A method, device, apparatus and storage medium for dynamic optimization of resource configuration

Abstract

The application belongs to the technical field of artificial intelligence, is suitable for medical field and financial field, and discloses a dynamic optimization method of resource configuration, which comprises the following steps: acquiring market data and target information, identifying market trend and potential risk through a pre-trained machine learning model, and generating market state labels for representing current market state; adopting a multi-objective optimization algorithm to dynamically adjust a resource configuration scheme, and generating multi-objective optimization configuration results; screening factors having a significant influence on yield by using the machine learning model, generating an optimal factor combination based on optimized factor weights; and adjusting the optimization configuration results according to the optimal factor combination, and outputting a final resource configuration scheme. The application solves the technical problems that static optimization cannot adapt to changes, single-target optimization ignores multi-target balance, and factor optimization lacks dynamic adjustment capability in the prior art.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence technology and is applicable to the medical and financial fields. In particular, it relates to a method, apparatus, device and storage medium for dynamic optimization of resource allocation. Background Technology

[0002] Resource allocation optimization techniques are widely used in various fields to improve overall efficiency by rationally allocating limited resources while meeting constraints. In the financial sector, for example, resource allocation optimization is often used to determine asset allocation ratios. However, traditional resource allocation optimization methods (such as the mean-variance model) typically rely on static optimization based on historical data, using the mean and covariance matrix of historical statistics as input to calculate the optimal allocation ratio. This approach assumes that data characteristics will remain unchanged in the future and cannot effectively cope with rapid changes in the data environment. When the data structure undergoes structural changes or experiences extreme fluctuations, the static optimization results often lag behind the actual data distribution, leading to poor overall performance of the allocation scheme.

[0003] Existing technologies often focus on single or dual-objective optimization. For example, mean-variance models aim to maximize expected returns and minimize volatility. However, in practical decision-making, multiple factors such as liquidity constraints, adjustment costs, and industry concentration must be considered. Ignoring these important factors can lead to inadequate allocation schemes. For instance, they may struggle to adjust quickly in environments with insufficient liquidity, or the high adjustment costs may erode overall efficiency. Existing methods are relatively simple in optimizing feature factor weights. They construct allocation schemes by selecting feature factors that have explanatory power for the target variable. However, existing technologies often use static factor weights or adjust based on simple rules, lacking the dynamic adjustment capabilities based on artificial intelligence. This makes it difficult to achieve effective factor selection and adaptive weight optimization in complex and ever-changing data environments. Existing schemes struggle to respond quickly to data fluctuations and lack mechanisms for real-time monitoring of data status and dynamic adjustment of resource allocation, resulting in poor performance and stability of the allocation schemes. Neither static optimization methods nor dynamic rebalancing systems based on fixed rules fully utilize artificial intelligence technologies such as machine learning and reinforcement learning to identify data status, balance multiple objectives, and adaptively optimize factor weights.

[0004] Existing resource allocation optimization techniques suffer from problems such as static optimization, single-objective optimization, and insufficient factor optimization, which limit their effectiveness in practical applications. There is an urgent need for a solution that can dynamically adjust, perform multi-objective optimization, and incorporate intelligent factor optimization to improve the efficiency, robustness, and environmental adaptability of resource allocation schemes. Summary of the Invention

[0005] This invention provides a method, apparatus, device, and storage medium for dynamic optimization of resource allocation, which solves the technical problems in existing resource allocation optimization technologies, such as static optimization being unable to adapt to changes, single-objective optimization neglecting multi-objective balance, and factor optimization lacking dynamic adjustment capabilities.

[0006] In a first aspect, the present invention provides a dynamic optimization method for resource allocation, comprising: Acquire market data and target information, and identify market trends and potential risks using a pre-trained machine learning model based on the market data and target information, generating market state labels to characterize the current market state; Based on the market status label and the target information, a multi-objective optimization algorithm is used to dynamically adjust the combined resource allocation scheme and generate multi-objective optimized allocation results. The machine learning model is used to screen factors that have a significant impact on returns. The weights of these factors are dynamically adjusted using a reinforcement learning algorithm. Based on the optimized factor weights, an optimal factor combination is generated. The optimized configuration result is adjusted based on the optimal factor combination to output the final resource combination configuration scheme.

[0007] In a second aspect, the present invention provides a resource allocation dynamic optimization device, comprising: The data acquisition module is used to acquire market data and target information, and to identify and classify market trends and potential risks based on the market data and target information through a pre-trained machine learning model, generating market status labels to characterize the current market state. The multi-objective optimization module is used to dynamically adjust the combined resource allocation scheme according to the market status label and the target information using a multi-objective optimization algorithm, and generate multi-objective optimized configuration results; The factor optimization module is used to use the machine learning model to screen factors that have a significant impact on returns, dynamically adjust the weights of the factors through a reinforcement learning algorithm, and generate the optimal factor combination based on the optimized factor weights. The result output module is used to adjust the optimized configuration result according to the optimal factor combination and output the final resource combination configuration scheme.

[0008] Thirdly, the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the dynamic optimization method for resource allocation.

[0009] Fourthly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the dynamic optimization method for resource allocation.

[0010] The aforementioned dynamic optimization method, apparatus, equipment, and storage medium for resource allocation, through a pre-trained machine learning model, identifies market trends and potential risks and generates market state labels. This enables real-time perception of market environment changes, overcoming the shortcomings of traditional static optimization methods that cannot adapt to dynamic market changes, making resource portfolio adjustments more forward-looking and timely. A multi-objective optimization algorithm dynamically adjusts the resource allocation scheme, achieving an optimal balance among multiple objectives such as returns, risks, and liquidity, avoiding unreasonable allocation problems caused by single-objective optimization. A machine learning model is used to screen factors that significantly impact returns, and reinforcement learning algorithms dynamically adjust factor weights, achieving adaptive optimization of factor weights and overcoming the shortcomings of existing technologies where factor weight adjustment is simple and lacks intelligent mechanisms. The optimized allocation result is adjusted based on the optimal factor combination, organically combining factor optimization with multi-objective optimization to form a synergistic enhancement optimization mechanism. This significantly improves the final portfolio allocation scheme in terms of profitability, risk control, liquidity management, and market adaptability, realizing intelligent, dynamic, and multi-objective synergistic resource portfolio optimization. This significantly improves the overall performance of the resource portfolio and reduces manual rebalancing costs and operational risks. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a schematic diagram of an application environment for a dynamic optimization method for resource allocation in one embodiment of the present invention.

