Investment optimization system based on artificial intelligence investment research

Abstract

The invention discloses an investment optimization system based on artificial intelligence investment research, relates to the technical field of artificial intelligence, and realizes comprehensive utilization of cross-frequency information through decomposition and fusion of multi-frequency original data streams, simultaneous capture of trend factors and resonance factors and calculation of dynamic cooperation coefficients. On the basis of market volatility calculation and rolling time window analysis, a volatility reference value and dispersion are further obtained, and a weighted combination and scaling processing mechanism is introduced, so that the generation of a dynamic threshold can give consideration to the overall trend and local fluctuation characteristics. Through a boundary constraint link, excessive deviation of a threshold value is effectively avoided, and decision stability is kept. And finally, identifying a market resonance state and generating an investment portfolio adjustment instruction in combination with a relationship between the dynamic cooperation coefficient and a dynamic threshold value.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, specifically to an investment optimization system based on artificial intelligence investment research. Background Technology

[0002] With the rapid development of financial markets, investment and trading activities are becoming increasingly frequent and complex. Traditional investment research methods often rely on statistical analysis of single-frequency data or the construction of empirical rules, which have limitations in capturing short-term market fluctuations and long-term trends. In recent years, artificial intelligence technology has been widely introduced into the field of investment research to assist in processing large-scale, multi-frequency market data, aiming to improve the scientific rigor and timeliness of investment decisions through modeling and intelligent methods.

[0003] In intelligent investment research, threshold mechanisms are often used to help identify market conditions and trigger portfolio adjustments. However, existing methods generally use fixed thresholds or simply rely on a single volatility indicator, making it difficult to adapt to dynamic changes in the market environment. Especially when facing fluctuations of varying intensity and frequency, fixed thresholds often lead to lag or oversensitivity in strategy responses, reducing the effectiveness of investment decisions. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an investment optimization system based on artificial intelligence investment research.

[0005] To achieve the above objectives, the technical solution of the present invention is as follows:

[0006] This invention discloses an investment optimization system based on artificial intelligence investment research, comprising:

[0007] The data acquisition module is used to acquire multi-frequency raw data streams from the investment market. The multi-frequency raw data streams include at least high-frequency tick-by-tick transaction data, which are tick-by-tick transaction data at the second or tick level.

[0008] The coordination coefficient calculation module is used to decompose and fuse the multi-frequency raw data stream, generate trend factors and resonance factors, and calculate the dynamic coordination coefficient between the two.

[0009] The volatility calculation module is used to calculate market volatility in real time based on the high-frequency transaction data.

[0010] The benchmark dispersion calculation module is used to calculate the mean and variance of the market volatility within a window based on a preset time length or the number of transaction events, by dynamically sliding according to the data updates within the window, to obtain the volatility benchmark value and volatility dispersion.

[0011] The weighted combination module is used to weight and combine the volatility benchmark value with a preset decision threshold to obtain a first adjustment value;

[0012] The scaling processing module is used to calculate a scaling factor based on the volatility dispersion, and to perform a calculation between the first adjustment value and the scaling factor to obtain a second adjustment value;

[0013] The boundary constraint module is used to constrain the second adjustment value within a preset upper and lower limit range to obtain a dynamic decision threshold.

[0014] The instruction generation module is used to determine the market resonance state and generate a resonance state signal based on whether the dynamic synergy coefficient is greater than the dynamic decision threshold, and to generate an investment portfolio adjustment instruction based on the resonance state signal.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0016] 1. By calculating the volatility benchmark value and volatility dispersion that reflect real-time market changes, and then converting the two types of information into threshold adjustment results through weighted combination and scaling, and implementing boundary constraints between preset upper and lower limits, the generated dynamic threshold can be flexibly adjusted with market conditions and kept within a reasonable range to avoid decision distortion caused by extreme fluctuations.

[0017] 2. It can identify the resonance effect between different frequency factors and generate portfolio adjustment instructions based on the resonance state; it can effectively improve the response speed and stability of investment strategies to complex market fluctuations, and in practical applications, it can improve the level of risk control and the rationality of asset allocation. Attached Figure Description

[0018] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts. Wherein:

[0019] Figure 1 This is a system architecture diagram of the present invention;

[0020] Figure 2 This is a data flow diagram of the present invention. Detailed Implementation

[0021] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0022] Application Overview

[0023] Existing trading signal generation methods generally suffer from rigid threshold settings, insufficient sensitivity to market fluctuations, and difficulty in accurately identifying market resonance states under the synergistic effect of multiple frequency factors. Most existing methods rely on fixed thresholds or single factor-driven approaches, making it difficult to take into account the dynamic changes of high-frequency tick-by-tick data and low-frequency trend factors, thus leading to signal distortion or decision delays.

