converted to arctangent by a phase modulator, and the digital-to-analog converter converts the time-effect parameters into a phase-modulated signal, the specific formula being: β is the time-scaling factor, which has a non-linear mapping with the proportion of remaining validity period; it also drives the phase modulator to generate a phase offset, the phase angle of which has a non-linear mapping with the proportion of remaining validity period. Simultaneously, a non-volatile memory stores historical data of the phase modulation signal to support retrospective analysis; finally, the three-dimensional modulation signals are fused into a regular feature vector through an analog superposition circuit. A V ,f R ,φ T The signals are amplitude, frequency, and phase modulated, respectively. The chip temperature is monitored by a temperature sensor, and the signal reference value is corrected in real time by a second-order filtering algorithm in the compensation circuit to eliminate signal drift caused by ambient temperature and ensure the stability of feature vector generation.

[0015] Preferably, in the market behavior capture stage of the multi-stage state machine, user transaction behavior features are collected at a fixed sampling frequency. The transaction behavior is converted into digital signals using an analog-to-digital converter, and the transaction frequency per unit time is statistically analyzed in real time using a circular buffer and sliding window algorithm. Specifically, each sampling updates the data at the tail of the buffer and triggers an accumulator to calculate the mean and variance of the data within the window. The statistical results are transmitted to the rule space positioning module via a bus. In the rule space positioning stage, a hardware accelerator is used to optimize the nearest neighbor search algorithm, calculating the Euclidean distance between the current market state and all rule nodes in the three-dimensional feature space. The three-dimensional coordinates are based on value intensity, risk correlation, and time decay rate. The error tolerance is set to a preset threshold to reduce computational complexity. Nearby node information is cached in a high-speed cache and a fast access channel is established. The nearest neighbor search algorithm is used to calculate the coordinates of the current market state. With all rule nodes The Euclidean distance is: Where V, R, and T represent value intensity, risk correlation, and time-related decay rate, respectively; during the association rule activation phase, the feature similarity of adjacent nodes is compared using a parallel comparator array. When the similarity exceeds the activation threshold, a cross-connection matrix establishes an association link. The link weight is generated by a digital-to-analog converter based on the similarity value, which drives a variable resistor to adjust the connection strength; during the topology adjustment phase, an accumulator calculates the algebraic sum of the connection strengths of all nodes. If the algebraic sum is not zero, the weights of each node are scaled proportionally, specifically: , where w i Let N be the original connection strength of node i, and N be the total number of nodes. The top three links are retained through a priority queue, and the strength of the remaining links is set to zero. The updated topology is written to the register. At the same time, the dynamic balance constraint rules monitor the changes in connection strength in real time through the feedback loop to ensure system stability.

[0016] Preferably, the dynamic constraint feedback loop parses regulatory text data into constraint elements, encodes them into time-continuous control signals using a finite state machine, and the pulse width is proportional to the constraint strength. The control signals are filtered by a programmable low-pass filter to eliminate high-frequency noise. The filter cutoff frequency is dynamically adjusted according to the rule update frequency. Specifically, the digital signal processor calculates the rule update rate in real time and drives the voltage-controlled filter to adjust its parameters. When the user transaction frequency statistics are higher than a preset threshold, the cutoff frequency is inversely proportional to the frequency value to suppress high-frequency interference. The processed control signal is decomposed into three independent channels: the value gain calibration channel adds the control signal to the reference voltage of the amplitude modulation signal through an adder; the risk attenuation compensation channel adjusts the duty cycle of the frequency modulation signal through a subtractor; and the timeliness synchronization correction channel aligns the phase modulation signal with the system clock through a phase comparator. The closed-loop adjustment mechanism monitors the deviation of the corrected feature vector in real time. If the deviation exceeds the threshold, the control signal regeneration process is triggered. At the same time, the feedback loop ensures that the adjustment signal is strictly synchronized with the market rhythm through a phase-locked loop circuit.