[0013] Figure 2 This is a flowchart illustrating a dynamic optimization method for resource allocation in one embodiment of the present invention.

[0014] Figure 3 yes Figure 2 A schematic diagram of a specific implementation method for step S20.

[0015] Figure 4 yes Figure 2 A schematic diagram of a specific implementation method for step S30.

[0016] Figure 5 yes Figure 2 A schematic diagram of a specific implementation method for step S50.

[0017] Figure 6 This is a schematic diagram of a resource allocation dynamic optimization device in one embodiment of the present invention.

[0018] Figure 7 This is a schematic diagram of the structure of a computer device according to an embodiment of the present invention.

[0019] Figure 8 This is another structural schematic diagram of a computer device according to one embodiment of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. 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.

[0021] The dynamic optimization method for resource allocation provided in this invention can be applied to, for example... Figure 1 In the application environment, Figure 1This is a schematic diagram of an application environment for a data processing method for insurance business according to an embodiment of the present invention; wherein, the client communicates with the server via a network. The server can acquire market data and target information, and based on the market data and target information, identify market trends and potential risks using a pre-trained machine learning model to generate a market state label representing the current market state; based on the market state label and the target information, a multi-objective optimization algorithm is used to dynamically adjust the combined resource allocation scheme to generate a multi-objective optimized allocation result; the machine learning model is used to screen factors that have a significant impact on returns, and the weights of the factors are dynamically adjusted using a reinforcement learning algorithm, and an optimal factor combination is generated based on the optimized factor weights; the optimized allocation result is adjusted based on the optimal factor combination, and the final resource combination allocation scheme is output. In this invention, a pre-trained machine learning model identifies market trends and potential risks and generates market state labels, enabling real-time perception of market environment changes. This overcomes the shortcomings of traditional static optimization methods, which cannot adapt to dynamic market changes, making resource portfolio adjustments more forward-looking and timely. A multi-objective optimization algorithm dynamically adjusts the asset allocation scheme, achieving an optimal balance among multiple objectives such as returns, risk, and liquidity, avoiding the problem of unreasonable allocation caused by single-objective optimization. A machine learning model is used to screen factors that significantly affect returns, and reinforcement learning algorithms are used to dynamically adjust factor weights, achieving adaptive optimization of factor weights. This overcomes the shortcomings of existing technologies, such as simple factor weight adjustment and a lack of intelligent mechanisms, allowing the resource portfolio to better capture market opportunities. The optimized allocation result is adjusted based on the optimal factor combination, organically combining factor optimization with multi-objective optimization to form a synergistic enhancement optimization mechanism. This significantly improves the final resource portfolio allocation scheme in terms of profitability, risk control, liquidity management, and market adaptability, realizing intelligent, dynamic, and multi-objective synergy in resource portfolio optimization. This significantly improves the overall performance of the resource portfolio and reduces manual rebalancing costs and operational risks. The invention will be described in detail below through specific embodiments.

[0022] Please see Figure 2 As shown, Figure 2 This is a flowchart illustrating a dynamic resource allocation optimization method according to an embodiment of the present invention. The dynamic resource allocation optimization method specifically includes the following steps: S10 acquires market data and target information, and identifies market trends and potential risks based on the market data and target information using a pre-trained machine learning model, generating a market state label to characterize the current market state. The acquisition of market data and target information includes: Real-time data collection of asset prices, macroeconomic indicators, and trading volume data is achieved through data interfaces or data platforms, and information on return targets, risk preferences, and restrictions is obtained.

[0023] Specifically, market data is collected in real time via API interfaces or data platforms, including data reflecting changes in the market environment such as asset price fluctuations and macroeconomic indicators. Target information includes return objectives and risk preferences. The collected market data and target information are input into a pre-trained machine learning model. This model identifies and classifies market trends, such as upward, downward, and volatile movements, as well as potential risks, such as high volatility and tightening liquidity, and outputs a market state label characterizing the current market state. This market state label serves as the basis for subsequent dynamic adjustments to the resource portfolio, enabling the adjustment process to respond to changes in the market environment. By identifying market trends and potential risks and generating market state labels through a pre-trained machine learning model, real-time perception of market environment changes can be achieved, overcoming the shortcomings of traditional static optimization methods that cannot adapt to dynamic market changes, making resource portfolio adjustments more forward-looking and timely.

[0024] In financial investment scenarios, real-time market data and investor target information can be obtained. Market data includes asset price fluctuations, macroeconomic indicators, and trading volume data, while investor target information includes return objectives, risk preferences, investment horizons, and liquidity needs. This market data is collected in real-time via API interfaces or data platforms and combined with investor-entered or system-stored target information, then input into a pre-trained machine learning model. This model identifies and classifies market trends, such as upward trends, downward trends, and volatile markets, as well as potential risks, such as high volatility, liquidity tightening risks, and macroeconomic downturn risks, generating a market state label to characterize the current market state. This market state label reflects real-time changes in the market environment.

[0025] In healthcare management scenarios, patient health monitoring data and health management goals can be acquired. Health monitoring data includes physiological indicators, exercise data, sleep quality, and medical records, while health management goals include weight control targets, blood glucose control ranges, and exercise targets. This health monitoring data is collected in real-time via wearable devices, medical information systems, or user input, and combined with user-defined health management goals, then fed into a pre-trained machine learning model. This model identifies and classifies health trends, such as weight changes, blood glucose fluctuations, changes in exercise habits, and potential health risks, such as the risk of hypertension, diabetes complications, and cardiovascular events, generating a health status label that represents the user's current health condition. This health status label reflects real-time changes in the user's health status.

[0026] S20 uses a multi-objective optimization algorithm to dynamically adjust the resource combination configuration scheme based on the market status label and the target information, and generates a multi-objective optimized configuration result; Specifically, the market state labels and target information generated by S10 are used as input parameters for the multi-objective optimization algorithm. The objective function of the multi-objective optimization algorithm includes at least two of the objectives of maximizing returns, minimizing risk, and optimizing liquidity. Constraints are set during the optimization process, including factors such as transaction costs, industry restrictions, and geographical limitations. The algorithm uses methods such as genetic algorithms or particle swarm optimization to solve the objective function, generating a Pareto optimal solution set. This solution set contains multiple asset allocation schemes that achieve equilibrium among multiple objectives. Based on the target information, the preferred allocation result is selected from the Pareto optimal solution set as the output of the multi-objective optimized allocation result. By dynamically adjusting the resource allocation scheme through the multi-objective optimization algorithm, optimal balance is achieved among multiple objectives such as returns, risk, and liquidity, avoiding the unreasonable allocation problems caused by single-objective optimization. Simultaneously, by setting constraints such as transaction costs and resource limitations, the optimization results are made closer to real-world scenarios, improving the executability of the optimization scheme.

