Intelligent post-investment management method and system

Abstract

The invention provides an intelligent post-investment management method and system, and the method comprises the steps: S1, obtaining structured financial data and unstructured financial information, and constructing a multi-source heterogeneous map; s2, carrying out situation modeling on the financial event based on a multi-source heterogeneous graph through a graph neural network, and extracting a causal situation vector; s3, constructing a deep reinforcement learning model containing a plurality of optimization targets, and generating a candidate strategy set based on the extracted causal scene vector; s4, performing crossover variation, fitness evaluation and optimization on the candidate strategy set by using an evolution strategy algorithm; s5, identifying the market situation in real time, and matching the strategy version with the highest adaptability to carry out asset configuration adjustment; wherein the structured financial data comprises a net value, a yield rate and an industry ratio; the unstructured financial information comprises policy announcements and news texts.

Description

Technical Field

[0001] The present invention relates to the interdisciplinary field of financial technology and artificial intelligence, and more specifically, to an intelligent post-investment management method and system, and more specifically, to a scenario-driven multi-objective strategy evolution system and its application method in intelligent asset portfolio management. Background Art

[0002] Existing post-investment management systems generally rely on fixed rules or static models, such as Black-Litterman and mean-variance models, for asset allocation and risk control. These models lack the ability to adaptively perceive and dynamically respond to unexpected market scenarios. As the financial environment becomes increasingly complex, traditional models often lag in responding to unstructured factors such as policy changes and emergencies, resulting in untimely strategy adjustments and inadequate risk prevention and control.

[0003] In addition, many current AI-enabled systems focus more on pre-investment screening or trend prediction, and there is still a lack of effective solutions for the "strategy evolution" and "real-time self-adjustment" of post-investment portfolios.

[0004] Patent document CN118735691A (application number: 202411223681.8) discloses an intelligent post-investment management method and system based on asset monitoring and dynamic balance, including: step S1: regularly collecting structured data including net value, investment strategy and asset allocation, as well as unstructured data including news and announcements that meet preset requirements; and preprocessing the collected structured data and unstructured data; step S2: constructing a fund monitoring indicator system based on the preprocessed structured data and unstructured data, and using the fund monitoring indicator system to monitor assets; step S3: dynamically optimizing asset allocation based on asset monitoring results and market risk conditions; step S4: establishing a performance evaluation system based on multiple dimensions including return, risk, and asset allocation, performing performance evaluation on the investment portfolio based on the performance evaluation system to obtain performance evaluation results, performing attribution analysis based on the performance evaluation results, and optimizing the investment strategy based on the attribution analysis results.

[0005] Therefore, there is an urgent need for an intelligent post-investment management method that can understand situational changes and dynamically evolve strategies to adapt to high-frequency fluctuations and complex coupled market environments. Summary of the Invention

[0006] In view of the defects in the prior art, the purpose of the present invention is to provide an intelligent post-investment management method and system.

[0007] An intelligent post-investment management method provided by the present invention includes:

[0008] Step S1: Acquire structured financial data and unstructured financial information and construct a multi-source heterogeneous graph;

[0009] Step S2: Use graph neural networks to perform context modeling on financial events based on multi-source heterogeneous graphs and extract causal context vectors;

[0010] Step S3: Build a deep reinforcement learning model with multiple optimization objectives and generate a set of candidate strategies based on the extracted causal scenario vectors;

[0011] Step S4: Using the evolutionary strategy algorithm to perform crossover mutation, fitness evaluation and optimization on the candidate strategy set;

[0012] Step S5: Identify market scenarios in real time and adjust asset allocation based on the most adaptable strategy version;

[0013] Among them, structured financial data includes net value, rate of return and industry allocation; the unstructured financial information includes policy announcements and news texts.

[0014] Preferably, the graph edge weight in the multi-source heterogeneous graph is defined as:

[0015] w ij =α·mutual information(i,j)+β·temporal co-occurrence frequency(i,j)

[0016] Among them, α and β are dynamically adjusted weight factors, reflecting the degree of joint influence between causal nodes.

[0017] Preferably, step S2 includes: identifying the semantic associations and causal paths of financial events based on multi-source heterogeneous graphs through a graph neural network to generate a scenario state vector; the scenario state vector includes: industry public opinion score, macro-policy sentiment intensity, market volatility and time trend factor.