[0017] Preferably, the main verification channel of the physical signal cross-validation module detects the phase difference of the feature vector through a phase-locked loop circuit. If the absolute value of the phase difference exceeds the threshold, an alarm signal is triggered and the strategy execution is interrupted. The auxiliary verification channel counts the duty cycle distribution of the pulse signal through a counter. If the ratio of the high-level duration to the period does not meet the preset logic conditions, the strategy is marked as invalid. The dual verification results are judged by a logic AND gate. Only when the main channel does not alarm and the auxiliary channel passes the verification is the strategy execution port activated and a valid mark is written to the non-volatile memory. During the verification process, the reference clock synchronizes the timing of the main and auxiliary channels through a phase-locked loop circuit to ensure the time consistency of the verification results. At the same time, the clock deviation monitoring circuit detects the clock signal of each module in real time. If the deviation exceeds the tolerance, the calibration process is triggered, and data writing is suspended to ensure data integrity.

[0018] Preferably, the reference clock is generated by a high-precision crystal oscillator and distributed to each module through a phase-locked loop circuit; the clock division coefficient of the three-dimensional feature conversion module is set to an integer multiple of the data sampling rate to ensure synchronization between analog-to-digital conversion and signal modulation; the clock division coefficient of the multi-stage state machine is matched with the sliding window update cycle to seamlessly connect the market behavior capture and rule positioning stages; the timing monitoring circuit detects the clock deviation of each module in real time, and if the deviation exceeds the preset tolerance, the phase-locked loop is triggered to resynchronize the clock signal; during the calibration process, the non-volatile memory suspends write operations and resumes them after the clock stabilizes to avoid data misalignment; at the same time, the temperature sensor monitors environmental changes and adjusts the clock signal reference value through a compensation circuit to further improve timing accuracy.

[0019] Preferably, the three-dimensional feature conversion module uses a comparator to detect the difference in the amplitude of value intensity signals under different rules in real time. If the difference exceeds the conflict threshold, an amplitude limiting mechanism is triggered. The limiting mechanism is implemented by a clamping circuit, which limits the signal amplitude to not exceed the system safety threshold through a diode array. At the same time, the dynamic range compression mechanism is suspended to avoid signal distortion. The limited signal is resampled by an analog-to-digital converter and updated to a non-volatile memory. The status register records the limiting status for reference during the topology adjustment stage. If an abnormal temperature is detected, the second-order filtering algorithm in the compensation circuit automatically adjusts the signal reference value, which, together with the limiting mechanism, suppresses environmental interference.

[0020] Preferably, the sliding window algorithm stores user transaction behavior data through a circular queue, with the queue length matching the window size. During each sampling, new data overwrites the oldest data at the tail of the queue. The accumulator calculates the sum of the data within the window in real time, sums the squares of the differences between each data point and the mean, and then divides by the window size to generate the variance. The variance of transaction frequency within the sliding window is calculated as follows: ,in N is the window mean, and N is the window size. The statistical results are transmitted to the rule space positioning module via the bus to dynamically adjust the error tolerance threshold of the nearest neighbor search algorithm. The update cycle of the sliding window is controlled by a hardware timer to ensure strict synchronization with the sampling frequency of the market behavior capture stage. At the same time, the statistical results are written to shared memory for multi-stage state machine calls.

[0021] Preferably, the association rule activation phase accelerates feature similarity calculation through a parallel comparator array. Each comparator independently processes the feature vectors of a pair of rule nodes and outputs the cosine similarity to the priority encoder. If the similarity exceeds the activation threshold, the encoder generates a link establishment instruction. The association rule activation phase calculates the node feature vector F. i With F j The cosine similarity is If so, then establish an associated link; link weight w ijLinear mapping from similarity to The cross-connection matrix connects to the corresponding nodes according to instructions; the link weight is converted from similarity value to analog voltage signal by digital-to-analog converter, which drives variable resistor to adjust the connection strength; the state machine controls the timing of comparator array and switch matrix according to system clock signal to ensure low latency response; neighbor node cache information is preloaded to comparator through high-speed bus to further shorten response time.