[0027] Specifically, such as Figure 3 The above, Figure 3 This is a flowchart illustrating a specific implementation of step S20, which specifically includes the following steps S21-S23: S21 incorporates at least two of the objectives of maximizing returns, minimizing risks, and optimizing liquidity into the objective function; Specifically, in multi-objective optimization algorithms, the objective function is a quantitative standard for evaluating the merits of resource portfolio allocation schemes. The profit maximization objective is typically quantified by the expected rate of return of the resource portfolio, which can be calculated using a weighted average of historical returns or an expected return prediction model. The risk minimization objective is represented by risk metrics such as the volatility, value at risk, or maximum drawdown of the resource portfolio. The liquidity optimization objective is measured by liquidity metrics such as the average daily trading volume, bid-ask spread, or turnover rate of the assets in the resource portfolio. Based on specific needs and market conditions, at least two of these three objectives are selected as optimization objectives to construct a multi-objective objective function. The objective function can be implemented using weighted summation, Pareto ranking, or constraint methods, incorporating multiple objectives into a unified optimization framework.

[0028] In financial investment scenarios, at least two of the objectives—maximizing returns, minimizing risk, and optimizing liquidity—can be incorporated into the objective function. Maximizing returns can be quantified by the expected annualized rate of return of resource allocation, calculated based on historical return data, asset pricing models, or machine learning predictions. Minimizing risk can be measured by risk indicators such as annualized volatility, value at risk, or maximum drawdown of resource allocation. Optimizing liquidity can be characterized by indicators such as average daily trading volume, bid-ask spread, turnover rate, or liquidity coverage ratio of the assets in resource allocation. Depending on the investor's risk appetite, investment horizon, and liquidity needs, at least two of these three objectives can be selected as optimization goals. For example, conservative investors can choose to minimize risk and optimize liquidity, prioritizing capital safety and liquidity; aggressive investors can choose to maximize returns and minimize risk, pursuing higher returns while controlling risk.

[0029] In healthcare management, at least two of the following objectives—maximizing health improvement, minimizing health risk, and optimizing intervention cost—can be incorporated into the objective function: maximizing health improvement can be quantified using indicators such as weight loss, blood glucose control achievement rate, blood pressure reduction, or improved exercise capacity, calculated based on wearable device monitoring data, physical examination reports, or user self-assessment results. Minimizing health risk can be measured using risk indicators such as cardiovascular disease risk score, diabetes complication risk index, fall risk level, or disease recurrence probability. Optimizing intervention cost can be characterized by the direct costs of the health intervention plan, such as medication costs, nutritional meal costs, and rehabilitation equipment costs, as well as indirect costs such as time costs and transportation costs to medical facilities. Based on the patient's health status, chronic disease type, personal health goals, and financial capacity, at least two of these three objectives can be selected as optimization objectives. For example, for patients with prediabetes, maximizing health improvement and minimizing health risk can be chosen as optimization objectives, prioritizing blood glucose control and reducing the risk of developing diabetes; for patients with stable chronic diseases, minimizing health risk and optimizing intervention cost can be chosen as optimization objectives, reducing medical expenses while maintaining disease stability.

[0030] S22 sets constraints, which include at least one of transaction costs, industry restrictions, and geographical restrictions; Specifically, in multi-objective optimization, constraints are used to limit the space of the optimization solution, ensuring that the generated asset allocation plan meets the feasibility requirements of actual operation. Transaction cost constraints refer to considering the commissions, stamp duties, impact costs, and other expenses incurred in buying and selling assets when adjusting the resource portfolio, avoiding excessive costs that erode returns due to frequent portfolio adjustments or excessively large transaction volumes. Industry restrictions refer to setting upper or lower limits on the allocation ratio of specific industries to prevent the resource portfolio from being overly concentrated in a certain industry and facing systemic risks from that industry. Geographical restrictions refer to limiting the asset allocation ratio of specific countries or regions to control geographical concentration risk or meet cross-border investment compliance requirements. The above constraints can be set according to risk appetite, regulatory requirements, or strategies, and serve as necessary conditions for a feasible solution in the multi-objective optimization process.

[0031] In financial investment scenarios, investment constraints can be set, including at least one of transaction costs, industry investment restrictions, and geographical investment restrictions. Transaction cost constraints can be based on the client's account type (e.g., ordinary account, institutional account, high-frequency trading account), trading volume, and market liquidity to estimate commissions, stamp duty, impact costs, etc., incurred from buying and selling assets. These costs are then incorporated into the optimization model as constraints to prevent excessive costs from eroding investment returns due to frequent portfolio adjustments or excessively large trading volumes. Industry investment restrictions can be based on the client's industry preferences (e.g., avoiding highly polluting industries, focusing on technology industries, or industry risk exposure control needs) to set upper or lower limits on the allocation ratio of specific industries such as finance, technology, energy, and healthcare, preventing excessive concentration of resources in a single industry and exposure to systemic industry risks. Geographical investment restrictions can be based on the client's cross-border investment compliance requirements, geographical risk preferences (e.g., avoiding regions with high geopolitical risks), or tax considerations to limit the asset allocation ratio in specific countries or regions, thereby controlling geographical concentration risk and meeting regulatory requirements. By setting the above investment constraints, the asset allocation plan generated by multi-objective optimization can better meet the actual investment needs and risk control requirements of clients, which helps to reduce transaction costs, diversify industry and regional risks, and improve the feasibility and stability of resource allocation.