[0018] Preferably, step S3 includes:

[0019] Build a reinforcement learning strategy framework with multiple objective functions, including maximizing the Sharpe ratio, controlling the maximum drawdown, and reducing volatility and transaction costs;

[0020] Use multi-objective reinforcement learning algorithm to train the initial policy set;

[0021] Generate a set of candidate strategies based on the extracted causal scenario vectors.

[0022] Preferably, step S4 includes: using CMA-ES or genetic algorithm to cross-mutate the candidate strategy set and evaluate the fitness, retaining high-performance strategies that meet preset requirements for the next round of deployment, and realizing intergenerational optimization of strategies.

[0023] Preferably, step S5 includes: when identifying the market scenario in real time and matching the strategy version with the highest adaptability to adjust the asset allocation, in order to control the transaction frequency, introducing a minimum adjustment threshold:

[0024] δ min =max(0.5%,min(1.5%,σ 组合收益 ×1.2)).

[0025] Preferably, the method further includes: feeding actual operation data back into strategy training to achieve closed-loop incremental learning; and strategy adaptive evolution achieves version updates through periodic online training.

[0026] An intelligent post-investment management system provided by the present invention includes:

[0027] Module M1: Acquire structured financial data and unstructured financial information and construct a multi-source heterogeneous graph;

[0028] Module M2: Uses graph neural networks to perform context modeling of financial events based on multi-source heterogeneous graphs and extract causal context vectors;

[0029] Module M3: Build a deep reinforcement learning model with multiple optimization objectives and generate a set of candidate policies based on the extracted causal scenario vectors;

[0030] Module M4: Use the evolutionary strategy algorithm to perform crossover mutation, fitness evaluation and optimization on the candidate strategy set;

[0031] Module M5: Identify market scenarios in real time and adjust asset allocation based on the most adaptable strategy.

[0032] Among them, structured financial data includes net value, rate of return and industry allocation; the unstructured financial information includes policy announcements and news texts.

[0033] Preferably, the graph edge weight in the multi-source heterogeneous graph is defined as:

[0034] w ij =α·mutual information(i,j)+β·temporal co-occurrence frequency(i,j)

[0035] Among them, α and β are dynamically adjusted weight factors, reflecting the degree of joint influence between causal nodes;

[0036] The module M2 includes: identifying the semantic associations and causal paths of financial events based on multi-source heterogeneous graphs through graph neural networks, and generating a scenario state vector; the scenario state vector includes: industry public opinion score, macro-policy sentiment intensity, market volatility and time trend factor.

[0037] Preferably, the module M3 includes:

[0038] Build a reinforcement learning strategy framework with multiple objective functions, including maximizing the Sharpe ratio, controlling the maximum drawdown, and reducing volatility and transaction costs;

[0039] Use multi-objective reinforcement learning algorithm to train the initial policy set;

[0040] Generate a set of candidate strategies based on the extracted causal scenario vectors;

[0041] The module M4 includes: using CMA-ES or genetic algorithm to cross-mutate the candidate strategy set and evaluate the fitness, retaining high-performance strategies that meet preset requirements for the next round of deployment, and realizing intergenerational optimization of strategies.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] 1. This invention uses automated data processing and model building to quickly respond to market changes, reduce manual decision-making time, and improve overall decision-making efficiency;

[0044] 2. This invention uses structured and unstructured information to construct a multi-source heterogeneous graph, which helps to more comprehensively identify the inherent connections and potential risks of financial events, making investment decisions more scientific;

[0045] 3. This invention identifies market scenarios in real time and matches the most adaptable strategy version, ensuring that asset allocation can flexibly respond to market changes and improve risk management capabilities;

[0046] 4. Through intergenerational optimization, the present invention enables investment strategies to self-adjust and improve based on market feedback in each round, ensuring the long-term effectiveness of the strategy;

[0047] 5. By combining advanced technologies such as deep learning, graph neural networks, and reinforcement learning, this invention not only improves the efficiency and accuracy of financial decision-making, but also provides investors with more flexible and effective asset management solutions. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Other features, objects and advantages of the present invention will become more apparent upon reading the detailed description of non-limiting embodiments with reference to the following drawings:

[0049] Figure 1 This is a flow chart of the intelligent post-investment management method. DETAILED DESCRIPTION

[0050] The present invention will be described in detail below with reference to specific embodiments. The following examples will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those skilled in the art, several changes and improvements can be made without departing from the scope of the present invention. These all fall within the scope of protection of the present invention.

