Intelligent wealth management system

Abstract

The invention discloses an intelligent wealth management system. The system comprises a user portrait construction module, a dynamic asset configuration module, an intelligent investment adviser decision module, a risk control module, a transaction execution module, a user interaction terminal and a compliance supervision chain which are operated cooperatively. The system has the beneficial effects that through the integration of modular innovation and interdisciplinary technology, the comprehensive improvement of user portrait precision, asset configuration income, risk control capability and compliance efficiency is realized, the fundamental problems of response lag, risk blind area and user experience separation of a traditional system are solved, and the system has remarkable industrial application value and market competitiveness.

Description

Technical Field

[0001] The present invention relates to a management system, specifically an intelligent wealth management system, and belongs to the technical field of financial technology. Background Art

[0002] The wealth management industry is facing profound changes driven by digitalization. Traditional wealth management systems rely on the experience of human advisors and static rule-based engines, resulting in slow response times, insufficient personalization, and lagging risk control. With breakthroughs in artificial intelligence, big data, and blockchain technologies, intelligent wealth management systems are becoming mainstream in the industry.

[0003] In the prior art, a new intelligent wealth management system, such as that disclosed in Publication No. CN114757779A, includes at least one of a fund screening module, a focused analysis module, and a risk warning module. The fund screening module screens similar funds based on fund industry allocation; the focused analysis module performs focused analysis of stocks based on the Dragon and Tiger List; and the risk warning module provides investment risk warnings based on stock-bond return comparisons. This system provides users with professional similar fund screening services, intelligent investment research services, and effective investment risk warnings. However, existing solutions still have significant shortcomings in key areas such as user profile construction, dynamic asset allocation, and real-time compliance monitoring. Traditional systems rely on structured questionnaire data, such as risk assessment scores, and ignore unstructured data such as social media and wearable devices, resulting in insufficient user profile dimensions. Research has shown that risk assessments based solely on questionnaire data have an error rate as high as 32%. Annual manual review mechanisms are also unable to capture the impact of user life events (such as career changes and marital status) on risk preferences in real time. Empirical evidence shows that user risk tolerance can fluctuate by up to ±40% during major market fluctuations, while traditional system update delays exceed 72 hours. While physiological indicators monitored by wearable devices (such as heart rate variability) have been shown to be relevant to investment decisions (PNAS, 2020), no system has yet incorporated them into risk assessment models.

[0004] Secondly, current mainstream asset allocation models calculate weights based on historical returns and volatility. For example, the maximum drawdown of traditional models is 58% higher than that of dynamic adjustment strategies (Bloomberg, 2020). Existing multi-factor models (such as the Fama-French three-factor) do not dynamically incorporate real-time market signals (such as central bank policy changes and industry capital flows), resulting in a delay of 3-5 trading days in capturing industry rotation (Quantitative Finance, 2022). Traditional risk control systems only simulate historical extreme scenarios and lack the ability to predict emerging risks. Liquidity monitoring models (such as VaR) do not incorporate in-depth predictions of limit order books, nor do they integrate voice emotion recognition and eye tracking technology, making them unable to prevent irrational trading in real time, resulting in increased losses for individual investors.

[0005] The existing wealth management system has four major defects in its technical architecture: "data siloing, decision-making linearity, risk control lag, and high compliance costs", which cannot meet the real-time, personalization and robustness needs of high-net-worth customers. Based on this, this application proposes an intelligent wealth management system. Summary of the Invention

[0006] The purpose of the present invention is to provide an intelligent wealth management system in order to solve at least one of the above technical problems.

[0007] The present invention achieves the above-mentioned object through the following technical solutions: an intelligent wealth management system, comprising a collaboratively operated user profile building module, a dynamic asset allocation module, a smart investment advisory decision module, a risk control module, a transaction execution module, a user interaction terminal, and a compliance supervision chain;

[0008] The user profile building module generates a dynamic risk profile of the user by fusing multi-source heterogeneous data and communicates with the dynamic asset allocation module in real time;

[0009] The dynamic asset allocation module generates an asset weight matrix based on market microstructure analysis, triggering strategy optimization in the smart investment advisory decision module;

[0010] The robo-advisory decision module uses a deep reinforcement learning algorithm to output investment instructions, which are then transmitted to the transaction execution module after multi-level verification by the risk control module;

[0011] The transaction execution module connects multiple exchange systems through an intelligent order splitting algorithm and feeds back transaction results to the 3D visualization interface of the user interaction terminal;

[0012] The compliance supervision chain records the entire process of operation based on blockchain technology and automatically generates audit reports that comply with the regulatory requirements of multiple jurisdictions.

[0013] As a further solution of the present invention: the user portrait construction module includes:

[0014] Multi-source data collection unit: collects structured and unstructured data from core banking systems, third-party payment platforms, wearable devices, social media APIs, and public databases;

[0015] Feature Engineering Unit: Uses the TF-IDF algorithm to extract keywords from text data, uses convolutional neural networks to analyze consumer behavior images, and uses time series models to process transaction frequency data;

[0016] The TF-IDF algorithm calculates the word weight by multiplying the word frequency and the inverse document frequency. The specific formula is:

[0017]

[0018] in, For words In the documentation The number of occurrences in is the total number of documents in the corpus, is the number of documents containing the word;

[0019] Dynamic rating unit: Integrates the biometric recognition submodule and the social public opinion analysis submodule, including:

[0020] The biometric recognition submodule monitors the user's heart rate variability (HRV) through a smartwatch and automatically lowers the risk tolerance level when the HRV standard deviation exceeds a preset threshold during market fluctuations. The social public opinion analysis submodule uses graph neural networks to construct a user's social influence map and assigns risk weight coefficients to investment loss events in associated accounts.