[0032] In healthcare management scenarios, constraints can be set for health interventions. These constraints may include at least one of the following: intervention cost, treatment plan restrictions, and regional medical resource restrictions. Intervention cost constraints can estimate direct costs such as drug costs, examination fees, and rehabilitation equipment costs based on the patient's medical insurance type, economic situation, and affordability. These costs can then be incorporated into the health management optimization model as constraints to prevent patients from being unable to adhere to intervention plans long-term or from incurring financial burdens due to excessively high costs. Treatment plan restrictions can limit the application of different treatment methods, such as drug therapy, lifestyle interventions, and surgical treatments, based on the patient's disease type, complications, allergy history, or treatment contraindications. This prevents the use of inappropriate treatment plans that could pose additional risks to the patient's health. Regional medical resource restrictions can constrain treatment plans requiring specific medical resource support based on the distribution of medical resources in the patient's region, ensuring the feasibility of the intervention plan within the patient's actual residential area.

[0033] S23 uses a genetic algorithm or particle swarm optimization algorithm to solve the objective function under the constraints, generates a Pareto optimal solution set, and selects a multi-objective optimization configuration result from the Pareto optimal solution set according to the objective.

[0034] Specifically, the objective function constructed in S21 and the constraints set in S22 are input into the multi-objective optimization solver. The genetic algorithm, by simulating natural selection and genetic mutation, encodes, selects, crosses over, and mutates candidate solutions, iteratively generating solutions with high fitness. The particle swarm optimization algorithm, by simulating the foraging behavior of a flock of birds, uses each particle to represent a candidate solution, continuously updating its own position by tracking the optimal position of the individual and the optimal position of the group, gradually approaching the optimal solution set. Both algorithms are applicable to multi-objective optimization problems and can generate a set of non-dominated solutions, i.e., the Pareto optimal solution set, in a single run. Each solution in this set satisfies the condition that no other objective can be further optimized without harming other objectives. Based on objective information, such as risk preference, expected return, and liquidity requirements, the solution that best matches the preference is selected from the Pareto optimal solution set and output as the multi-objective optimization configuration result.

[0035] In financial investment scenarios, genetic algorithms or particle swarm optimization algorithms can be used to solve the objective function under constraints, generating a Pareto optimal solution set. The resulting multi-objective optimal allocation is then selected from this set based on the investor's goals. Specifically, at least two of the objectives—maximizing returns, minimizing risk, and optimizing liquidity—are incorporated into the objective function. Constraints such as transaction costs, industry investment restrictions, and geographical investment restrictions are set. The algorithm then iteratively optimizes the asset allocation scheme by simulating natural selection and genetic mutation using a genetic algorithm, or by simulating bird flocking foraging behavior using a particle swarm optimization algorithm. In each iteration, the algorithm generates a set of candidate asset allocation schemes. By evaluating the performance of each scheme across multiple objectives, non-dominated solutions are selected—schemes that cannot be further optimized for any objective without compromising the performance of other objectives. After multiple iterations, the algorithm outputs a Pareto optimal solution set consisting of multiple non-dominated solutions. Each solution in this set corresponds to an asset allocation ratio, reflecting different combinations of returns, risks, and liquidity characteristics. Based on the investor's return objectives, risk tolerance, and liquidity needs, the solution that best matches the investor's preferences is selected from the Pareto optimal solution set as the multi-objective optimal allocation result.

[0036] In healthcare management scenarios, genetic algorithms or particle swarm optimization algorithms can be used to solve the objective function under constraints, generating a Pareto optimal solution set. The result of multi-objective optimization is then selected from this set based on the patient's health goals. Specifically, at least two of the objectives—maximizing health improvement, minimizing health risk, and optimizing intervention cost—are incorporated into the objective function. Constraints such as intervention cost, treatment plan limitations, and geographical medical resource limitations are set. The algorithm then iteratively optimizes the health intervention plan by simulating natural selection and genetic variation using a genetic algorithm, or by simulating bird flock foraging behavior using a particle swarm optimization algorithm. In each iteration, the algorithm generates a set of candidate intervention plans, including drug combinations, exercise plans, dietary adjustments, and regular monitoring. By evaluating the performance of each plan across multiple health goals, non-dominated solutions are selected—intervention plans that cannot be further optimized for any single goal without compromising the performance of other health goals. After multiple iterations, the algorithm outputs a Pareto optimal solution set composed of multiple non-dominated solutions. Each solution in this set corresponds to a combination of health intervention plans, reflecting different health improvement effects, risk control levels, and cost characteristics. Based on the patient's health status, type of chronic disease, personal health goals, and economic affordability, the solution that best matches the patient's preferences is selected from the Pareto optimal solution set as the multi-objective optimization configuration result.

[0037] S30 uses the machine learning model to screen factors that have a significant impact on returns, dynamically adjusts the weights of the factors through a reinforcement learning algorithm, and generates an optimal factor combination based on the optimized factor weights. Specifically, a pre-trained machine learning model is used to analyze the candidate factor set. Candidate factors include value factors, momentum factors, size factors, quality factors, and volatility factors. The machine learning model analyzes the correlation between each factor and returns, calculates the importance score of each factor, and selects factors with a correlation exceeding a preset threshold as target factors, while eliminating factors with weak impact on returns or exhibiting multicollinearity. A reinforcement learning algorithm is then used to dynamically optimize the weights of the selected target factors. In the reinforcement learning framework, the current market state is used as the state space, the factor weight adjustment actions (such as increasing weight, decreasing weight, or keeping it unchanged) are used as the action space, and return performance, such as Sharpe ratio and excess returns, are used as reward signals. The agent learns the optimal factor weight adjustment strategy under different market states by interacting with the environment, enabling the factor weights to adaptively adjust according to market changes. Based on the dynamically adjusted factor weights from the reinforcement learning algorithm, the target factors are combined according to their weights to generate the optimal factor combination. This factor combination reflects the relative importance of each factor's contribution to returns under the current market state and is used for subsequent adjustments to the combination configuration. By using machine learning models to screen factors that have a significant impact on returns, effective factors can be automatically identified from a large number of candidate factors, reducing the subjectivity and limitations of manual screening.