[0051] Example 1

[0052] According to the present invention, an intelligent post-investment management method and system are provided. Figure 1 As shown, it includes: causal scenario modeling through graph neural networks, combined with multi-objective reinforcement learning and evolutionary strategies, to achieve adaptive generation, dynamic evolution and real-time feedback adjustment of asset allocation strategies, and enhance the resilience management capabilities of post-investment asset portfolios.

[0053] The intelligent post-investment management method includes:

[0054] Step 1: Obtain structured data and unstructured data to construct a multi-source heterogeneous graph. In this embodiment, the structured data includes net value, rate of return, and industry allocation; the unstructured data includes policy announcements and news texts.

[0055] The edge weight of the multi-source heterogeneous graph is defined as follows:

[0056] w ij =α·mutual information(i,j)+β·temporal co-occurrence frequency(i,j)

[0057] Among them, α and β are dynamically adjusted weight factors, reflecting the degree of joint influence between causal nodes.

[0058] Step 2: Use a graph neural network to identify the semantic associations and causal paths of financial events and generate a contextual state vector. In this embodiment, the graph neural network includes any one of GAT, GCN, or R-GCN.

[0059] The contextual state vector includes, but is not limited to: industry public opinion score; macroeconomic policy sentiment intensity; market volatility (VIX); and time trend factors (such as the 5-day EMA slope). The industry public opinion score is a text sentiment score calculated using the BERT-LSTM model.

[0060] Step 3: Construct a reinforcement learning strategy framework with multiple objective functions. The optimization objectives include maximizing the Sharpe ratio, controlling maximum drawdown, and reducing volatility and transaction costs. Use a multi-objective reinforcement learning algorithm to train the initial strategy set. The multi-objective reinforcement learning algorithm can be any of TD3, PPO, and DDPG.

[0061] Step 4: Use an evolutionary algorithm to cross-mutate candidate strategies and evaluate their fitness. High-performing strategies are retained for the next round of deployment, achieving generational optimization. The strategy pool is capped at 20, and a "elite retention + round-by-round mutation rate reduction" mechanism is employed. The initial mutation rate is set at 20%, decreasing by 5% every three rounds. The evolutionary algorithm can be any of CMA-ES, genetic algorithm, differential evolution algorithm (DE), or particle swarm optimization algorithm (PSO).

[0062] Step 5: Identify market scenarios in real time and match the most adaptable strategy version to adjust asset allocation or risk intervention. To control transaction frequency, introduce a minimum adjustment threshold:

[0063] δ min =max(0.5%,min(1.5%,σ 组合收益 ×1.2))

[0064] In this embodiment, the variance of strategy returns is calculated through cross-scenario Monte Carlo backtesting to evaluate the generalization ability of the strategy in different scenarios.

[0065] Step 6: Feed actual operational data (returns, volatility, and rebalances) back to the strategy training module to achieve closed-loop incremental learning. Strategy adaptive evolution is achieved through periodic online training module version updates, continuously improving robustness and responsiveness.

[0066] The present invention is deployed in a microservice architecture, and each module is independently serviced, which facilitates strategy iteration and integration.

[0067] The present invention can be used for intelligent post-investment management of various financial products such as funds, wealth management, and private equity portfolios.

[0068] The present invention also provides an intelligent post-investment management system, which can be implemented by executing the process steps of the intelligent post-investment management method. That is, those skilled in the art can understand the intelligent post-investment management method as a preferred implementation of the intelligent post-investment management system.

[0069] Those skilled in the art will appreciate that, in addition to implementing the system, device, and various modules provided by the present invention in purely computer-readable program code, it is entirely possible to implement the same program in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, embedded microcontrollers, and the like by logically programming the method steps. Therefore, the system, device, and various modules provided by the present invention can be considered a hardware component, and the modules included therein for implementing various programs can also be considered structures within the hardware component; the modules for implementing various functions can also be considered both software programs for implementing the method and structures within the hardware component.

[0070] The above describes specific embodiments of the present invention. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art may make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. The embodiments of this application and the features in the embodiments may be combined with each other in any manner unless there is a conflict.