[0022] Preferably, during the topology adjustment phase, the accumulator calculates the algebraic sum of the connection strengths of all nodes. If the algebraic sum is not zero, the divider calculates a scaling factor based on the total number of nodes and adjusts the weights of each node proportionally. The scaled strength values ​​are written to the connection weight register via a digital-to-analog converter, while the priority queue selects the top three links and sets the remaining links to zero. The dynamic balance constraint monitors the changes in connection strength in real time through a feedback loop. If an imbalance is detected, a renormalization process is triggered. During the normalization process, the accumulator and divider work together through a hardware pipeline to ensure computational efficiency and real-time performance.

[0023] Preferably, the cutoff frequency of the low-pass filter is dynamically adjusted according to the rule update frequency. The digital signal processor calculates the rule update rate in real time and drives the voltage-controlled filter to adjust its parameters. When the user's transaction frequency statistics are higher than a preset threshold, the cutoff frequency is inversely proportional to the frequency value to suppress high-frequency noise, which is specifically achieved through an inverting proportional amplifier. The filtering timing is strictly aligned with the system master clock through a hardware synchronization circuit to ensure that the control signal is synchronized with the market rhythm. The filtered signal is decomposed into three independent channels by an analog-to-digital converter, which respectively apply to the feature correction of value intensity, risk correlation, and time decay rate. Each channel is equipped with an independent gain adjustment circuit to adapt to the calibration requirements of different dimensions.

[0024] Preferably, the strategy persistence operation is implemented through the page writing mechanism of non-volatile memory. The feature vector and execution mark of the effective strategy are written in a page-aligned format. Before writing, the memory controller checks the erase status of the target page. If it has not been erased, a sector erase command is triggered. During the writing process, the temperature sensor data is adjusted in real time by the compensation circuit to adjust the signal reference value to ensure data reliability. After the writing is completed, the verification circuit compares the header check code with the tail redundancy check bit. If they are inconsistent, a rewrite process is triggered. The memory address mapping table records the storage location of each strategy, supporting fast retrieval and updates.

[0025] (III) Beneficial Effects

[0026] This invention provides a method and system for building an inclusive finance business system based on artificial intelligence. It has the following beneficial effects:

[0027] 1. This invention maps financial rule data into multi-dimensional modulated signals through a three-dimensional feature conversion module, and uses logarithmic, inverse proportional, and arctangent functions to achieve nonlinear feature expression, significantly enhancing the ability to represent complex market behaviors; the multi-stage state machine, through hardware-level event triggering and parallel computing architecture, compresses operations such as market behavior capture, rule location, and topology adjustment to millisecond-level response, greatly improving real-time performance; the dynamic constraint feedback loop, combined with control signals and feature correction mechanisms, ensures that the system can quickly calibrate strategies when regulatory rules change, adapting to the high-frequency fluctuations of the financial market; the physical signal cross-validation module, through time sequence consistency detection and logical completeness verification, provides dual assurance of the credibility of strategy generation, reducing the need for manual intervention, and achieving fully automated and efficient operation of the entire process from data parsing to strategy execution.

[0028] 2. The dynamic balance constraint rules of this invention force the algebraic sum of connection strength to zero through a normalization algorithm, combined with a priority link screening mechanism, effectively preventing strategy deviations caused by topology imbalance; the compensation circuit and the second-order filtering algorithm work together to eliminate signal reference drift caused by changes in ambient temperature in real time, ensuring the stability of feature vector generation; the amplitude limiting mechanism and dynamic range compression strategy are automatically activated when there are rule conflicts or severe market fluctuations, suppressing the risk of signal overload and avoiding hardware damage; the page writing and redundancy verification mechanism of the non-volatile memory ensures that the strategy data is completely and persistently stored in the event of abnormal power failure or system failure; the phase-locked loop synchronization and timing monitoring circuit of the reference clock strictly aligns the operation rhythm of each module, reduces the accumulation of timing deviations, and ultimately achieves high robustness and long-term reliable operation of the system in complex environments. Detailed Implementation