[0038] In healthcare management, machine learning models can be used to screen intervention factors that significantly impact health improvement. Reinforcement learning algorithms dynamically adjust the weights of these intervention factors, generating an optimal combination of intervention factors based on the optimized weights. Specifically, the candidate intervention factor set includes exercise intervention factors (e.g., daily step goals, aerobic exercise frequency, strength training intensity), dietary intervention factors (e.g., calorie intake control, nutrient ratios, dietary fiber intake), medication intervention factors (e.g., drug type, dosage, medication adherence), sleep intervention factors (e.g., sleep duration, sleep quality, sleep onset time), and psychological intervention factors (e.g., stress management, meditation practice, social activities). Machine learning models are used to analyze the correlation between each intervention factor and health improvement effects, such as blood glucose control levels, blood pressure changes, body mass index changes, and quality of life scores, selecting factors with correlations exceeding a preset threshold as target intervention factors. Subsequently, the current health status, such as blood glucose level, blood pressure value, and weight change trend, is used as the state space; the adjustment actions of the intervention factor weights, such as increasing weight, decreasing weight, or keeping it unchanged, are used as the action space; and the health improvement effects, such as blood glucose target achievement rate, blood pressure reduction magnitude, and weight loss percentage, are used as reward signals. Through iterative training using a reinforcement learning algorithm, the optimal intervention factor weight adjustment strategy is learned under different health states. After training, the weights of each target intervention factor are dynamically adjusted based on the learned strategy, and the intervention factors are weighted and combined according to the adjusted weights to generate the optimal intervention factor combination.

[0039] Specifically, such as Figure 4 The above, Figure 4 This is a flowchart illustrating a specific implementation of step S30, which specifically includes the following steps S31-S33: S31 uses the machine learning model to analyze the preset set of candidate factors, sorts them according to the degree of correlation between each factor and the return, and selects the factors with a correlation degree exceeding a preset threshold as target factors. Specifically, the pre-defined candidate factor set includes common factor types such as value factors, momentum factors, size factors, quality factors, and volatility factors. These candidate factors are used as input features, and returns, such as future returns or excess returns, are used as prediction targets. Machine learning models, such as random forests, gradient boosting trees, and LSTMs, are used to quantitatively analyze the correlation between each factor and returns. The machine learning model calculates feature importance indices, such as Gini importance, permutation importance, and model weight coefficients, to score the importance of each candidate factor, reflecting its explanatory or predictive power for returns. Factors are ranked according to their importance scores, and those with scores exceeding a pre-defined threshold are selected as target factors. Factors with low scores or redundant factors highly correlated with other factors are removed, retaining only the set of relatively independent factors that have a significant impact on returns. Quantitative analysis and screening of candidate factors through machine learning models can automatically identify factors with a significant impact on returns from a large number of candidates, reducing the subjectivity and reliance on experience in manual screening.

[0040] S32 uses the current market state as the state space, the factor weight adjustment action as the action space, and the return performance as the reward signal. Through iterative training using a reinforcement learning algorithm, it learns the optimal factor weight adjustment strategy under different market states and dynamically adjusts the weights of each target factor based on the learned strategy.

[0041] Specifically, in the reinforcement learning framework, the state space is defined as the market state labels generated by S10, including market trend categories and potential risk levels. The action space is defined as the adjustment operations on the target factor weights, including increasing weights, decreasing weights, keeping them unchanged, and the specific magnitude of the adjustment. The reward signal is defined as a quantitative indicator of return performance, which can be an indicator such as the Sharpe ratio, excess return, information ratio, or cumulative return, reflecting the portfolio's return performance under the current factor weight configuration.

[0042] Reinforcement learning algorithms employ iterative training through agent-environment interaction. At each time step, the agent observes the current market state, selects an action to adjust factor weights based on its current strategy, receives a reward for executing the action, and then proceeds to the next market state. The agent updates its strategy based on the reward signal, aiming to maximize long-term cumulative rewards. Through multiple rounds of iterative training, the agent gradually learns how to adjust factor weights to obtain optimal returns under different market states. After training, based on the learned strategy, the weights of each target factor are dynamically adjusted according to the current market state in practical applications. By modeling factor weight adjustment as a sequential decision problem through reinforcement learning, the agent can autonomously learn the optimal adjustment strategy through continuous interaction with the market.

[0043] S33 weights and combines each target factor according to the factor weights dynamically adjusted by the reinforcement learning algorithm, and calculates the relative proportion of each target factor in the factor combination. Specifically, the target factor set obtained after screening in S31 contains multiple factors. Each factor, in S32, receives a dynamically adjusted weight value based on the current market state using a reinforcement learning algorithm. This weight value reflects the relative importance of the factor's contribution to returns under the current market environment. Weighted combination refers to linearly combining the target factors according to their weight values ​​to form a unified factor combination expression. When calculating the relative proportion, the weight value of each target factor is divided by the sum of the weight values ​​of all target factors to obtain the proportion of each target factor in the factor combination. This proportion is between 0 and 1, and the sum of the proportions of all target factors is 1, characterizing the distribution of the relative importance of each factor in the factor combination.

[0044] S34 adjusts the weights of the weighted combination factors based on the correlation analysis results between the target factors to eliminate the multicollinearity effect between the factors. Specifically, within the target factor set selected by S31, some factors may exhibit high correlation, i.e., multicollinearity. For example, value factors and quality factors may show a strong positive correlation under certain market conditions, while momentum factors and volatility factors may have a negative correlation. If the weighted portfolio weights generated by S33 are used directly, highly correlated factors may make duplicate contributions to the factor portfolio, causing the actual exposure of the factor portfolio to deviate from expectations and affecting the accuracy of subsequent resource portfolio adjustments.

[0045] Correlation analysis calculates correlation coefficients between target factors using historical data, such as Pearson correlation coefficient and Spearman rank correlation coefficient, to identify the degree of linear or nonlinear association between factors. A significant correlation is determined when the absolute value of the correlation coefficient between two or more factors exceeds a preset threshold. Decorrelation adjustment can be achieved using various methods, including: factor orthogonalization, which constructs orthogonal factors using methods such as Schmidt orthogonalization to eliminate linear correlations between factors; weight redistribution, which treats highly correlated factors as a group and redistributes weights based on the importance of each factor within the group, maintaining the original combined weights; and principal component analysis, which projects the original factors onto a principal component space and uses principal components to replace the original factors for combination.

[0046] By adjusting for correlation, the factors in the factor portfolio remain relatively independent, avoiding weight redundancy caused by repeated contributions from factors, and ensuring that the factor portfolio can truly reflect the independent impact of each factor on the return.

[0047] S35 normalizes the adjusted factor weights to generate the optimal factor combination where the sum of the weights of all target factors is 1.

[0048] Specifically, after the S34 decorrelation adjustment, the weights of each target factor may no longer satisfy the constraint that the sum of the weights must equal 1. For example, the adjusted weights might be 0.6, 0.4, and 0.3, with a sum of 1.3. Normalization involves dividing the weights of each factor by their sum, so that the sum of the transformed weights equals 1. The normalization formula is: Normalized weight_i = Adjusted weight_i / (Sum of all adjusted weights). Through normalization, the weights of each target factor are rescaled to between 0 and 1, and the sum of all factor weights is 1, forming a standardized optimal factor combination.