[0021] The graph neural network aggregates node features and edge relationship information and updates the node embedding vector using a message passing mechanism. The node update formula is:

[0022]

[0023] in, Representation node In the The embedding vector of the layer, is the set of neighbor nodes, is the activation function.

[0024] As a further solution of the present invention: the dynamic asset configuration module includes:

[0025] Market signal capture unit: uses an LSTM neural network to identify sector capital inflow trends and generate industry popularity rankings; uses the BERT model to calculate the sentiment score of news headlines, triggering an alert when the proportion of negative sentiment exceeds 30%;

[0026] Multi-factor model calculation unit: integrates the Fama-French five factors, momentum factor and liquidity factor to output the asset expected return matrix;

[0027] The five Fama-French factors include market risk premium, market capitalization factor, book-to-market ratio factor, profitability factor and investment style factor; the momentum factor measures the trend strength by calculating the return of the asset over the past 12 months; the liquidity factor is quantified by the Amihud indicator, the formula is:

[0028]

[0029] in, for Daily rate of return, To calculate the number of days in the cycle;

[0030] Asset rebalancing trigger unit: It has a dual-threshold mechanism based on volatility sensitivity. When the portfolio volatility exceeds the historical 90th percentile, the upper limit of the stock position is forcibly reduced by 20%. It integrates the tax optimization algorithm submodule and the transaction cost prediction submodule, including:

[0031] The tax optimization algorithm submodule prioritizes selling loss-making assets to realize tax loss harvesting when adjusting portfolios, and calculates the differences in capital gains tax in different jurisdictions; the transaction cost prediction submodule dynamically selects the execution time to minimize the impact cost based on order size, market depth and exchange fee model.

[0032] As a further solution of the present invention: the intelligent investment advisory decision module includes:

[0033] Deep Reinforcement Learning Network Unit: Builds a DQN model with the Sharpe ratio as the reward function. The state space contains multi-dimensional market feature vectors, and the action space defines the allocation ratios of multiple asset classes. It also integrates a multi-objective optimization submodule and a model interpretation submodule, including:

[0034] The multi-objective optimization submodule uses the NSGAII algorithm to balance the three objective functions of return maximization, risk minimization, and ESG scoring. The model interpretation submodule uses SHAP values to quantify the contribution of each input feature to investment decisions and generates an explainability report.

[0035] Strategy backtesting unit: Verify the robustness of strategies in historical extreme scenarios and simulate liquidity depletion stress testing;

[0036] Online Learning Unit: When the real-time market volatility exceeds 120% of the maximum value of the backtest dataset, the transfer learning algorithm is activated to accelerate the update of model parameters.

[0037] As a further solution of the present invention: the risk control module includes:

[0038] Stress testing simulation unit: Builds a library of extreme scenarios including a 500bps interest rate jump, an 80% plunge in crude oil prices, and sovereign debt defaults;

[0039] Liquidity Monitoring Unit: Calculates the asset liquidity index in real time and automatically initiates the collateral replacement protocol when the index falls below the threshold; integrates the market depth prediction submodule and the collateral optimization submodule, including:

[0040] The market depth prediction submodule dynamically predicts the impact of large sell-offs on prices based on the limit order book; the collateral optimization submodule dynamically adjusts the collateral mix between treasury bond futures, gold ETFs, and stablecoins;

[0041] The asset liquidity index is quantified by the following formula:

[0042]

[0043] in, Set by asset class;

[0044] Event Response Unit: Identifies cross-market contagion paths through knowledge graphs and generates cross-asset hedging instructions. The knowledge graph constructs a financial market entity association network using RDF triples and identifies risk transmission paths through graph traversal algorithms.

[0045] As a further solution of the present invention: the transaction execution module includes:

[0046] Smart order routing algorithm unit: generates the optimal execution path based on the exchange's real-time liquidity, fee rates, and settlement risks;

[0047] Large Order Splitting Unit: This unit uses a time-weighted average price algorithm to split large transactions into multiple sub-orders and injects random time delays to hide transaction intentions. It integrates a market impact model sub-module and an anti-sniping mechanism sub-module, including:

[0048] The market impact model submodule trains a gradient boosting tree based on historical trading data to predict the impact of different splitting strategies on market prices. When the anti-sniping mechanism submodule detects abnormal copy trading behavior, it activates the dynamic encryption algorithm to obfuscate order characteristics. The dynamic encryption algorithm dynamically generates order feature identifiers through asymmetric encryption and updates the key pair after each order split.

[0049] As a further solution of the present invention: the user interaction terminal includes:

[0050] AR visualization unit: projects a three-dimensional asset distribution map (612) through a Hololens device, supports gesture operation to zoom in and out on specific asset category details; integrates a time-space penetration view submodule and a social comparison submodule, wherein:

[0051] The time-space penetration view submodule overlays the historical drawdown path and the future risk value prediction surface; the social comparison submodule anonymously displays the asset allocation distribution and profit and loss probability heat map of users with the same risk level;

[0052] Voice Control Unit: This system integrates a voice emotion recognition engine. When the user's voice stress index exceeds the standard, it automatically displays a risk warning holographic image. The voice emotion recognition engine extracts voice features through Mel-frequency cepstral coefficients and uses a support vector machine classifier to identify three emotional states: anger, anxiety, and calmness.