[0029] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] This invention provides a method and system for constructing an inclusive financial business system based on artificial intelligence. Specifically, upon system startup, a three-dimensional feature conversion module receives raw text data from a financial business rules database. It then parses the text information using a natural language processing model, extracting economic value parameters, risk transmission parameters, and timeliness impact parameters, corresponding to the three dimensions of value intensity, risk correlation, and timeliness decay rate, respectively. The economic value parameters are processed by a logarithmic conversion unit and input into a logarithmic amplifier circuit to generate an amplitude modulation signal. The dynamic range of this signal is dynamically adjusted by the volatility of the financial market. When the variance of the price series increases, the signal amplitude gradient is compressed to suppress excessive volatility. Simultaneously, an analog-to-digital converter converts the amplitude signal... The signal is written to a specified address in the non-volatile memory; the risk transmission parameter is converted into a frequency modulation signal by a voltage-controlled oscillator, the frequency value is inversely proportional to the risk transmission speed, and the deviation is calibrated in real time by a frequency counter to ensure that the signal frequency is strictly matched with the risk parameter; the time-effect parameter is generated into a phase offset signal by a phase modulator, the phase angle of which has a non-linear mapping relationship with the proportion of remaining validity period, and the digital-to-analog converter writes the phase parameter into memory and stores historical data to support retrospective analysis; the three-dimensional modulation signals are fused into a regular feature vector by an analog superposition circuit, and at the same time, the temperature sensor monitors the chip temperature change, and the signal reference value is corrected in real time by a second-order filter compensation circuit to eliminate drift caused by ambient temperature;

[0031] The multi-stage state machine module employs a hardware-level event-triggered mechanism. In the market behavior capture stage, user transaction behavior data is collected at a fixed sampling frequency. After the analog-to-digital converter converts the transaction behavior into digital signals, the frequency of transactions per unit time is statistically analyzed in real time using a circular buffer and sliding window algorithm. Each sampling updates the data at the tail of the buffer and triggers an accumulator to calculate the mean and variance of the data within the window. The statistical results are transmitted to the rule space positioning module via a bus. In the rule space positioning stage, a hardware accelerator optimizes the nearest neighbor search algorithm, calculating the Euclidean distance between the current market state and all rule nodes in the three-dimensional feature space. The error tolerance is set to a preset threshold to reduce computational complexity. The information of neighboring nodes is cached in a high-speed cache and a fast access channel is established. During the association rule activation phase, the feature similarity of neighboring nodes is compared through a parallel comparator array. When the similarity exceeds the activation threshold, an association link is established by a cross-switching matrix. The link weight is generated by an analog voltage signal by a digital-to-analog converter based on the similarity value, which drives a variable resistor to adjust the connection strength. During the topology adjustment phase, the algebraic sum of the connection strength of all nodes is calculated through an accumulator. If the algebraic sum is not zero, the weight of each node is scaled proportionally, and the top three links are retained through a priority queue. The strength of the remaining links is set to zero. The updated topology is written to a register and fed back to the dynamic balance constraint module in real time.

[0032] After receiving regulatory text data, the dynamic constraint feedback module parses the constraint elements and encodes them into continuous control signals using a finite state machine. The pulse width is proportional to the constraint strength. The control signals are filtered by a programmable low-pass filter to eliminate high-frequency noise. The filter cutoff frequency is dynamically adjusted according to the rule update frequency. When the user's transaction frequency statistics are higher than the preset threshold, the cutoff frequency is inversely proportional to the frequency value to suppress high-frequency interference. The processed control signals are decomposed into three independent adjustment channels: the value gain calibration channel adds the control signal to the reference voltage of the amplitude modulation signal through an adder; the risk attenuation compensation channel adjusts the duty cycle of the frequency modulation signal through a subtractor; and the timeliness synchronization correction channel aligns the phase modulation signal with the system clock through a phase comparator. The closed-loop adjustment mechanism monitors the deviation of the corrected feature vector in real time. If the deviation exceeds the threshold, the control signal regeneration process is triggered. At the same time, the feedback loop ensures that the adjustment signal is strictly synchronized with the market rhythm through a phase-locked loop circuit.