Claims

1. An intelligent post-investment management method, characterized in that: include: Step S1: Acquire structured financial data and unstructured financial information and construct a multi-source heterogeneous graph; Step S2: Use graph neural networks to perform context modeling on financial events based on multi-source heterogeneous graphs and extract causal context vectors; Step S3: Build a deep reinforcement learning model with multiple optimization objectives and generate a set of candidate strategies based on the extracted causal scenario vectors; Step S4: Using the evolutionary strategy algorithm to perform crossover mutation, fitness evaluation and optimization on the candidate strategy set; Step S5: Identify market scenarios in real time and adjust asset allocation based on the most adaptable strategy version; Among them, structured financial data includes net value, rate of return and industry allocation; the unstructured financial information includes policy announcements and news texts.

2. The intelligent post-investment management method according to claim 1, characterized in that: The graph edge weight in the multi-source heterogeneous graph is defined as: w ij =α·mutual information(i,j)+β·temporal co-occurrence frequency(i,j) Among them, α and β are dynamically adjusted weight factors, reflecting the degree of joint influence between causal nodes.

3. The intelligent post-investment management method according to claim 1, characterized in that: The step S2 includes: identifying the semantic associations and causal paths of financial events based on multi-source heterogeneous graphs through a graph neural network, and generating a scenario state vector; the scenario state vector includes: industry public opinion score, macro-policy sentiment intensity, market volatility, and time trend factor.

4. The intelligent post-investment management method according to claim 1, characterized in that: The step S3 comprises: Build a reinforcement learning strategy framework with multiple objective functions, including maximizing the Sharpe ratio, controlling the maximum drawdown, and reducing volatility and transaction costs; Use multi-objective reinforcement learning algorithm to train the initial policy set; Generate a set of candidate strategies based on the extracted causal scenario vectors.

5. The intelligent post-investment management method according to claim 1, characterized in that: The step S4 includes: using CMA-ES or genetic algorithm to cross-mutate the candidate strategy set and evaluate the fitness, retaining high-performance strategies that meet preset requirements for the next round of deployment, and realizing intergenerational optimization of strategies.

6. The intelligent post-investment management method according to claim 1, characterized in that: The step S5 includes: when identifying the market scenario in real time and matching the most adaptable strategy version to adjust the asset allocation, in order to control the transaction frequency, introducing a minimum adjustment threshold: d min =max(0.5%,min(1.5%,σ 组合收益 ×1.2)).

7. The intelligent post-investment management method according to claim 1, characterized in that: The method also includes: feeding actual operation data back to strategy training to achieve closed-loop incremental learning; and strategy adaptive evolution achieves version updates through periodic online training.

8. An intelligent post-investment management system, characterized in that: include: Module M1: Acquire structured financial data and unstructured financial information and construct a multi-source heterogeneous graph; Module M2: Uses graph neural networks to perform context modeling of financial events based on multi-source heterogeneous graphs and extract causal context vectors; Module M3: Build a deep reinforcement learning model with multiple optimization objectives and generate a set of candidate policies based on the extracted causal scenario vectors; Module M4: Use the evolutionary strategy algorithm to perform crossover mutation, fitness evaluation and optimization on the candidate strategy set; Module M5: Identify market scenarios in real time and adjust asset allocation based on the most adaptable strategy. Among them, structured financial data includes net value, rate of return and industry allocation; the unstructured financial information includes policy announcements and news texts.

9. The intelligent post-investment management system according to claim 8, characterized in that: The graph edge weight in the multi-source heterogeneous graph is defined as: w ij =α·mutual information(i,j)+β·temporal co-occurrence frequency(i,j) Among them, α and β are dynamically adjusted weight factors, reflecting the degree of joint influence between causal nodes; The module M2 includes: identifying the semantic associations and causal paths of financial events based on multi-source heterogeneous graphs through graph neural networks, and generating a scenario state vector; the scenario state vector includes: industry public opinion score, macro-policy sentiment intensity, market volatility and time trend factor.

10. The intelligent post-investment management system according to claim 8, characterized in that: The module M3 includes: Build a reinforcement learning strategy framework with multiple objective functions, including maximizing the Sharpe ratio, controlling the maximum drawdown, and reducing volatility and transaction costs; Use multi-objective reinforcement learning algorithm to train the initial policy set; Generate a set of candidate strategies based on the extracted causal scenario vectors; The module M4 includes: using CMA-ES or genetic algorithm to cross-mutate the candidate strategy set and evaluate the fitness, retaining high-performance strategies that meet preset requirements for the next round of deployment, and realizing intergenerational optimization of strategies.

Images