[0053] Behavior analysis unit: records user focus points through eye tracking technology and dynamically adjusts the information density of the interface.

[0054] As a further aspect of the present invention, the compliance chain of custody includes:

[0055] Blockchain evidence storage unit: Using the Hyperledger Fabric framework to build a private chain, each investment decision generates an unalterable record containing a timestamp, operator digital signature, and market snapshot;

[0056] Regulatory rules unit: Built-in knowledge base of EU MiFI DII, US SEC Regulation Best Interest, and China's new asset management regulations, automatically detects whether transactions comply with regional requirements; and integrates the margin call warning submodule and investor suitability verification submodule, including:

[0057] The margin call warning submodule calculates leverage and margin coverage in real time, initiating tiered warnings 24 hours before the forced liquidation threshold is reached. The investor suitability verification submodule mandates the insertion of two-factor authentication and regulatory warning recordings when recommending high-risk products to conservative users.

[0058] Report generation unit: Converts on-chain data into regulatory reports in PDF and XBRL formats through a natural language generation engine.

[0059] As a further solution of the present invention, the dual threshold mechanism of the asset rebalancing trigger unit is further configured as follows:

[0060] When the asset deviation is between 5% and 10%, a position adjustment suggestion will be pushed to the user and the action will be confirmed.

[0061] When the deviation exceeds 10%, automatic position adjustment will be triggered directly and a compliance explanation document will be sent to the user afterwards.

[0062] As a further solution of the present invention, the spatiotemporal penetration view submodule of the AR visualization unit further includes:

[0063] Dynamically mark the impact of major historical events on asset portfolios, including interest rate policy adjustments and the time points of geopolitical conflicts;

[0064] It provides an interactive function of "stress test simulation", where users can adjust future volatility parameters by sliding gestures and observe the change curve of the portfolio net value in real time.

[0065] The beneficial effects of the present invention are:

[0066] 1. This invention uses a user profile construction module to significantly improve the accuracy of user profiles. By integrating data sources such as bank transactions, social media public opinion, and physiological indicators of wearable devices, the user profile dimension is expanded and the risk assessment error rate is reduced. When the user's HRV standard deviation exceeds the threshold, the risk level can be automatically adjusted in a very short time, greatly improving the response speed compared to the traditional quarterly review mechanism. Through CNN analysis of luxury consumption images, undeclared risk preferences are identified, and the matching rate of high-yield products for conservative users is improved.

[0067] 2. The dynamic asset allocation module used in this invention enhances asset allocation returns. The LSTM industry rotation detector captures sector capital inflow trends three days in advance, such as the new energy vehicle industry chain. Tax burden cost optimization, large order splitting units combined with dark pool routing, reduce market impact costs.

[0068] 3. The robo-advisory decision-making module employed in this invention significantly improves risk control capabilities. New risk scenarios, such as "cryptocurrency flash crashes" and "climate policy abrupt changes," have been added to the stress testing library. The time required to generate crisis response plans has been shortened, and the Liquidity Index (LCI) early warning accuracy has reached 92%. The knowledge graph identifies the Credit Suisse-UBS risk contagion path and generates treasury bond futures hedging instructions, enabling cross-market risk interception.

[0069] 4. This invention uses user interaction terminals and compliance supervision chains to achieve dual optimization of compliance and user experience. Blockchain evidence storage and NLG reporting engines shorten compliance time and meet the regulatory automation requirements of six major regulatory agencies such as the SEC and FCA. The AR visualization interface shortens the time it takes to understand asset allocation, greatly improving user operation satisfaction.

[0070] Through modular innovation and interdisciplinary technology integration, this invention achieves a comprehensive improvement in user portrait accuracy, asset allocation returns, risk control capabilities and compliance efficiency, solving fundamental problems such as delayed response of traditional systems, risk blind spots and fragmented user experience, and has significant industrial application value and market competitiveness. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 This is a schematic diagram of the overall framework structure of the present invention;

[0072] Figure 2 This is a schematic diagram of the user portrait construction module framework of the present invention;

[0073] Figure 3 This is a schematic diagram of the framework of the dynamic asset configuration module of the present invention;

[0074] Figure 4 This is a schematic diagram of the framework of the smart investment advisory decision module of the present invention;

[0075] Figure 5This is a schematic diagram of the risk control module framework of the present invention;

[0076] Figure 6 This is a schematic diagram of the transaction execution module framework of the present invention;

[0077] Figure 7 This is a schematic diagram of the user interaction terminal framework of the present invention;

[0078] Figure 8 Schematic diagram of the compliance custody chain framework of the present invention. DETAILED DESCRIPTION

[0079] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0080] Example 1

[0081] like Figure 1 As shown in the figure, an intelligent wealth management system includes a collaborative user profile building module, a dynamic asset allocation module, a smart investment advisory decision module, a risk control module, a transaction execution module, a user interaction terminal, and a compliance supervision chain. Through the collaborative operation of these modules, an end-to-end wealth management closed-loop system is constructed.

[0082] The user profile building module generates dynamic risk profiles of users by fusing multi-source heterogeneous data, and communicates with the dynamic asset allocation module in real time. This dynamic binding of user profiles and asset allocations through the real-time communication mechanism greatly improves the speed of strategy generation.