[0033] The physical signal cross-validation module detects the phase difference of the feature vector through the main validation channel. If the absolute value of the phase difference exceeds the threshold, an alarm signal is generated and the strategy execution is interrupted. The auxiliary validation channel counts the duty cycle distribution of the pulse signal through a counter. If the ratio of the high-level duration to the period does not meet the preset logic conditions, the strategy is marked as invalid. The dual validation results are judged by a logic AND gate. Only when the main channel does not alarm and the auxiliary channel passes the verification is the strategy execution port activated and a valid mark is written to the non-volatile memory. During the validation process, the reference clock synchronizes the timing of the main and auxiliary channels through a phase-locked loop circuit. The clock deviation monitoring circuit detects the clock signal of each module in real time. If the deviation exceeds the tolerance, the calibration process is triggered and data writing is paused.

[0034] The timing synchronization module generates a master clock signal using a high-precision crystal oscillator, and the phase-locked loop circuit distributes the clock to each module. The clock division factor of the three-dimensional feature conversion module is set to an integer multiple of the data sampling rate to ensure synchronization between analog-to-digital conversion and signal modulation. The clock division factor of the multi-stage state machine is matched with the sliding window update cycle, enabling seamless connection between the market behavior capture and rule positioning stages. The timing monitoring circuit detects the clock deviation of each module in real time, and if the deviation exceeds the preset tolerance, it triggers the phase-locked loop to resynchronize. During the calibration process, the non-volatile memory suspends write operations and resumes after the clock stabilizes. The temperature sensor monitors environmental changes and adjusts the clock signal reference value through a compensation circuit, and further eliminates temperature drift by combining a second-order filtering algorithm.

[0035] After receiving the verified strategy signal, the strategy execution and storage module executes financial business operations through the strategy activation port. The strategy parameters and execution results are written to the non-volatile memory in a page-aligned format. Before writing, the memory controller checks the erase status of the target page. If it is not erased, a sector erase command is triggered. During the writing process, the temperature sensor data is adjusted in real time by the compensation circuit to adjust the signal reference value to ensure data reliability. After writing, the verification circuit compares the header checksum with the tail redundancy check bit. If they are inconsistent, a rewrite process is triggered. The memory address mapping table records the storage location of each strategy, supporting fast retrieval and updates. During system operation, the dynamic balance constraint rule forces the algebraic sum of connection strength to zero through a normalization algorithm. The priority link filtering mechanism retains critical connections. The amplitude limiting mechanism and dynamic range compression strategy are automatically activated when there are rule conflicts or market fluctuations to suppress the risk of signal overload, ultimately achieving fully automated, efficient, and highly robust operation.

[0036] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for building an artificial intelligence-based inclusive finance business system, characterized in that, Comprise the following steps: S1 receives financial service rule data and generates a rule feature vector containing value strength, risk correlation degree, and time decay rate through a three-dimensional feature conversion module, the rule feature vector expresses the dynamic change relationship of the three dimensions of value strength, risk correlation degree, and time decay rate in the form of analog modulation signals, the specific steps comprising: S11 based on the financial service rule data received by S1, parses the text information corresponding to the input financial service rule data, and extracts the value strength, risk correlation degree, and time decay rate into semantic elements of quantifiable economic value parameters, risk transmission parameters, and time effect parameters through a natural language processing model; S12 applies a logarithmic conversion relationship to the value parameter to generate an amplitude modulation signal, the signal strength of which is nonlinearly positively correlated with the rule economic contribution degree; S13 applies an inverse proportional conversion relationship to the risk parameter to generate a frequency modulation signal, the signal change rate of which is inversely proportional to the risk transmission speed; S14 applies an inverse tangent conversion relationship to the time parameter to generate a phase modulation signal, the phase angle of which is nonlinearly mapped to the remaining effective period ratio, and is written into a non-volatile memory through the three-dimensional feature conversion module; S2 processes the feature vector using a multi-stage state machine, sequentially performing market behavior capture, rule space positioning, associated rule activation, topology structure adjustment, business strategy generation, and strategy persistence operation, each stage triggering state transition through hardware-level events, wherein the market behavior capture includes collecting user transaction behavior characteristics; S3 receives regulatory text data and constructs a dynamic constraint feedback loop, converts the requirements in the regulatory text data into a continuous control signal, and injects the control signal into the feature vector generation logic of the three-dimensional feature conversion module, forming a closed-loop adjustment mechanism for value gain calibration, risk decay compensation, and time synchronization correction; S4 double-verify the generated strategy through a physical signal cross-validation module, and activate the strategy execution when the waveform characteristics of the main verification channel and the timing characteristics of the auxiliary verification channel meet the preset coupling conditions; S5 maintains system timing synchronization, coordinates the operation rhythm of each module through a reference clock, so that the timing deviation of feature conversion, state transition, and constraint feedback is below the preset tolerance; the reference clock shares a clock source with the physical signal cross-validation module, and synchronizes the operation rhythm of each module through a phase-locked loop circuit.