[0024] To address this, this invention proposes a method for identifying market resonance states and dynamically adjusting thresholds based on multi-source factor synergy. This method introduces a joint modeling mechanism of high-frequency trading factors and low-frequency trend factors, combined with the dynamic extraction of resonance factors. An initial adjustment value is generated using a weighted combination module, further corrected through scaling based on volatility dispersion, and finally, a dynamic decision threshold is obtained within a preset boundary range. By comparing this dynamic threshold with the synergy coefficient, the market resonance state can be identified in real time, and corresponding trading signals can be generated.

[0025] Compared with existing technologies, this invention achieves dynamic adaptation to market conditions through the joint design of multi-level factor fusion, scaling correction and boundary constraints, which significantly improves the stability and flexibility of signal generation and can better meet the intelligent decision-making needs in high-frequency fluctuation environments.

[0026] like Figures 1-2 As shown, this invention discloses an investment optimization system based on artificial intelligence investment research, comprising:

[0027] The module includes a data acquisition module, a coordination coefficient calculation module, a volatility calculation module, a benchmark discrete calculation module, a weighted combination module, a scaling processing module, a boundary constraint module, and an instruction generation module.

[0028] The data acquisition module is used to acquire multi-frequency raw data streams from the investment market, including high-frequency tick-by-tick transaction data, mid-frequency technical indicator data, and low-frequency macroeconomic fundamental data. High-frequency tick-by-tick data can be transaction data at the second or tick level, mid-frequency technical indicator data can be technical indicator data at the minute or hour level, and low-frequency macroeconomic fundamental data can be macroeconomic fundamental data at the daily or weekly level. Specifically, high-frequency data can be acquired in real time via exchange APIs, mid-frequency data can be generated using a technical indicator calculation library, and low-frequency data can be downloaded periodically from a macroeconomic database.

[0029] The synergy coefficient calculation module decomposes and fuses multi-frequency raw data streams to generate trend factors and resonance factors, and calculates the dynamic synergy coefficient between them. The trend factor can be obtained through moving average or wavelet decomposition of low-frequency data, while the resonance factor can be extracted through Fourier transform of high-frequency data. The dynamic synergy coefficient can be calculated using Pearson correlation coefficient or dynamic time warping algorithm.

[0030] The volatility calculation module calculates market volatility in real time based on high-frequency tick-by-tick trading data. For example, it can use the GARCH model or existing volatility calculation methods to calculate volatility in a 1-minute window.

[0031] The benchmark dispersion calculation module dynamically calculates the mean and variance of market volatility within a window of preset time length or number of trading events, obtaining the benchmark volatility value and volatility dispersion. The window length can be set to 20 trading days, and the calculation uses an exponentially weighted moving average method.

[0032] The weighted combination module combines the volatility benchmark value with a preset decision threshold to obtain the first adjustment value. The weights can be dynamically adjusted according to market conditions; for example, the benchmark value can be given higher weights during periods of high volatility.

[0033] The scaling module calculates a scaling factor based on volatility dispersion and then calculates a second adjustment value by combining the first adjustment value with the scaling factor. The scaling factor can be a linear or non-linear function, and the adjustment magnitude is amplified when the dispersion exceeds a threshold.

[0034] The boundary constraint module applies boundary constraints to the second adjustment value within a preset upper and lower limit range to obtain a dynamic decision threshold. The upper and lower limits can be set based on historical fluctuation extreme values ​​to prevent the threshold from deviating excessively from a reasonable range.

[0035] The instruction generation module determines the market resonance state and generates a resonance state signal based on whether the dynamic coordination coefficient is greater than the dynamic decision threshold, and then generates portfolio adjustment instructions.

[0036] This technical solution addresses the decision-making lag caused by single-frequency data analysis and fixed thresholds in traditional investment research through multi-frequency data fusion and a dynamic threshold mechanism. Compared to existing technologies, it can more accurately capture the synergistic relationship between market trends and short-term fluctuations, enabling timely optimization and adjustment of investment portfolios. Specifically, multi-frequency data processing improves the comprehensiveness of market state identification, while the dynamic threshold mechanism enhances the strategy's adaptability to different market environments.

[0037] The core purpose of calculating the first adjustment value is to dynamically determine a weight for merging a volatility benchmark (representing the current volatility level) and a preset decision threshold (representing an empirically fixed threshold) based on the current market volatility level, thereby generating an intelligent first adjustment value that adapts to the market environment. This process specifically includes the following steps:

[0038] Acquiring historical multi-frequency raw data streams: The system reads historical multi-frequency raw data streams of the target investment market (e.g., a stock index) over a relatively long historical period (e.g., the past 3 to 5 years) from the built-in database or external data interface. This data stream also includes high-frequency tick-by-tick data, mid-frequency technical indicator data, and low-frequency macroeconomic fundamental data, thereby ensuring consistency between historical and current data analysis foundations.