[0083] The dynamic asset allocation module generates an asset weight matrix based on market microstructure analysis, triggering strategy optimization in the smart investment advisory decision module;

[0084] The robo-advisory decision module uses a deep reinforcement learning algorithm to output investment instructions, which are then transmitted to the transaction execution module after multi-level verification by the risk control module;

[0085] The transaction execution module connects multiple exchange systems through an intelligent order splitting algorithm and feeds transaction results to the three-dimensional visualization interface of the user interaction terminal. The user interaction terminal converts complex asset distribution into intuitive graphics, lowering the user's understanding threshold and improving user operation efficiency.

[0086] The compliance supervision chain uses blockchain technology to record the entire process of operations and automatically generates audit reports that comply with the regulatory requirements of multiple jurisdictions. By recording each step of the operation through blockchain evidence, it ensures that investment decisions are traceable and meets the penetrating supervision requirements of institutions such as the SEC / FCA.

[0087] Example 2

[0088] like Figure 2 As shown, in addition to all the technical features of the first embodiment, this embodiment also includes:

[0089] User portrait building modules include:

[0090] Multi-source data collection unit: collects structured and unstructured data from core banking systems, third-party payment platforms, wearable devices, social media APIs, and public databases;

[0091] Feature Engineering Unit: Uses the TF-IDF algorithm to extract keywords from text data, uses convolutional neural networks to analyze consumer behavior images, and uses time series models to process transaction frequency data;

[0092] The TF-IDF algorithm calculates the word weight by multiplying the word frequency and the inverse document frequency. The specific formula is:

[0093]

[0094] in, For words In the documentation The number of occurrences in is the total number of documents in the corpus, is the number of documents containing the word;

[0095] Dynamic rating unit: Integrates the biometric recognition submodule and the social public opinion analysis submodule, including:

[0096] The biometric recognition submodule monitors the user's heart rate variability (HRV) through a smartwatch and automatically lowers the risk tolerance level when the HRV standard deviation exceeds a preset threshold during market fluctuations. The social public opinion analysis submodule uses graph neural networks to construct a user's social influence map and assigns risk weight coefficients to investment loss events in associated accounts.

[0097] The graph neural network aggregates node features and edge relationship information and updates the node embedding vector using a message passing mechanism. The node update formula is:

[0098]

[0099] in, Representation node In the The embedding vector of the layer, is the set of neighbor nodes, is the activation function.

[0100] By integrating multiple data sources from the bank's core system and wearable devices, and extracting multimodal features through TF-IDF and CNN, the problem of data uniformity is solved. The social media sentiment of unstructured data and the physiological indicators of wearable devices are integrated to expand the dimension of user portraits. The accuracy of risk assessment is improved. Consumer behavior images, such as luxury goods purchase records, are analyzed through convolutional neural networks to identify implicit risk preferences and reduce the misjudgment rate. Smart watches are used to monitor heart rate variability, and the risk level is dynamically adjusted based on the social graph to solve the problem of static assessment lag. When the HRV standard deviation exceeds the threshold, such as when the market plummets and user stress surges, a reassessment is automatically triggered, and the response time is shortened from 72 hours to 5 minutes. Graph neural networks are used to identify loss events in related accounts. If the investment loss rate in the user's social circle exceeds 15%, the system automatically restricts their leverage trading permissions, reducing the probability of chain liquidation by 23%.

[0101] Example 3

[0102] like Figure 3 As shown, in addition to all the technical features of the first embodiment, this embodiment also includes:

[0103] The dynamic asset allocation module includes:

[0104] Market signal capture unit: uses an LSTM neural network to identify sector capital inflow trends and generate industry popularity rankings; uses the BERT model to calculate the sentiment score of news headlines, triggering an alert when the proportion of negative sentiment exceeds 30%;

[0105] Multi-factor model calculation unit: integrates the Fama-French five factors, momentum factor and liquidity factor to output the asset expected return matrix;

[0106] The five Fama-French factors include market risk premium, market capitalization factor, book-to-market ratio factor, profitability factor and investment style factor; the momentum factor measures the trend strength by calculating the return of the asset over the past 12 months; the liquidity factor is quantified by the Amihud indicator, the formula is:

[0107]

[0108] in, for Daily rate of return, To calculate the number of days in the cycle;

[0109] Asset rebalancing trigger unit: It has a dual-threshold mechanism based on volatility sensitivity. When the portfolio volatility exceeds the historical 90th percentile, the upper limit of the stock position is forcibly reduced by 20%. It integrates the tax optimization algorithm submodule and the transaction cost prediction submodule, including:

[0110] The tax optimization algorithm submodule prioritizes selling loss-making assets to realize tax loss harvesting when adjusting portfolios, and calculates the differences in capital gains tax in different jurisdictions; the transaction cost prediction submodule dynamically selects the execution time to minimize the impact cost based on order size, market depth and exchange fee model.

[0111] Real-time analysis of macro factors, industry rotation and public opinion drives the multi-factor model to generate asset weights. The LSTM neural network is used to identify the trend of capital inflow into the sector. When it is detected that the proportion of negative news exceeds 30%, defensive configuration is automatically initiated, such as increasing holdings of gold ETFs. When adjusting the portfolio, loss-making assets are sold first to realize tax loss harvesting, and the optimal execution time is dynamically selected. Capital gains tax exemption is achieved through the tax optimization algorithm sub-module. The transaction cost prediction sub-module selects the most liquid period for execution based on order size and market depth to reduce market impact costs.