[0049] In the normalized optimal factor combination, the weight of each target factor represents its relative importance within the overall factor combination. This combination can be directly used for subsequent resource allocation adjustments, such as mapping the weights of each factor to the asset level to determine the asset's overall score across the factor dimensions, thereby adjusting the asset allocation ratio.

[0050] In one embodiment of the present invention, before identifying market trends and potential risks using a pre-trained machine learning model based on the market data and the target information, the method further includes: the pre-trained machine learning model is trained through the following steps: Acquire historical market data, which includes at least one of asset price series, trading volume data, macroeconomic indicators, and market sentiment indicators; Specifically, asset price series refers to the price records of various financial assets over a historical time period, including opening price, closing price, highest price, and lowest price, used to calculate basic indicators such as asset returns and volatility. Trading volume data refers to the number or amount of assets traded within a specific time period, reflecting market activity and liquidity, and can be used to help judge the reliability of price trends. Macroeconomic indicators include GDP growth rate, inflation rate, interest rate level, unemployment rate, and purchasing managers' index, reflecting changes in the overall economic environment and having a systematic impact on asset prices. Market sentiment indicators include confidence index, fear index, fund flows, margin trading balance, and news sentiment analysis results, reflecting the psychological expectations and behavioral tendencies of market participants and can serve as predictive factors for short-term market fluctuations.

[0051] The historical market data is labeled, and the market status of different time periods is divided into several categories according to historical market performance. The categories include at least one of upward trend, downward trend, oscillating market, high volatility, and low volatility. Specifically, the annotation of historical market data involves converting raw market data into labeled data required for supervised learning, which is then used to train machine learning models to identify market conditions. The annotation process assigns a corresponding market condition category to each time period based on historical market performance.

[0052] An uptrend is defined as a sustained increase in asset prices over a period of time, which can be determined by quantitative conditions such as the slope of the price trend, the alignment of moving averages, or a cumulative positive return. A downtrend is defined as a sustained decrease in asset prices over a period of time, which can be determined by conditions such as a negative slope of the price trend, a bearish alignment of moving averages, or a cumulative negative return. A sideways market is defined as an asset price with no clear directional trend over a period of time, fluctuating repeatedly within a certain range, which can be determined by conditions such as the ratio of price volatility to trend strength or the frequency of alternating positive and negative returns. A high volatility state is defined as an asset price volatility significantly higher than the historical average, which can be determined by a volatility index exceeding a preset threshold. A low volatility state is defined as an asset price volatility significantly lower than the historical average, which can be determined by a volatility index below a preset threshold. Labeling can be done using rule-based methods to automatically classify market states based on the above quantitative conditions, or by combining expert experience for manual verification of boundary cases. The labeled results serve as true labels for training machine learning models, used by the models to learn the mapping relationship from market data to market states.

[0053] An initial machine learning model is constructed, using the historical market data as input features and the corresponding market state category as training labels, and supervised learning training is performed on the initial machine learning model. Specifically, the initial machine learning model construction includes selecting a model architecture and defining model parameters. The model architecture can employ classification models such as Support Vector Machines, Random Forests, Gradient Boosting Trees, and Long Short-Term Memory Networks. Input features are preprocessed historical market data, including asset price sequences, trading volume data, macroeconomic indicators, and market sentiment indicators. These features are standardized, normalized, or feature-engineered before being used as input variables. Training labels are market state categories generated in the aforementioned labeling steps, such as upward trend, downward trend, sideways market, high volatility, and low volatility, serving as the target output for model learning. The supervised learning training process involves inputting the input features and corresponding training labels into the model, adjusting the model parameters through optimization algorithms to minimize the difference between the model's predicted output and the true labels. During training, a loss function is used to measure prediction error, and the model weights are updated through backpropagation. Training data is typically divided into a training set and a validation set. The training set is used for parameter learning, and the validation set is used to evaluate the model's generalization ability. An early stopping mechanism can be set during training, stopping training when the validation set error no longer decreases to prevent overfitting. After training, the model learns the mapping relationship from historical market data to market state categories, enabling it to identify and classify market states when new market data inputs are received.

[0054] The model parameters are optimized through cross-validation until the model's classification accuracy on the validation set reaches a preset threshold, thus obtaining the pre-trained machine learning model.

[0055] Specifically, cross-validation is a model evaluation and parameter optimization method used to assess the generalization ability of machine learning models and prevent overfitting. A commonly used cross-validation method is K-fold cross-validation, which randomly divides the training data into K equal-sized subsets. In each iteration, one subset is selected as the validation set, and the remaining K-1 subsets are used as the training set for model training. The classification accuracy of the model on the validation set is calculated. This process is repeated K times, each time selecting a different subset as the validation set. Finally, the average of the K validation accuracies is taken as the comprehensive evaluation metric for model performance. During cross-validation, the model's hyperparameters are tuned. Hyperparameters include, but are not limited to, the learning rate, the maximum depth of the tree model, the regularization coefficient, and the number of neurons in the hidden layer. Different combinations of hyperparameters are explored using methods such as grid search, random search, or Bayesian optimization. K-fold cross-validation is performed on each set of hyperparameters, and the combination of hyperparameters that maximizes the average validation accuracy is selected as the optimal hyperparameters. After the optimal hyperparameters are determined, the model is retrained using all the training data, and the classification accuracy of the model is evaluated on an independent validation set. If the validation accuracy reaches the preset threshold, the model training is complete, and a deployable pre-trained machine learning model is obtained. If the threshold is not reached, the model architecture is adjusted or the training data is expanded, and the above optimization process is repeated until the requirements are met.

[0056] In healthcare management scenarios, the pre-trained machine learning model can be obtained through the following steps: Historical health data is acquired, including at least one of the following: physiological indicator sequences, activity data, medical records, and health behavior indicators. Physiological indicator sequences include continuous monitoring data of changes over time, such as heart rate, blood pressure, blood glucose, weight, and body temperature; activity data includes daily steps, exercise duration, calorie consumption, and exercise type; medical records include past medical history, diagnoses, medication records, hospitalization records, and surgical records; and health behavior indicators include dietary records, sleep duration and quality, smoking and alcohol consumption, and self-reported stress levels.