[0039] Calculate the historical price volatility series: Based on the price data in the aforementioned historical multi-frequency raw data stream (usually the daily closing price series of the mid-day frequency), calculate its historical volatility. Volatility can be calculated using methods well-known in the field, such as calculating the standard deviation of cycle returns (daily returns) or using more complex realized volatility models based on high-frequency data aggregation.

[0040] By performing daily rolling calculations, a historical price volatility sequence corresponding one-to-one with historical dates is finally obtained.

[0041] Classify the historical volatility series into states and determine quantiles: Perform statistical analysis on the historical price volatility series obtained in the above steps to quantitatively define the historical "high volatility" and "low volatility" states.

[0042] Calculate the specific quantiles for this sequence. The 80th quantile can be chosen as the upper quantile for volatility. This characterizes historical high volatility; the 20th percentile is selected as the lower volatility quantile. It represents a historical low-volatility state. By converting absolute volatility values ​​into relative positions with respect to their historical levels, the system can adapt to various assets with different absolute volatility levels.

[0043] The dynamic weights are calculated based on relative positional relationships to obtain the first adjustment value. The principle is that when the volatility benchmark value approaches the upper quantile, a higher weight is assigned to it, making the first adjustment value more sensitive to the current high volatility state; when it approaches the lower quantile, its weight is reduced to avoid over-adjustment in low-volatility environments. Compared with existing technologies, this scheme can more accurately adjust the decision threshold dynamically according to the intensity of market volatility, avoiding the lag in response of fixed thresholds during high volatility periods and preventing overreaction during low volatility periods.

[0044] Therefore, this technical solution solves the problem that traditional fixed thresholds are difficult to adapt to changes in market fluctuations by introducing historical fluctuation state division and dynamic weight calculation mechanism.

[0045] The process of obtaining the weighted portfolio aims to precisely quantify the relative position of the current volatility benchmark value within its historical distribution using a data-driven approach, and intelligently allocate its weight when weighted in a combination with a preset decision threshold. This process is achieved through the following steps:

[0046] First, we will perform three basic calculations:

[0047] 1. Calculate the difference. Its value is the aforementioned volatility benchmark value. With the lower quantile of the volatility The difference, that is:

[0048] ;

[0049] This difference reflects the extent to which current market volatility exceeds historical low volatility benchmarks.

[0050] 2. Calculate the difference two Its value is the upper quantile of volatility. With lower quantile of volatility The difference, that is:

[0051] ;

[0052] This difference represents the entire range of fluctuations between high and low volatility states in historical data, serving as a benchmark.

[0053] 3. Calculate the normalized intermediate value :

[0054] Linear normalization was achieved.

[0055] when hour, This indicates that the current volatility is at or below historical low volatility levels.

[0056] when hour, This indicates that the current volatility is at a historically moderate level.

[0057] when hour, This indicates that the current volatility is at or above historical high volatility levels.

[0058] This transforms an absolute volatility value into a scalar value that is theoretically unbounded relative to its own historical distribution.

[0059] Then, the obtained normalized intermediate values Input a predefined function mapping relationship The calculations are performed to obtain the final weight values.

[0060] Function mapping relationship Selection criteria: Its output value range must be within the range Within this range, the result is used to ensure that it can be used as a valid weight value. This function is used to "squeeze" normalized intermediate values ​​that may be less than 0 or greater than 1 into a legal weight range, and may introduce non-linear effects.

[0061] As a preferred embodiment, the above-described function mapping relationship It could be:

[0062] Linear cutoff function: This function maps values ​​less than 0 to 0, values ​​greater than 1 to 1, and maintains a linear relationship in between. This method is simple and direct.

[0063] The sigmoid function provides a smooth, non-linear mapping. This approach is better suited to simulating the non-linear psychological threshold in decision-making.

[0064] The output value of the function mapping relationship is directly used as the weight value assigned to the volatility benchmark. ,Right now: ;

[0065] Calculate the weight value assigned to the preset decision threshold. Its value is the difference between 1 and the weight value one, that is:

[0066] .

[0067] Thus, a complete weight combination is obtained: , This combination accurately reflects the system's trust ratio between "current observations" and "historical experience."

[0068] By establishing a dynamic weighting mechanism based on the relative positions of quantiles, the problem of traditional fixed weights being unable to adapt to changes in market conditions is solved. Mathematical modeling quantifies the impact of market conditions on weights, avoiding subjective bias; normalization enhances adaptability to different market environments; and the continuous nature of the function mapping ensures smooth weight changes, reducing transaction costs associated with frequent strategy adjustments.