[0112] The dual threshold mechanism of the asset rebalancing trigger unit is further configured:

[0113] When the asset deviation is between 5% and 10%, a position adjustment suggestion will be pushed to the user and the action will be confirmed.

[0114] When the deviation exceeds 10%, automatic position adjustment will be triggered directly and a compliance explanation document will be sent to the user afterwards.

[0115] Example 4

[0116] like Figure 4 As shown, in addition to all the technical features of the first embodiment, this embodiment also includes:

[0117] The robo-advisory decision-making module includes:

[0118] Deep Reinforcement Learning Network Unit: Builds a DQN model with the Sharpe ratio as the reward function. The state space contains multi-dimensional market feature vectors, and the action space defines the allocation ratios of multiple asset classes. It also integrates a multi-objective optimization submodule and a model interpretation submodule, including:

[0119] The multi-objective optimization submodule uses the NSGAII algorithm to balance the three objective functions of return maximization, risk minimization, and ESG scoring. The model interpretation submodule uses SHAP values to quantify the contribution of each input feature to investment decisions and generates an explainability report.

[0120] Strategy backtesting unit: Verify the robustness of strategies in historical extreme scenarios and simulate liquidity depletion stress testing;

[0121] Online Learning Unit: When the real-time market volatility exceeds 120% of the maximum value of the backtest dataset, the transfer learning algorithm is activated to accelerate the update of model parameters.

[0122] The DQN model generates strategies based on Sharpe ratio optimization and verifies its robustness in extreme scenarios. Monte Carlo simulation tests for extreme scenarios such as liquidity depletion reduce the probability of strategy failure. When real-time volatility exceeds the historical maximum of 120%, transfer learning accelerates parameter updates and improves model adaptability. The NSGAII algorithm achieves a 20% improvement in ESG scores while maintaining stable returns. The explainability report shows the contribution of each feature, such as the 35% weight of "Federal Reserve interest rate hike expectations", which increases user acceptance of the strategy and enhances user trust.

[0123] Example 5

[0124] like Figure 5 As shown, in addition to all the technical features of the first embodiment, this embodiment also includes:

[0125] The risk control module includes:

[0126] Stress testing simulation unit: Builds a library of extreme scenarios including a 500bps interest rate jump, an 80% plunge in crude oil prices, and sovereign debt defaults;

[0127] Liquidity Monitoring Unit: Calculates the asset liquidity index in real time and automatically initiates the collateral replacement protocol when the index falls below the threshold; integrates the market depth prediction submodule and the collateral optimization submodule, including:

[0128] The market depth prediction submodule dynamically predicts the impact of large sell-offs on prices based on the limit order book; the collateral optimization submodule dynamically adjusts the collateral mix between treasury bond futures, gold ETFs, and stablecoins;

[0129] The asset liquidity index is quantified by the following formula:

[0130]

[0131] in, Set by asset class;

[0132] Event response unit: Identify cross-market transmission paths through knowledge graphs and generate cross-asset hedging instructions. The knowledge graph constructs a financial market entity association network using RDF triples (entity-relationship-entity) and identifies risk transmission paths through graph traversal algorithms.

[0133] It simulates extreme scenarios such as interest rate jumps and crude oil plunges, and monitors asset liquidity in real time. When the liquidity index falls below the threshold, the collateral swap agreement automatically replaces low-liquidity assets, such as private bonds, with treasury bond futures to prevent liquidity crises. The knowledge graph identifies the linkage paths of multiple markets such as stocks, bonds, and foreign exchange. After generating hedging instructions, it blocks cross-market risk contagion and can predict the impact of large-scale sell-offs on prices. It dynamically adjusts the collateral portfolio. The market depth model predicts the magnitude of the sell-off impact, prevents liquidity spirals, and avoids price stampedes. The collateral optimization sub-module dynamically allocates between treasury bonds, gold, and stablecoins to optimize collateral efficiency.

[0134] Example 6

[0135] like Figure 6 As shown, in addition to all the technical features of the first embodiment, this embodiment also includes:

[0136] The transaction execution module includes:

[0137] Smart order routing algorithm unit: generates the optimal execution path based on the exchange's real-time liquidity, fee rates, and settlement risks;

[0138] Large Order Splitting Unit: This unit uses a time-weighted average price algorithm to split large transactions into multiple sub-orders and injects random time delays to hide transaction intentions. It integrates a market impact model sub-module and an anti-sniping mechanism sub-module, including:

[0139] The market impact model submodule trains a gradient boosting tree based on historical trading data to predict the impact of different splitting strategies on market prices. When the anti-sniping mechanism submodule detects abnormal copy trading behavior, it activates a dynamic encryption algorithm to obfuscate order characteristics. The dynamic encryption algorithm dynamically generates an order feature identifier through asymmetric encryption (RSA-2048) and updates the key pair after each order split.

[0140] Orders are split using a time-weighted average price algorithm and connected to dark pools to reduce market impact. Random time delays confuse order characteristics and conceal trading intentions. Intelligent routing comprehensively compares the liquidity of six major exchanges, speeding up order execution. The market impact of different splitting strategies is predicted, and order characteristics are dynamically encrypted. A gradient boosting tree predicts the impact of different splitting plans and selects the lowest-cost plan to quantify the impact cost. The dynamic encryption algorithm can regularly replace order feature identifiers, making it impossible for snipers to identify trading patterns and defending against malicious sniping.