[0057] The historical health data is labeled, and the health status of different time periods is divided into several categories based on historical health performance. These categories include at least one of the following: health improvement period, health stability period, health decline period, high-risk state, and low-risk state. The health improvement period is defined as a period in which physiological indicators show a continuous positive trend, such as a sustained decrease in blood sugar and steady weight loss. The health stability period is defined as a period in which physiological indicators remain within the normal range without significant fluctuations. The health decline period is defined as a period in which physiological indicators show a continuous deterioration trend, such as a gradual increase in blood pressure and a decline in exercise capacity. The high-risk state is defined as a period in which physiological indicators exceed the danger threshold or acute symptoms appear. The low-risk state is defined as a period in which all health indicators are within the safe range without significant abnormalities.

[0058] An initial machine learning model is constructed using the historical health data as input features and the corresponding health status categories as training labels. Supervised learning training is then performed on this initial machine learning model. The input features include the temporal, statistical, and trend characteristics of the multi-dimensional health data, and the training labels are the aforementioned labeled health status categories. The model learns the mapping relationship between health data and health status, thereby acquiring the ability to identify health status from monitoring data.

[0059] The model parameters are optimized through cross-validation until the model's classification accuracy on the validation set reaches a preset threshold, thus obtaining the pre-trained machine learning model. Cross-validation uses the K-fold validation method to evaluate the model's generalization ability on different data subsets. The model's hyperparameters are tuned through grid search or random search to ensure that the model's classification accuracy on unseen data meets application requirements.

[0060] The ability to obtain deployable pre-trained machine learning models for real-time identification of patients' health status provides a basis for dynamic adjustment of subsequent health intervention programs, which helps improve the accuracy and response speed of health status identification.

[0061] As can be seen, in the above scheme, the pre-trained machine learning model identifies market trends and potential risks and generates market state labels, enabling real-time perception of changes in the market environment. This overcomes the shortcomings of traditional static optimization methods in adapting to dynamic market changes, making resource portfolio adjustments more forward-looking and timely. The multi-objective optimization algorithm dynamically adjusts the asset allocation scheme, achieving an optimal balance among multiple objectives such as returns, risk, and liquidity, avoiding the problem of unreasonable allocation caused by single-objective optimization. The machine learning model screens factors that significantly impact returns, and the reinforcement learning algorithm dynamically adjusts factor weights, achieving adaptive optimization of factor weights. This overcomes the shortcomings of existing technologies, such as simple factor weight adjustment and a lack of intelligent mechanisms, allowing the resource portfolio to better capture market opportunities. Adjusting the optimized allocation results based on the optimal factor combination organically combines factor optimization with multi-objective optimization, forming a synergistic enhancement optimization mechanism. This significantly improves the final resource portfolio allocation scheme in terms of profitability, risk control, liquidity management, and market adaptability, realizing intelligent, dynamic, and multi-objective synergy in resource portfolio optimization. This significantly improves the overall performance of the resource portfolio and reduces manual rebalancing costs and operational risks.

[0062] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0063] In one embodiment, a resource allocation dynamic optimization device is provided, which corresponds one-to-one with the resource allocation dynamic optimization method described in the above embodiments. For example... Figure 5 As shown, Figure 5 This is a schematic diagram of a resource allocation dynamic optimization device according to an embodiment of the present invention. The device includes a data acquisition module 51, a multi-objective optimization module 52, a factor optimization module 53, and a result output module 54. Detailed descriptions of each functional module are as follows: The data acquisition module 51 is used to acquire market data and target information, and to identify and classify market trends and potential risks based on the market data and target information through a pre-trained machine learning model, and generate market status labels to characterize the current market status. The multi-objective optimization module 52 is used to dynamically adjust the resource allocation and asset allocation scheme according to the market status label and the target information, and generate multi-objective optimized allocation results. The factor optimization module 53 is used to use the machine learning model to screen factors that have a significant impact on returns, dynamically adjust the weights of the factors through a reinforcement learning algorithm, and generate the optimal factor combination based on the optimized factor weights. The result output module 54 is used to adjust the optimized configuration result according to the optimal factor combination and output the final resource configuration scheme.

[0064] In one embodiment, the multi-objective optimization module 52 is specifically used for: Incorporate at least two of the objectives of maximizing returns, minimizing risks, and optimizing liquidity into the objective function; Set constraints, which include at least one of transaction costs, industry restrictions, and geographical restrictions; The objective function is solved under the constraints using a genetic algorithm or a particle swarm optimization algorithm to generate a Pareto optimal solution set, and a multi-objective optimization configuration result is selected from the Pareto optimal solution set according to the objective.

[0065] In one embodiment, the factor optimization module 53 is specifically used for: The machine learning model is used to analyze the preset set of candidate factors, and the factors are sorted according to the degree of correlation between each factor and the return. Factors with a correlation degree exceeding a preset threshold are selected as target factors. Using the current market state as the state space, the adjustment of factor weights as the action space, and the performance as the reward signal, the algorithm learns the optimal factor weight adjustment strategy under different market states through iterative training using reinforcement learning, and dynamically adjusts the weights of each target factor based on the learned strategy.

[0066] In one embodiment, the result output module 54 is specifically used for: The weights of each factor in the optimal factor combination are mapped to the corresponding assets or asset classes to determine the exposure of each asset in the factor dimension. The comprehensive factor score of each asset is calculated based on the exposure, and the asset allocation ratio in the optimized allocation result is weighted and adjusted to favor assets with higher factor scores. The adjusted asset allocation ratio is normalized to ensure that the constraints are met, and the final resource allocation scheme is generated.

[0067] In one embodiment, the resource allocation dynamic optimization device is further configured to: Acquire historical market data, which includes at least one of asset price series, trading volume data, macroeconomic indicators, and market sentiment indicators; The historical market data is labeled, and the market status of different time periods is divided into several categories according to historical market performance. The categories include at least one of upward trend, downward trend, oscillating market, high volatility, and low volatility. An initial machine learning model is constructed, using the historical market data as input features and the corresponding market state category as training labels, and supervised learning training is performed on the initial machine learning model. The model parameters are optimized through cross-validation until the model's classification accuracy on the validation set reaches a preset threshold, thus obtaining the pre-trained machine learning model.