[0069] The functional mapping relationship can specifically take the form of a sigmoid function. Its core advantage lies in providing a smooth, continuous, and diminishing marginal returns nonlinear mapping, thus more precisely simulating the psychological threshold in investment decisions. This mapping relationship is not established based on manual experience in setting parameters, but rather through a data-driven approach, enabling the function's key parameters to adapt to the volatility characteristics of different investment markets. The establishment process is as follows:

[0070] Calculate slope parameters :

[0071] slope parameter This determines the steepness of the S-curve, i.e., the sensitivity of the weight values ​​to changes in the normalized median.

[0072] The calculation formula is as follows:

[0073] ,in, This is an adjustable constant adjustment factor (for example, it can be optimized and set via historical backtesting). ).

[0074] The range of fluctuations in market history Inversely proportional:

[0075] When the market has a large historical fluctuation range (such as a highly volatile market), the denominator... The value is large, resulting in The value decreases. Smaller. The value will make the S-curve flatter. This means the weight value... normalized intermediate value The changes are slow, and the system responds more robustly and cautiously to changes in volatility, avoiding decision-making jitter caused by noise in a wide-ranging market.

[0076] When the market's historical volatility range is small (such as a stable trending market), the denominator... The value is small, resulting in The value increases. A larger value. This value makes the S-curve steeper. This means the system is more sensitive to changes in volatility and can switch decision weights more quickly, thus capturing trend opportunities in a timely manner.

[0077] therefore, It is an adaptive parameter that enables the system's decision sensitivity to automatically match the inherent volatility characteristics of the market being analyzed.

[0078] Calculate the center point position parameters :

[0079] Center point position parameters This determines the center point of the S-curve, which is the position of the normalized median value corresponding to a weight of 0.5. The calculation formula is as follows:

[0080] ;

[0081] in, This is a typical volatility value calculated based on the historical price volatility series. This value is typically taken as the median or mean of the series, preferably the median, as it is less sensitive to extreme values ​​and better represents the typical level of volatility.

[0082] This formula uses historical typical volatility values Mapped to normalized interval The significance of this parameter lies in its objective definition of the equilibrium point for system decision-making. For example, if the median volatility of a market is closer to its lower quantile (i.e., lower volatility most of the time), the calculated... The value will be less than 0.5. This means that when the volatility benchmark is still at a relatively low level, the weight value is already greater than 0.5, indicating that the decision-making process is beginning to favor the current volatility observation.

[0083] therefore, This ensures that the balance point of the system's decision-making is consistent with the historical norm of the analyzed market, avoiding the influence of subjective concepts.

[0084] The calculated slope parameters Center point position parameters Substituting these into the sigmoid function forms a complete, parameter-adaptive function mapping relationship.

[0085] ,in For normalized intermediate values, The weight values ​​assigned to the volatility benchmark in the final output.

[0086] This function provides a smooth non-linear mapping, and the output value is strictly limited to between 0 and 1, and due to the slope parameter Center point position parameters Driven by historical data, the function mapping relationship can automatically adapt to any investment product being analyzed (such as stocks, futures, foreign exchange, etc.) without the need for manual parameter readjustment. It has a high degree of intelligence and strong generalization ability.

[0087] The calculation of the second adjustment value introduces a threshold judgment mechanism based on volatility dispersion, enabling conditional scaling of the first adjustment value, thereby significantly improving the system's robustness in extreme market environments. This process specifically includes the following steps:

[0088] The calculated volatility dispersion (i.e., the standard deviation or variance of volatility within a set time window) is compared with a preset dispersion threshold. The value of the preset dispersion threshold can be determined by analyzing historical data. For example, a high quantile (such as the 75th or 90th quantile) of the historical volatility dispersion sequence can be calculated and set as the threshold. The technical implication is that this threshold represents the critical point at which market volatility transitions from an "orderly change" to a "disorderly chaos" state.

[0089] Based on the comparison results above, execute the conditional branch logic to calculate the scaling factor:

[0090] If the volatility dispersion exceeds a preset dispersion threshold, the market is determined to be in a "high dispersion and disorder state." This state not only signifies volatile market fluctuations (characterized by the volatility benchmark value) but also indicates that such fluctuations are unstable and unpredictable (characterized by high dispersion), implying a high tail risk. In this state, the system activates a scaling mechanism. The scaling factor is calculated as the product of the volatility dispersion and a fixed coefficient. This fixed coefficient is a magnification factor greater than 0 (e.g., set to 1.2), and its specific value can be determined through historical backtesting optimization. The purpose is that the higher the market disorder (dispersion), the larger the scaling factor, and the greater the subsequent scaling adjustment. This is a measure to strengthen risk control.