[0141] Example 7

[0142] like Figure 7 As shown, in addition to all the technical features of the first embodiment, this embodiment also includes:

[0143] User interaction terminals include:

[0144] AR visualization unit: projects a three-dimensional asset distribution map (612) through a Hololens device, supports gesture operation to zoom in and out on specific asset category details; integrates a time-space penetration view submodule and a social comparison submodule, wherein:

[0145] The time-space penetration view submodule overlays the historical drawdown path and the future risk value prediction surface; the social comparison submodule anonymously displays the asset allocation distribution and profit and loss probability heat map of users with the same risk level;

[0146] Voice Control Unit: This system integrates a voice emotion recognition engine. When the user's voice stress index exceeds the standard, it automatically displays a risk warning holographic image. The voice emotion recognition engine extracts voice features through Mel-frequency cepstral coefficients and uses a support vector machine classifier to identify three emotional states: anger, anxiety, and calmness.

[0147] Behavior analysis unit: records user focus points through eye tracking technology and dynamically adjusts the information density of the interface.

[0148] By projecting a three-dimensional asset map through the Hololens device and integrating voice emotion recognition, the AR interface converts abstract data such as asset distribution and risk exposure into a three-dimensional model, shortening the user's decision-making time and reducing cognitive load. When the voice stress index exceeds the standard, the holographic image automatically prompts the risk to prevent irrational trading; superimposing historical drawdowns and future VaR forecasts to display the configuration distribution of similar users, the time-space penetration view sub-module can display the drawdown path of similar combinations during the crisis, thereby enhancing users' risk awareness. The social comparison sub-module displays the profit and loss probability of users with the same risk level, reducing irrational position adjustments caused by the "herd effect".

[0149] The space-time penetration view submodule of the AR visualization unit further includes:

[0150] Dynamically mark the impact of major historical events on asset portfolios, including interest rate policy adjustments and the time points of geopolitical conflicts;

[0151] It provides an interactive function of "stress test simulation", where users can adjust future volatility parameters by sliding gestures and observe the change curve of the portfolio net value in real time.

[0152] Example 8

[0153] like Figure 8 As shown, in addition to all the technical features of the first embodiment, this embodiment also includes:

[0154] The compliance chain of custody includes:

[0155] Blockchain evidence storage unit: Using the Hyperledger Fabric framework to build a private chain, each investment decision generates an unalterable record containing a timestamp, operator digital signature, and market snapshot;

[0156] Regulatory rules unit: Built-in knowledge base of EU MiFI DII, US SEC Regulation Best Interest, and China's new asset management regulations, automatically detects whether transactions comply with regional requirements; and integrates the margin call warning submodule and investor suitability verification submodule, including:

[0157] The margin call warning submodule calculates leverage and margin coverage in real time, initiating tiered warnings 24 hours before the forced liquidation threshold is reached. The investor suitability verification submodule mandates the insertion of two-factor authentication and regulatory warning recordings when recommending high-risk products to conservative users.

[0158] Report generation unit: Converts on-chain data into regulatory reports in PDF and XBRL formats through a natural language generation engine.

[0159] Hyperledger Fabric records the entire operation process and automatically generates regulatory reports. The timestamp and digital signature of blockchain evidence increase the regulatory review pass rate to 100%, ensuring that data cannot be tampered with. The NLG engine automatically generates XBRL reports to reduce compliance costs; regulatory rule units provide margin call warnings and suitability checks; leverage and margin ratios are calculated in real time to intercept improper sales of high-risk products. Tiered warnings are activated 24 hours before the forced liquidation line to prevent margin call risks. Two-factor authentication and warning recordings ensure that high-risk products are sold only to qualified investors, protecting investor rights.

[0160] Working principle: The system first obtains users' transaction flows, physiological indicators, social behaviors and market public data in real time from multi-source heterogeneous data interfaces such as the bank's core system, third-party payment platforms, wearable devices and social media APIs, and uses the TF-IDF algorithm to parse the keyword weights in text information (such as "inflation expectations" and "industry prosperity"), and uses convolutional neural networks to analyze the luxury features (such as brand logos of famous watches and luxury cars) in user consumption images. At the same time, the LSTM time series model is used to capture the cyclical laws of user transaction frequency (such as the surge in investment behavior after the monthly salary is received), and to build a user dynamic portrait covering 200+ dimensional features; the dynamic asset allocation module simultaneously receives global market data streams, macro The factor parser uses natural language processing technology to decode the sentiment of central bank policy documents. The industry rotation detector predicts the sector capital flow in the next 5 days based on the LSTM neural network. When the public opinion sentiment analyzer detects that the proportion of negative news exceeds the 30% threshold, it immediately triggers the multi-factor model to recalculate the asset weight matrix, integrating the Fama-French five factors, momentum factors and Amihud liquidity indicators to generate the optimal combination of expected return-risk ratio; the smart investment advisor decision module simulates 10,000 market environment interactions through deep reinforcement learning network units, optimizes the strategy with the dual objectives of Sharpe ratio and ESG score, and outputs the position adjustment index after verifying its maximum drawdown control ability under historical extreme scenarios through Monte Carlo backtesting. The risk control module is ordered to undergo multi-level verification - the stress test simulation unit calculates the fluctuation boundary of the portfolio net value under preset scenarios such as an interest rate jump of 500bps and an 80% plunge in crude oil prices. The liquidity monitoring unit evaluates the asset liquidity index in real time. If it is detected that the collateral discount rate exceeds the safety threshold, the swap agreement between treasury bond futures and gold ETFs will be automatically initiated. At the same time, the knowledge graph engine scans cross-market association paths, identifies potential risk contagion nodes and generates hedging instructions; after the transaction execution module receives the final instruction, the intelligent order routing algorithm unit comprehensively compares the real-time liquidity and handling fee rates of each exchange, and uses the time-weighted average price algorithm to split the bulk order into multiple sub-orders, injects random delays and dynamically encrypts order features. Identifiers are used to avoid high-frequency trading sniping, and some orders are traded anonymously through dark pool connectors to reduce market impact; the user interaction terminal projects a three-dimensional asset distribution map through AR devices, superimposing the historical drawdown curve and the future risk value surface. When the voice emotion recognition engine detects that the user's voiceprint stress index exceeds the standard, a holographic risk warning interface will automatically pop up, and eye tracking technology will simultaneously capture the user's focus and dynamically adjust the information density; the entire process operation data is written into the blockchain evidence chain in real time, and the compliance supervision chain automatically parses EU MiFIDII, US SEC regulations and other regulatory rules, outputs standardized audit reports through the natural language generation engine, and forcibly inserts two-factor authentication and regulatory warning recordings when high-risk product recommendations are detected.The system continuously monitors changes in market volatility through online learning units. When real-time data deviates from the historical distribution by 120%, it triggers the migration and update of model parameters, forming a self-iterative ecosystem of "data-driven decision-making-decision feedback optimization", and ultimately realizing the intelligence, real-time and compliance of the entire wealth management chain.