2. The method for building an artificial intelligence-based inclusive financial service system according to claim 1, characterized in that: The dynamic adjustment of the logarithmic conversion relationship in S12 comprises: S121 automatically compresses the signal dynamic range according to the variance value of the price sequence in the user transaction behavior characteristics, the signal amplitude change gradient is negatively correlated with the financial market volatility; S122 starts the amplitude limiting mechanism when a rule conflict is detected, and constrains the maximum output signal to be less than the system safety threshold; S123 the three-dimensional feature conversion module has a built-in chip temperature sensor and a compensation circuit, the temperature sensor data is compensated by the compensation circuit in real time, and the signal reference value drift caused by external environmental temperature change is eliminated through the second-order filtering algorithm in the compensation circuit; S124 when the amplitude limiting mechanism is started, the dynamic range compression mechanism is automatically suspended.

3. The method for building an artificial intelligence-based inclusive financial service system according to claim 1, characterized in that: The control logic of the multi-stage state machine in S2 includes: S21, the market behavior capture stage, collects user transaction behavior characteristics at a fixed sampling frequency and converts them into digital signals, and the unit time frequency statistical value of the user transaction behavior characteristics is calculated in real time through a sliding window algorithm; S22, the rule space positioning stage, determines the current market state coordinates in a three-dimensional feature space using a nearest neighbor search algorithm; S23, the associated rule activation stage, compares the feature similarity of adjacent rule nodes in parallel, and establishes an associated link when the similarity exceeds the activation threshold; S24, the topology adjustment stage, dynamically adjusts the connection strength between rule nodes according to market changes, and the connection strength adjustment needs to meet the dynamic balance constraint rule, and through a normalization algorithm, the algebraic sum of the connection strength changes of all nodes is zero, and the top three links in the connection strength value are reserved.

4. The method for building an artificial intelligence-based inclusive financial service system according to claim 3, characterized in that: The execution process of the nearest neighbor search algorithm includes: S221, calculate the multi-dimensional space distance between the current market state and all rule nodes, and the multi-dimensional space distance is calculated using the three-dimensional coordinates of the value intensity, risk correlation degree, and time decay rate; S222, use the nearest neighbor search algorithm to accelerate the distance sorting process, and set the error tolerance to a preset threshold to reduce the computational complexity under the condition of ensuring accuracy; S223, cache the adjacent node information including the connection strength sorting result, which is used to support the operation of reserving the strength link and establish a fast access channel to shorten the response time of the subsequent associated rule activation stage.

5. The method for building an artificial intelligence-based inclusive financial service system according to claim 4, characterized in that: The construction process of the dynamic constraint feedback loop in S3 includes: S31, analyze the supervision text and extract the constraint elements, and encode them into control signals with time continuity; S32, eliminate high-frequency noise components in the control signal through a filtering algorithm, and retain effective adjustment components synchronized with market rhythm; S33, decompose the processed control signal into three independent adjustment channels, and apply them to feature correction in the three dimensions of value intensity, risk correlation degree, and time decay rate.