[0091] If the volatility dispersion is less than or equal to a preset dispersion threshold, the system determines that the market is currently in a "low-dispersion ordered state." In this state, market volatility may be dramatic, but its patterns of change are relatively stable and predictable. In this state, the system considers no additional scaling adjustments necessary. Therefore, the scaling factor is set to a default value of 1. In subsequent calculations, any number multiplied by 1 equals itself. Therefore, when the market is not chaotic, this scaling calculation step is essentially "skipped," and the system directly adopts the result of the first adjustment value. This avoids introducing unnecessary operations in normal or trending markets, maintaining the purity and aggressiveness of the strategy.

[0092] Finally, the first adjustment value is multiplied by the scaling factor to obtain the second adjustment value.

[0093] When the market is orderly, the system makes decisions based on the first adjustment value (weighted by volatility level and historical experience).

[0094] When the market is in turmoil, the system will further amplify the final dynamic decision threshold based on the first adjustment value, making the conditions for generating trading signals more stringent (for example, requiring a stronger correlation coefficient to trigger a "bullish" signal), thereby automatically entering a more conservative and risk-controlled operating mode.

[0095] The purpose of trend factors and resonance factors is to decompose the mixed raw market data into characteristic components representing different market forces, thereby providing clear information input for subsequent collaborative analysis. This process is achieved through the following steps:

[0096] First, the acquired multi-frequency raw data streams are preprocessed to generate standardized market data that can be used for analysis:

[0097] Data with different frequencies can be unified to the same timestamp using interpolation or aggregation methods. For example, tick-level data can be aggregated into minute-level data to align with the time frequency of minute-level technical indicator data.

[0098] Methods such as z-score standardization or maximum-minimum normalization can be used to eliminate differences in units and numerical ranges between different data sources. For example, price data, trading volume data, and technical indicator data can be standardized to have a mean of 0 and a standard deviation of 1.

[0099] Fill in or remove any missing values ​​in the data to ensure data continuity.

[0100] The standardized market data is subjected to signal decomposition processing to generate multiple intrinsic mode components of different frequencies:

[0101] Empirical Mode Decomposition (EMD) or its improved algorithms (such as EEMD and CEEMDAN) are used as the preferred implementation method. This algorithm can adaptively decompose a complex non-stationary time series signal into several intrinsic mode functions (IMFs), and these IMF components are arranged in descending order of frequency.

[0102] Each IMF component represents an oscillation mode within a specific frequency range in the original signal, and satisfies the following conditions:

[0103] Throughout the entire data range, the number of extreme points is equal to or differs from the number of zero-crossing points by at most one;

[0104] At any point, the mean of the upper envelope defined by the local maxima and the lower envelope defined by the local minima is zero.

[0105] The trend factor is generated by superimposing the eigenmode components with the lowest frequency (e.g., the first 3).

[0106] These low-frequency components represent the long-term trend components in the original signal after removing short-term fluctuations and noise. This factor can clearly depict the long-term directional movement of the market and filter out short-term noise interference.

[0107] The resonance factor is generated by superimposing the eigenmode components with the highest predetermined number of frequencies (e.g., the last two).

[0108] These high-frequency components contain short-term fluctuations and emotional trading elements from the original signal. This factor effectively captures short-term emotional pulses and minute fluctuations in the market, reflecting the immediate trading behavior of market participants.

[0109] Dynamic correlation coefficients are used to accurately quantify the morphological synergy between trend factors and resonance factors, even if the two series have phase delays or stretching on the time axis. This process addresses the limitation of traditional correlation coefficients being overly sensitive to phase differences when processing financial time series by introducing the Dynamic Time Warping (DTW) algorithm.

[0110] The specific process is as follows:

[0111] Obtain the trend factor sequence and resonance factor sequence generated after decomposition and fusion. These two sequences are numerical sequences of the same length and with aligned timestamps, representing the long-term trend movement and short-term emotional fluctuations of the market, respectively.

[0112] The Dynamic Time Warping (DTW) algorithm is used to calculate the minimum warping path cumulative distance. Its goal is to find the optimal nonlinear alignment between two sequences and calculate its cumulative cost.

[0113] First, calculate the Euclidean distance (or other distance metric, such as Manhattan distance) between any two points in the two sequences, forming a... Distance matrix , where matrix elements , Represents the first in the trend factor sequence One element, The first in the resonance factor sequence The matrix quantifies the trend factor sequence of the nth element. Point and resonance factor sequence Local differences between points.

[0114] The goal of the dynamic time warping algorithm is to find a path through the distance matrix. A regular path, where each point Points representing the trend factor sequence Points of the sequence resonance factor sequence Matching.

[0115] This path must satisfy:

[0116] Boundary conditions: The starting point of the path is The destination is .

[0117] Monotonicity and continuity: The path must progress monotonically over time.

[0118] The algorithm uses dynamic programming to find a path that minimizes the cumulative distance of the regularized path. This cumulative distance... The calculation formula is:

[0119] ;in, From the starting point Time The minimum cumulative distance.