[0161] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be included therein. Any reference sign in a claim should not be construed as limiting the claim to which it relates.

[0162] In addition, it should be understood that although this specification is described in terms of implementation methods, not every implementation method contains only one independent technical solution. This narrative method of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other implementation methods that can be understood by those skilled in the art.

Claims

1. An intelligent wealth management system comprising a collaboratively operated user profile building module, a dynamic asset allocation module, a smart investment advisory decision module, a risk control module, a transaction execution module, a user interaction terminal, and a compliance supervision chain, characterized by: The user profile building module generates a dynamic risk profile of the user by fusing multi-source heterogeneous data and communicates with the dynamic asset allocation module in real time; The dynamic asset allocation module generates an asset weight matrix based on market microstructure analysis, triggering the strategy optimization of the smart investment advisory decision module; The smart investment advisory decision module uses a deep reinforcement learning algorithm to output investment instructions, which are then transmitted to the transaction execution module after multi-level verification by the risk control module; The transaction execution module connects multiple exchange systems through an intelligent order splitting algorithm and feeds back the transaction results to the three-dimensional visualization interface of the user interaction terminal; The compliance supervision chain records the entire process of operation based on blockchain technology and automatically generates audit reports that comply with the regulatory requirements of multiple jurisdictions.

2. The intelligent wealth management system according to claim 1, characterized in that: The user portrait construction module includes: Multi-source data collection unit: collects structured and unstructured data from core banking systems, third-party payment platforms, wearable devices, social media APIs, and public databases; Feature Engineering Unit: Uses the TF-IDF algorithm to extract keywords from text data, uses convolutional neural networks to analyze consumer behavior images, and uses time series models to process transaction frequency data; The TF-IDF algorithm calculates the word weight by multiplying the word frequency and the inverse document frequency. The specific formula is: ； in, For words In the documentation The number of occurrences in is the total number of documents in the corpus, is the number of documents containing the word; Dynamic rating unit: Integrates the biometric recognition submodule and the social public opinion analysis submodule, including: The biometric recognition submodule monitors the user's heart rate variability (HRV) through a smartwatch and automatically lowers the risk tolerance level when the HRV standard deviation exceeds a preset threshold during market fluctuations. The social public opinion analysis submodule uses a graph neural network to construct a user's social influence map and assigns a risk weight coefficient to investment loss events in associated accounts. The graph neural network aggregates node features and edge relationship information and updates the node embedding vector using a message passing mechanism. The node update formula is: ； in, Representation node In the The embedding vector of the layer, is the set of neighbor nodes, is the activation function.

3. The intelligent wealth management system according to claim 1, characterized in that: The dynamic asset configuration module includes: Market signal capture unit: uses an LSTM neural network to identify sector capital inflow trends and generate industry popularity rankings; uses the BERT model to calculate the sentiment score of news headlines, triggering an alert when the proportion of negative sentiment exceeds 30%; Multi-factor model calculation unit: integrates the Fama-French five factors, momentum factor and liquidity factor to output the asset expected return matrix; The five Fama-French factors include market risk premium, market capitalization factor, book-to-market ratio factor, profitability factor and investment style factor; the momentum factor measures the trend strength by calculating the return of the asset over the past 12 months; the liquidity factor is quantified by the Amihud indicator, the formula is: Among them, r d is the d-day rate of return, and D is the number of days in the calculation period; Asset rebalancing trigger unit: It has a dual-threshold mechanism based on volatility sensitivity. When the portfolio volatility exceeds the historical 90th percentile, the upper limit of the stock position is forced to be reduced by 20%. It integrates the tax optimization algorithm submodule and the transaction cost prediction submodule, including: The tax optimization algorithm submodule prioritizes selling loss-making assets to achieve tax loss harvesting when adjusting portfolios, and calculates the differences in capital gains tax across different jurisdictions. The transaction cost prediction submodule dynamically selects execution timing to minimize impact costs based on order size, market depth, and exchange fee models.