6. The method for building an artificial intelligence-based inclusive financial service system according to claim 5, characterized in that: The parameter adjustment logic of the filtering algorithm includes: S321, dynamically set the filter cutoff frequency according to the rule update frequency, so that the cutoff frequency value of the filter is in a linear positive correlation with the error tolerance threshold of the nearest neighbor search algorithm in S222; S322, automatically adjust the filtering strength based on the unit time frequency statistical value of the user transaction behavior characteristics, and realize through cutoff frequency adjustment, when the unit time frequency statistical value is higher than the preset threshold, the cutoff frequency is inversely proportional to the unit time frequency statistical value, wherein the preset threshold is dynamically adjusted based on the real-time frequency statistical value output by the sliding window algorithm in S21; S323, make the filtering timing strictly aligned with the system master clock through a hardware-level signal synchronization mechanism.

7. The method for building an artificial intelligence-based inclusive financial service system according to claim 1, characterized in that: The execution logic of the physical signal cross-validation in S4 includes: S41, the main verification channel analyzes the phase relationship of the feature vector, detects the timing consistency in the strategy generation process, and generates an alarm signal when the absolute value of the phase difference exceeds the threshold; S42, the auxiliary verification channel measures the duty cycle distribution of the pulse signal to verify the logical completeness of the strategy decision; S43 activate the strategy execution port and write a valid mark to the non-volatile memory when and only when the main verification channel does not generate an alarm signal and the review process of the auxiliary verification channel passes.

8. The system for building an artificial intelligence-based inclusive finance business system according to claim 1, characterized in that, Comprise the following modules: A three-dimensional feature conversion module for receiving financial service rule data and generating a rule feature vector containing value strength, risk correlation degree, and time decay rate, and dynamically expressing the three-dimensional feature relationship through analog modulation signals; Its built-in logarithmic conversion unit, inverse proportional conversion unit, and inverse tangent conversion unit convert economic value parameters, risk transmission parameters, and time impact parameters into amplitude, frequency, and phase modulation signals, and write them to non-volatile memory; The module integrates a temperature sensor and a second-order filter compensation circuit to eliminate signal drift caused by environmental temperature; A multi-stage state machine module uses a multi-stage control logic triggered by hardware-level events to sequentially perform market behavior capture, rule space positioning, associated rule activation, topology adjustment, business strategy generation, and persistent operation; it contains a sliding window algorithm unit to statistically analyze user transaction behavior frequency, a nearest neighbor search unit to locate market state coordinates, a parallel comparison unit to activate associated links, and a normalization adjustment unit to dynamically balance connection strength, retaining the top three links; A dynamic constraint feedback module for parsing regulatory text data and generating continuous control signals, which are decomposed into value gain calibration, risk decay compensation, and time synchronization correction after filtering algorithm to eliminate high-frequency noise; Its filter cutoff frequency is linearly related to the nearest neighbor search error tolerance, and the cutoff frequency is dynamically adjusted according to the user transaction behavior frequency, and aligned with the system master clock through a hardware-level synchronization mechanism; A physical signal cross-validation module contains a main verification channel and an auxiliary verification channel; the main channel verifies the consistency of the strategy timing through phase difference detection, and the auxiliary channel verifies the completeness of the logic through pulse duty cycle; after the double-channel signal coupling, the strategy execution port is activated, and a valid mark is written to the non-volatile memory; the module shares the clock source with the system reference clock, and maintains timing synchronization through a phase-locked loop circuit; A timing synchronization module coordinates the operation rhythm of each module through a reference clock to ensure that the timing deviation of feature conversion, state transfer, and constraint feedback is below the preset tolerance; it integrates a phase-locked loop circuit and a clock source distribution unit to achieve hardware-level synchronization of operation rhythm between modules; A strategy execution and storage module for receiving verified strategy signals and executing financial business operations, while persistently storing strategy parameters and execution results to non-volatile memory; the module contains a strategy activation port, a safety threshold limiting unit, and a fast access channel, supporting low-latency response in high-concurrency business scenarios.