[0120] final, .

[0121] Obtain the cumulative distance of the minimum regularized path Then, it needs to be transformed into a representation of the strength of synergy and its range. The dynamic synergy coefficient between them.

[0122] The calculation of the dynamic coherence coefficient can be expressed as converting the cumulative distance of the minimum regularized path into a value between 0 and 1 using a preset normalization function, where a larger value indicates a higher morphological coherence between the two sequences. As a preferred implementation, the normalization function can be an exponential decay function or a linear scaling function.

[0123] The dynamic decision threshold ensures that the decision thresholds generated by the system are always within a reasonable range, while providing important status information for risk control through a labeling signal mechanism. This process is achieved through the following steps:

[0124] The system compares the calculated second adjustment value with pre-set upper and lower boundary values; the upper and lower limits are safe operating ranges set based on historical backtesting and risk appetite. The upper boundary value typically corresponds to the maximum acceptable threshold in extremely high-risk market environments, while the lower boundary value corresponds to the minimum threshold during periods of extreme market calm. These boundary values ​​can be determined by analyzing historical extreme market data, for example, by taking the 95th and 5th percentiles of historical threshold distributions.

[0125] When the second adjustment value is greater than the upper limit value, the dynamic decision threshold is set to the upper limit value and a first flag signal is generated; this signal is a Boolean or numeric flag used to record that the current decision threshold has reached the upper limit allowed by the system.

[0126] When the second adjustment value is less than the lower limit, the dynamic decision threshold is set to the lower limit and a second marker signal is generated; this signal records that the current decision threshold has reached the lower limit allowed by the system.

[0127] The first or second flag signal is used to trigger additional risk control operations on the portfolio adjustment instruction.

[0128] The generation of the marker signal can be implemented through a state machine. When a boundary constraint is detected to be triggered, a corresponding marker is immediately generated and stored in the event queue. Additional risk control operations include, but are not limited to: reducing the position ratio, extending the rebalancing interval, and increasing hedging positions.

[0129] This technical solution effectively addresses the issue of threshold failure caused by extreme market volatility in existing technologies by introducing a boundary constraint mechanism. When market volatility is high, the decision threshold is automatically limited to a reasonable range to avoid generating overly aggressive or conservative trading signals. Simultaneously, additional risk control measures are triggered by flagged signals, enhancing system robustness while maintaining the core strategy logic. Compared to fixed threshold methods, this solution can adapt to different market volatility environments, controlling extreme risks while maintaining strategy sensitivity.

[0130] The resonance state signal addresses the problem of false signals caused by market noise. By introducing a signal confirmation mechanism, it ensures that high-confidence trading orders are triggered only when a resonance state has been consistently verified. This process is achieved through the following steps:

[0131] The system continuously assesses the relationship between the dynamic coordination coefficient and the dynamic decision threshold at multiple consecutive time points (e.g., 5 consecutive minute periods):

[0132] A series of consecutive time points constitute an observation window. The window length (i.e., the number of times, such as 3 to 5 times) can be set according to market characteristics. A shorter window makes the system more sensitive, while a longer window makes the system more robust.

[0133] At each point in time, the system performs a judgment: whether the dynamic coordination coefficient is greater than the dynamic decision threshold. This judgment aims to verify whether the market's coordinated state continues to hold.

[0134] Based on the results of continuous judgment, the system generates two types of resonance state signals with different confidence levels:

[0135] Type I resonance state signal (strong confirmation signal):

[0136] Generation conditions: In a predefined number of consecutive judgments (e.g., three consecutive cycles), the dynamic synergy coefficient is greater than the dynamic decision threshold. This indicates a market resonance state: short-term sentiment fluctuations are highly consistent with the long-term trend direction, which has been continuously verified and is not a transient noise phenomenon. This signal is a high-confidence trading signal, signifying a high probability of trend continuation opportunities.

[0137] Second type of resonance state signal (weakly confirmed / unconfirmed signal):

[0138] Generation condition: In continuous judgments, the number of times the dynamic synergy coefficient exceeds the dynamic decision threshold has not reached the predefined number. This indicates that the market resonance state has either just formed and has not yet been confirmed, or it has been disrupted. This could be due to market noise, short-term emotional fluctuations, or the early stages of a genuine trend reversal. This signal is a low-confidence or neutral state signal, suggesting that the system needs to remain cautious.

[0139] The system triggers differentiated portfolio adjustment instructions based on the type of signal generated:

[0140] When a first-type resonance state signal is generated: the system triggers a standard portfolio adjustment command. This is an operational command based on a complete model signal and normal strength, for example:

[0141] Adjust the portfolio by 80% to 100% of the preset positions;

[0142] Set stop-loss and take-profit levels according to the original risk-reward ratio.