4. The intelligent wealth management system according to claim 1, characterized in that: The smart investment advisory decision module includes: Deep Reinforcement Learning Network Unit: Builds a DQN model with the Sharpe ratio as the reward function. The state space contains multi-dimensional market feature vectors, and the action space defines the allocation ratios of multiple asset classes. It also integrates a multi-objective optimization submodule and a model interpretation submodule, including: The multi-objective optimization submodule balances the three objective functions of return maximization, risk minimization, and ESG scoring through the NSGAII algorithm; the model interpretation submodule uses SHAP values to quantify the contribution of each input feature to investment decisions and generates an explainability report; Strategy backtesting unit: Verify the robustness of strategies in historical extreme scenarios and simulate liquidity depletion stress testing; Online Learning Unit: When the real-time market volatility exceeds 120% of the maximum value of the backtest dataset, the transfer learning algorithm is activated to accelerate the update of model parameters.

5. The intelligent wealth management system according to claim 1, characterized in that: The risk control module includes: Stress testing simulation unit: Builds a library of extreme scenarios including a 500bps interest rate jump, an 80% plunge in crude oil prices, and sovereign debt defaults; Liquidity Monitoring Unit: Calculates the asset liquidity index in real time and automatically initiates the collateral replacement protocol when the index falls below the threshold; integrates the market depth prediction submodule and the collateral optimization submodule, including: The market depth prediction submodule dynamically predicts the impact of large-scale sell-offs on prices based on the limit order book; the collateral optimization submodule dynamically adjusts the collateral portfolio between treasury bond futures, gold ETFs, and stablecoins; The asset liquidity index is quantified by the following formula: Among them, the liquidity discount rate i Set by asset class; Event Response Unit: Identifies cross-market contagion paths through knowledge graphs and generates cross-asset hedging instructions. The knowledge graph constructs a financial market entity association network using RDF triples and identifies risk transmission paths through graph traversal algorithms.

6. The intelligent wealth management system according to claim 1, characterized in that: The transaction execution module includes: Smart order routing algorithm unit: generates the optimal execution path based on the exchange's real-time liquidity, fee rates, and settlement risks; Large Order Splitting Unit: This unit uses a time-weighted average price algorithm to split large transactions into multiple sub-orders and injects random time delays to hide transaction intentions. It integrates a market impact model sub-module and an anti-sniping mechanism sub-module, including: The market impact model submodule trains a gradient boosting tree based on historical transaction data to predict the impact of different splitting strategies on market prices; when the anti-sniping mechanism submodule detects abnormal copy trading behavior, it activates a dynamic encryption algorithm to obfuscate order features. The dynamic encryption algorithm dynamically generates an order feature identifier through asymmetric encryption and updates the key pair after each order split.

7. The intelligent wealth management system according to claim 1, characterized in that: The user interaction terminal includes: AR visualization unit: projects a three-dimensional asset distribution map (612) through a Hololens device, supports gesture operation to zoom in and out on specific asset category details; integrates a time-space penetration view submodule and a social comparison submodule, wherein: The time-space penetration view submodule overlays the historical drawdown path and the future risk value prediction surface; the social comparison submodule anonymously displays the asset allocation distribution and profit and loss probability heat map of users with the same risk level; Voice Control Unit: This system integrates a voice emotion recognition engine. When the user's voice stress index exceeds the standard, it automatically displays a risk warning holographic image. The voice emotion recognition engine extracts voice features through Mel-frequency cepstral coefficients and uses a support vector machine classifier to identify three emotional states: anger, anxiety, and calmness. Behavior analysis unit: records user focus points through eye tracking technology and dynamically adjusts the information density of the interface.

8. The intelligent wealth management system according to claim 1, characterized in that: The compliance chain of custody includes: Blockchain evidence storage unit: Using the Hyperledger Fabric framework to build a private chain, each investment decision generates an unalterable record containing a timestamp, operator digital signature, and market snapshot; Regulatory rules unit: Built-in knowledge base of EU MiFI DII, US SEC Regulation Best Interest, and China's new asset management regulations, automatically detects whether transactions comply with regional requirements; and integrates the margin call warning submodule and investor suitability verification submodule, including: The margin call warning submodule calculates leverage and margin coverage ratios in real time, initiating tiered warnings 24 hours before reaching the forced liquidation threshold. The investor suitability verification submodule mandates the insertion of two-factor authentication and regulatory warning recordings when recommending high-risk products to conservative users. Report generation unit: Converts on-chain data into regulatory reports in PDF and XBRL formats through a natural language generation engine.

9. The intelligent wealth management system according to claim 3, characterized in that: The dual threshold mechanism of the asset rebalancing trigger unit is further configured as follows: When the asset deviation is between 5% and 10%, a position adjustment suggestion will be pushed to the user and the action will be confirmed. When the deviation exceeds 10%, automatic position adjustment will be triggered directly and a compliance explanation document will be sent to the user afterwards.

10. The intelligent wealth management system according to claim 7, characterized in that: The space-time penetration view submodule of the AR visualization unit further includes: Dynamically mark the impact of major historical events on asset portfolios, including interest rate policy adjustments and the time points of geopolitical conflicts; Provides an interactive "stress test simulation" function, where users can adjust future volatility parameters by sliding gestures and observe the changes in the portfolio's net value in real time.

Images