Medium and long term adaptive transaction signal system and system based on large model

Abstract

The invention relates to the field of financial data, and discloses a medium-and-long-term adaptive transaction signal system based on a large model, which comprises the following steps: after obtaining target signal data, inquiring a hidden variable association graph of the target signal data, analyzing a structure association element and generating an evolution path, and by analyzing a path response sequence, identifying a sequence transition node and calculating a node chaos index, obtaining the target signal data; the method comprises the following steps: performing modal decomposition on target signal data to obtain a signal sub-data cluster, performing label coding to obtain a primary function label set, querying a sparse representation coefficient, marking a weight vector, coding the weight vector to a historical memory module, dynamically adjusting a hyper-parameter combination according to a convergence state, calculating a generalization performance ratio, generating a distribution characteristic key, and finally, performing data distribution on the basis of the distribution characteristic key. And constructing a self-adaptive transaction signal mechanism according to various identification elements in the secret key. According to the invention, the transaction signal adaptation accuracy can be improved.

Description

Technical Field

[0001] This invention relates to a medium- to long-term adaptive trading signal system based on a large model, belonging to the field of financial data. Background Technology

[0002] In the financial trading market, with the continuous development of the financial market and the accelerated advancement of financial innovation, the investment environment has become increasingly complex and volatile, affecting the formulation and optimization of medium- and long-term trading strategies, thus leading to higher standards for obtaining trading signals.

[0003] At present, traditional trading signal systems are inadequate in dealing with the complex and ever-changing financial market environment and the ever-increasing investment demands. Some trading information is delayed in different information interaction environments, which makes it impossible to capture market dynamics in a timely and accurate manner, resulting in financial losses for enterprises and investors. Therefore, a medium- to long-term adaptive trading signal system based on a large model is needed to improve the adaptability and accuracy of trading signals. Summary of the Invention

[0004] This invention provides a medium- to long-term adaptive trading signal system based on a large model, the main purpose of which is to improve the adaptive accuracy of trading signals.

[0005] To achieve the above objectives, this invention provides a medium- to long-term adaptive trading signal system based on a large model, comprising: Obtain target signal data, query the latent variable correlation graph corresponding to the target signal data, parse the structural correlation elements in the latent variable correlation graph, and generate the evolution path corresponding to the structural correlation elements; Analyze the path response sequence corresponding to the evolution path, identify the sequence transition nodes in the path response sequence, and calculate the node chaos index corresponding to the sequence transition node; Based on the node chaos index, the target signal data is decomposed into mode to obtain signal sub-data clusters. The signal sub-data clusters are then labeled to obtain a basis function label set. The sparse representation coefficients corresponding to the basis function label set are queried, and the representation weight vectors corresponding to the sparse representation coefficients are labeled. The representation weight vector is encoded into a preset historical memory module, and the convergence status of the data in the preset historical memory module is monitored. Based on the convergence status, the hyperparameter combination of the target signal data in the memory encoding process is dynamically adjusted, and the generalization efficiency ratio corresponding to the hyperparameter combination is calculated. Based on the generalization efficiency ratio, a distribution feature key corresponding to the target signal data is generated. The stable identifier, differentiable identifier, and invariant identifier in the distribution feature key are queried. Based on the stable identifier, differentiable identifier, and invariant identifier, an adaptive trading signal mechanism corresponding to the target signal data is constructed.

[0006] Optionally, generating the evolution path corresponding to the structural association element includes: Analyze the element connection relationships corresponding to the structural associated elements; Identify the connection evolution nodes in the element connection relationships; Map the associated element branches corresponding to the connection evolution nodes; The associated element branches are marked to obtain branch stage marks; By integrating the branch stage markers, the evolution path corresponding to the structural association elements is generated.

[0007] Optionally, the analysis of the path response sequence corresponding to the evolutionary path includes: Extract path change features from the evolutionary path; Based on the dynamic change characteristics, the path response stages corresponding to the evolution path are divided; Mark the stage marker points in the path response stage; Analyze the path response fragments corresponding to the stage markers; Based on the path response fragments, the path response sequences corresponding to the evolutionary path are analyzed.

[0008] Optionally, calculating the node chaos index corresponding to the sequence transition node includes: The node chaos index corresponding to the sequence transition node is calculated using the following formula: ; in, This represents the node chaos index corresponding to the sequence transition node. This indicates the number of nodes corresponding to the sequence transition nodes. This represents the index of the number of nodes corresponding to the sequence transition. Indicates the first The state variables corresponding to each sequence transition node and This represents the start and end times corresponding to the chaotic time interval of a node. Indicates the first The acceleration corresponding to the change of state variables at each sequence transition node Indicates the first The rate of change of the state variables at each sequence transition node. This represents the average speed value.

[0009] Optionally, the step of performing mode decomposition on the target signal data based on the node chaos index to obtain signal sub-data clusters includes: Divide the oscillation mode intervals corresponding to the chaotic index of the nodes; Extract chaotic signal segments within the oscillation mode range; The chaotic signal segment is frequency-domain aligned to obtain a signal alignment label; Based on the signal alignment label, the target signal data is subjected to mode decomposition to obtain signal sub-data clusters.