[0143] When a second type of resonance state signal is generated: the system triggers either a tentative portfolio adjustment instruction or an instruction to maintain the current holdings. This includes two processing methods:

[0144] Exploratory operations: Only execute operations with a relatively small position (such as 20% to 50% of the standard position) to test the market direction while controlling risk and to make advance arrangements for possible trends.

[0145] Maintain the status quo: Do ​​not perform any position adjustments, continue to hold existing positions and observe, and wait for further clearer signals.

[0146] The portfolio adjustment instruction generation process constructs a two-level decision-making mechanism, combining preliminary decisions based on market analysis with risk control based on system self-checks, thereby generating intelligent investment instructions that balance return pursuit and risk control. This process is achieved through the following steps:

[0147] The system first generates preliminary portfolio adjustment instructions based on the received resonance state signal:

[0148] If the resonance state signal is: strong resonance bullish, then a buy-in tendency order is generated, which includes operations such as increasing the position of risky assets and reducing the position of safe-haven assets.

[0149] If the resonance state signal is: strong resonance bearish, then a sell-in tendency order is generated, which includes operations such as reducing risky asset positions and increasing safe-haven asset positions.

[0150] If the resonance state signal is weak resonance or no resonance, an instruction to maintain the current position will be generated.

[0151] The specific content of the instruction includes: the initial instruction usually includes the underlying asset code, the trading direction (buy / sell), the suggested trading quantity or proportion, and the suggested transaction price range.

[0152] The system then checks whether a valid first or second marker signal exists:

[0153] The validity period of the marker signal can be set to a fixed time window (e.g., 30 minutes). Marker signals generated within this time window are considered valid signals.

[0154] The system distinguishes between the first and second marker signals, and performs differentiated processing based on the judgment result of the marker signals:

[0155] If a valid first-signal signal exists, it indicates that the threshold calculated by the system has reached its upper limit, and the market is in an extremely high-risk state. At this point, conservative risk control operations are triggered, such as:

[0156] Reduce the number of transactions in the initial order proportionally (e.g., reduce by 50%).

[0157] Attach stricter stop-loss conditions to trading orders (e.g., adjust the stop-loss range from -5% to -3%).

[0158] Reduce the maximum risk exposure of a single transaction.

[0159] If a valid second marker signal exists, it indicates that the threshold calculated by the system has reached its lower limit, and the market is in an abnormally calm state. At this point, prudent risk control operations are triggered, such as:

[0160] The model may underestimate potential risks;

[0161] Additional review of trading instructions;

[0162] Reduce the size of trades and use exploratory positions.

[0163] The risk control module outputs risk control correction parameters based on the type of the flag signal. These parameters correct the initial portfolio adjustment instructions and generate the final portfolio adjustment instructions.

[0164] When no valid marker signal exists:

[0165] The system determines that no special risk control measures are needed at present, and directly outputs the preliminary portfolio adjustment instruction as the final portfolio adjustment instruction.

[0166] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. An investment optimization system based on artificial intelligence investment research, characterized in that: include: The data acquisition module is used to acquire multi-frequency raw data streams from the investment market. The multi-frequency raw data streams include at least high-frequency tick-by-tick transaction data, which are tick-by-tick transaction data at the second or tick level. The coordination coefficient calculation module is used to decompose and fuse the multi-frequency raw data stream, generate trend factors and resonance factors, and calculate the dynamic coordination coefficient between the two. The volatility calculation module is used to calculate market volatility in real time based on the high-frequency transaction data. The benchmark dispersion calculation module is used to calculate the mean and variance of the market volatility within a window based on a preset time length or the number of transaction events, by dynamically sliding according to the data updates within the window, to obtain the volatility benchmark value and volatility dispersion. The weighted combination module is used to weight and combine the volatility benchmark value with a preset decision threshold to obtain a first adjustment value; The scaling processing module is used to calculate a scaling factor based on the volatility dispersion, and to perform a calculation between the first adjustment value and the scaling factor to obtain a second adjustment value; The boundary constraint module is used to constrain the second adjustment value within a preset upper and lower limit range to obtain a dynamic decision threshold. The instruction generation module is used to determine the market resonance state and generate a resonance state signal based on whether the dynamic synergy coefficient is greater than the dynamic decision threshold, and to generate an investment portfolio adjustment instruction based on the resonance state signal.

2. The investment optimization system based on artificial intelligence investment research according to claim 1, characterized in that: The calculation process for the first adjustment value is as follows: Obtain the historical multi-frequency raw data stream of the investment market; Based on the price data in the historical multi-frequency raw data stream, calculate the historical price volatility sequence; The historical price volatility series is divided into states to determine the upper quantile of volatility representing high volatility states and the lower quantile of volatility representing low volatility states. Based on the relative positional relationship between the current calculated volatility benchmark value and the upper and lower quantiles of volatility, a weight combination for weighting the volatility benchmark value and the preset decision threshold is calculated, and a first adjustment value is obtained by weighting the volatility benchmark value and the preset decision threshold based on the weight combination.