[0068] Specific limitations regarding the dynamic resource allocation optimization device can be found in the limitations of the dynamic resource allocation optimization method described above, and will not be repeated here. Each module in the aforementioned dynamic resource allocation optimization device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0069] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 6 As shown, Figure 6 This is a schematic diagram of a computer device according to an embodiment of the present invention. The computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with external clients via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a data processing method for insurance business on the server side.

[0070] In one embodiment, a computer device is provided, which may be a client, and its internal structure diagram may be as follows: Figure 7 As shown, Figure 7 This is another schematic diagram of a computer device according to an embodiment of the present invention. The computer device includes a processor, memory, network interface, display screen, and input device connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it implements the functions or steps on the client side of the method.

[0071] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps: Acquire market data and target information, and identify market trends and potential risks using a pre-trained machine learning model based on the market data and target information, generating market state labels to characterize the current market state; Based on the market status label and the target information, a multi-objective optimization algorithm is used to dynamically adjust the resource allocation and asset allocation scheme, generating multi-objective optimized allocation results; The machine learning model is used to screen factors that have a significant impact on returns. The weights of these factors are dynamically adjusted using a reinforcement learning algorithm. Based on the optimized factor weights, an optimal factor combination is generated. The optimized configuration result is adjusted based on the optimal factor combination to output the final resource configuration scheme.

[0072] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor: Acquire market data and target information, and identify market trends and potential risks using a pre-trained machine learning model based on the market data and target information, generating market state labels to characterize the current market state; Based on the market status label and the target information, a multi-objective optimization algorithm is used to dynamically adjust the resource allocation and asset allocation scheme, generating multi-objective optimized allocation results; The machine learning model is used to screen factors that have a significant impact on returns. The weights of these factors are dynamically adjusted using a reinforcement learning algorithm. Based on the optimized factor weights, an optimal factor combination is generated. The optimized configuration result is adjusted based on the optimal factor combination to output the final resource configuration scheme.

[0073] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0074] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0075] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0076] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A dynamic optimization method for resource allocation, characterized in that, include: Acquire market data and target information, and identify market trends and potential risks based on the market data and target information using a pre-trained machine learning model to generate market state labels that characterize the current market state; Based on the market status label and the target information, a multi-objective optimization algorithm is used to dynamically adjust the resource allocation scheme and generate multi-objective optimized allocation results. The machine learning model is used to screen factors that have a significant impact on returns. The weights of these factors are dynamically adjusted using a reinforcement learning algorithm, and an optimal factor combination is generated based on the optimized factor weights. The optimized configuration result is adjusted based on the optimal factor combination to output the final resource configuration scheme.

2. The dynamic optimization method for resource allocation according to claim 1, characterized in that, The acquisition of market data and target information includes: Real-time data collection of asset prices, macroeconomic indicators, and trading volume data is achieved through data interfaces or data platforms, and information on return targets, risk preferences, and restrictions is obtained.

3. The dynamic optimization method for resource allocation according to claim 1, characterized in that, The process of dynamically adjusting the resource allocation scheme using a multi-objective optimization algorithm to generate multi-objective optimized allocation results includes: Incorporate at least two of the objectives of maximizing returns, minimizing risks, and optimizing liquidity into the objective function; Set constraints, which include at least one of transaction costs, industry restrictions, and geographical restrictions; The objective function is solved under the constraints using a genetic algorithm or a particle swarm optimization algorithm to generate a Pareto optimal solution set, and a multi-objective optimization configuration result is selected from the Pareto optimal solution set according to the objective.

4. The dynamic optimization method for resource allocation according to claim 3, characterized in that, The step of using the machine learning model to screen factors that have a significant impact on returns, and dynamically adjusting the weights of the factors through a reinforcement learning algorithm, includes: The machine learning model is used to analyze the preset set of candidate factors, and the factors are sorted according to the degree of correlation between each factor and the return. Factors with a correlation degree exceeding a preset threshold are selected as target factors. Using the current market state as the state space, the adjustment of factor weights as the action space, and the performance as the reward signal, the algorithm learns the optimal factor weight adjustment strategy under different market states through iterative training using reinforcement learning, and dynamically adjusts the weights of each target factor based on the learned strategy.

5. The dynamic optimization method for resource allocation according to claim 4, characterized in that, The preset set of candidate factors includes at least one of the following: value factor, momentum factor, size factor, quality factor, and volatility factor.

6. The dynamic optimization method for resource allocation according to claim 4, characterized in that, The step of adjusting the optimized configuration result based on the optimal factor combination includes: The weights of each factor in the optimal factor combination are mapped to the corresponding assets or asset classes to determine the exposure of each asset in the factor dimension. The comprehensive factor score of each asset is calculated based on the exposure, and the asset allocation ratio in the optimized allocation result is weighted and adjusted to favor assets with higher factor scores. The adjusted asset allocation ratio is normalized to ensure that the constraints are met, and the final resource allocation plan is generated.

7. The dynamic optimization method for resource allocation according to claim 1, characterized in that, The pre-trained machine learning model is obtained through the following steps: Acquire historical market data, which includes at least one of asset price series, trading volume data, macroeconomic indicators, and market sentiment indicators; The historical market data is labeled, and the market status of different time periods is divided into several categories according to historical market performance. The categories include at least one of upward trend, downward trend, oscillating market, high volatility, and low volatility. An initial machine learning model is constructed, using the historical market data as input features and the corresponding market state category as training labels, and supervised learning training is performed on the initial machine learning model. The model parameters are optimized through cross-validation until the model's classification accuracy on the validation set reaches a preset threshold, thus obtaining the pre-trained machine learning model.

8. A dynamic optimization device for resource allocation, characterized in that, include: The data acquisition module is used to acquire market data and target information, and to identify and classify market trends and potential risks based on the market data and target information through a pre-trained machine learning model, generating market status labels to characterize the current market state. The multi-objective optimization module is used to dynamically adjust the resource allocation scheme based on the market status label and the target information using a multi-objective optimization algorithm, and generate multi-objective optimized configuration results. The factor optimization module is used to use the machine learning model to screen factors that have a significant impact on returns, dynamically adjust the weights of the factors through a reinforcement learning algorithm, and generate the optimal factor combination based on the optimized factor weights. The result output module is used to adjust the optimized configuration result according to the optimal factor combination and output the final resource configuration scheme.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the dynamic optimization method for resource allocation as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the dynamic optimization method for resource allocation as described in any one of claims 1 to 7.

Images