3. The investment optimization system based on artificial intelligence investment research according to claim 2, characterized in that: The process of obtaining the weight combination is as follows: The difference between the volatility benchmark value and the lower quantile of volatility is denoted as difference one, the difference between the upper quantile of volatility and the lower quantile of volatility is denoted as difference two, and the ratio of difference one to difference two is denoted as the normalized median value. The normalized intermediate value is input into a predefined function mapping relationship for calculation, and the output value of the function mapping relationship is between 0 and 1. The output value of the function mapping relationship is used as the first weight value assigned to the volatility benchmark value, and the difference between 1 and the first weight value is recorded as the second weight value of the first preset threshold.

4. The investment optimization system based on artificial intelligence investment research according to claim 3, characterized in that: The mapping process of the function mapping relationship is as follows: Obtain the upper quantile of volatility Volatility lower quantile And calculate the slope parameter. : ,in, The constant adjustment coefficient; The average of historical price volatility series is taken as the typical volatility value. And calculate the center point position parameters. : ； Based on slope parameter Center point position parameters Establish function mapping relationships: ,in For normalized intermediate values, The weights are for the output.

5. The investment optimization system based on artificial intelligence investment research according to claim 1, characterized in that: The calculation process for the second adjustment value is as follows: The volatility dispersion is compared with a preset dispersion threshold. If the volatility dispersion is greater than a preset dispersion threshold, the scaling factor is calculated as the product of the volatility dispersion and a fixed coefficient. If the volatility dispersion is less than or equal to a preset dispersion threshold, then the preset dispersion threshold sets the scaling factor to the default value of 1; The product of the first adjustment value and the scaling factor is used as the second adjustment value.

6. The investment optimization system based on artificial intelligence investment research according to claim 1, characterized in that: The generation process of the trend factor and resonance factor is as follows: The multi-frequency raw data stream is standardized and aligned to generate standardized market data; The standardized market data is subjected to signal decomposition processing to generate multiple intrinsic mode components of different frequencies; The trend factor is generated by superimposing the eigenmode components with the lowest frequency at a predetermined number of bits. The resonance factor is generated by superimposing the eigenmode components with the highest frequency at a predetermined number of bits.

7. The investment optimization system based on artificial intelligence investment research according to claim 1, characterized in that: The calculation process for the dynamic synergy coefficient is as follows: Obtain the trend factor sequence and the resonance factor sequence; The dynamic time warping algorithm is used to calculate the minimum warped path cumulative distance between two sequences; Based on the cumulative distance of the minimum regularized path, the dynamic coordination coefficient, which characterizes the morphological coordination of the two sequences, is calculated.

8. The investment optimization system based on artificial intelligence investment research according to claim 1, characterized in that: The process of obtaining the dynamic decision threshold is as follows: Determine whether the second adjustment value exceeds the preset upper and lower limits; If the second adjustment value is greater than the upper limit value, then the dynamic decision threshold is set to the upper limit value, and a first marker signal is generated; If the second adjustment value is less than the lower limit value, then the dynamic decision threshold is set to the lower limit value, and a second marker signal is generated; If the second adjustment value is within the range of the upper and lower limits, then the second adjustment value is directly used as the dynamic decision threshold; The first or second marker signal is used to trigger additional risk control operations on the portfolio adjustment instruction.

9. An investment optimization system based on artificial intelligence investment research according to claim 1, characterized in that: The process of generating the resonance state signal is as follows: At multiple consecutive time points, it is determined whether the dynamic coordination coefficient is continuously greater than the dynamic decision threshold; If the dynamic coordination coefficient is greater than the dynamic decision threshold in a predefined number of consecutive judgments, then a first type of resonance state signal is generated. If the number of consecutive judgments does not reach the predefined number, a second type of resonance state signal is generated; The first type of resonance state signal is used to trigger a standard portfolio adjustment instruction, and the second type of resonance state signal is used to trigger a tentative portfolio adjustment instruction or an instruction to maintain the current holdings.

10. An investment optimization system based on artificial intelligence investment research according to claim 1, characterized in that: Generating portfolio adjustment instructions based on the resonance state signal includes: Based on the resonance state signal, a preliminary portfolio adjustment instruction is generated; Determine whether a valid first or second marker signal exists; If a valid first or second marker signal exists, then according to the type of the marker signal, the corresponding additional risk control operation is triggered to modify the preliminary portfolio adjustment instruction and generate a final portfolio adjustment instruction. If no valid first and second marker signals exist, the preliminary portfolio adjustment instruction shall be used as the final portfolio adjustment instruction.

Images