[0010] Optionally, the step of label encoding the signal sub-data clusters to obtain a basis function label set includes: Analyze the time-frequency components of the data corresponding to the signal sub-data clusters; Construct the correlation component matrix corresponding to the time-frequency components of the data; Filter the data vector groups in the relevant component matrix; The vector parameters in the data vector group are normalized to obtain a standard basis function set; The data in the standard basis function set is labeled to obtain the basis function label set.

[0011] Optionally, the representation weight vector corresponding to the sparse representation coefficients includes: Parse the index of the sparse characterization coefficients; Based on the characterization coefficient index, the weight activation region corresponding to the sparse characterization coefficient is divided; Locate the active weight components in the weight activation region; Extract the non-zero feature terms corresponding to the active weight components; Based on the non-zero feature terms, label the representation weight vector corresponding to the sparse representation coefficients.

[0012] Optionally, the step of dynamically adjusting the hyperparameter combination of the target signal data during the memory encoding process based on the convergence state includes: Extract the state feature parameters corresponding to the convergence state; Generate a parameter adjustment list corresponding to the state feature parameters; Based on the parameter adjustment list, the correction hyperparameters corresponding to the target signal data are calculated item by item; The modified hyperparameters are segmented and filtered to obtain valid parameter items; Based on the effective parameter items, the hyperparameter combination of the target signal data during the memory encoding process is dynamically adjusted.

[0013] Optionally, calculating the generalization efficiency ratio corresponding to the hyperparameter combination includes: The generalization efficiency ratio corresponding to the hyperparameter combination is calculated using the following formula: ; in, This represents the generalization performance ratio corresponding to the hyperparameter combination. This represents the total number of types included in the hyperparameter combination. This represents the category index corresponding to the hyperparameter combination. Indicates the first The weight coefficients corresponding to the hyperparameters Indicates the first Standardized values ​​of hyperparameters This indicates the time period corresponding to training the hyperparameter combination. This represents the parameter performance index corresponding to training the hyperparameter combination at time t. This represents the predicted sample corresponding to the hyperparameter combination. This represents the verification sample corresponding to the hyperparameter combination. This represents the correction factor.

[0014] Optionally, generating the distribution feature key corresponding to the target signal data based on the generalization efficiency ratio includes: Analyze the performance range threshold corresponding to the generalization performance ratio; Based on the performance interval threshold, extract the dynamic distribution entropy value corresponding to the target signal data; Based on the dynamic distribution entropy value, a dynamic key sequence corresponding to the target signal data is constructed; The dynamic key sequence is iteratively encrypted to obtain the core key seed; Based on the core key seed, a distribution feature key corresponding to the target signal data is generated.

[0015] Compared to the problems described in the background art, this invention, by acquiring target signal data and querying the latent variable correlation graph corresponding to the target signal data, uncovers hidden relationships between data points, establishing connections between previously isolated data points. This allows for the precise capture of subtle market dynamics, enhancing the depth of market understanding and laying a solid foundation for developing accurate and effective trading strategies. Furthermore, by analyzing the path response sequence corresponding to the evolution path, this invention can accurately grasp the intensity and speed of the market's response to changes in structural correlation elements at different stages. This helps investors keenly capture subtle changes in market sentiment and predict market trends in advance. Moreover, based on the node chaos index, this invention performs modal decomposition on the target signal data to obtain signal sub-data clusters, which helps to accurately uncover hidden information in the signal, providing a basis for subsequent signal processing. This invention provides a clear and targeted data foundation for signal processing, feature extraction, and prediction, improving analytical efficiency and accuracy. Furthermore, the invention encodes the representation weight vector into a preset historical memory module and monitors the convergence status of the data in this module. This allows for timely detection of data trend stability, facilitating accurate judgments based on stable and reliable historical data in signal processing and decision-making. It avoids misjudgments caused by data fluctuations, improving overall work efficiency. Finally, based on the generalization performance ratio, the invention generates a distribution feature key corresponding to the target signal data, accurately extracting key signal distribution characteristics and effectively characterizing the signal's distribution patterns in different scenarios. This helps in quickly identifying signal features during data storage, transmission, and analysis, ensuring data security and accuracy. Therefore, the medium-to-long-term adaptive trading signal system based on a large model provided by this invention can improve the adaptive accuracy of trading signals. Attached Figure Description

[0016] Figure 1 A flowchart illustrating a medium- to long-term adaptive trading signal system based on a large model, provided as an embodiment of the present invention; Figure 2 This is a schematic diagram of a module for implementing a medium-to-long-term adaptive trading signal system based on a large model, as provided in an embodiment of the present invention.

[0017] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0018] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0019] This application provides a medium- to long-term adaptive trading signal system based on a large model. The executing entity of this large-model-based medium- to long-term adaptive trading signal system includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the large-model-based medium- to long-term adaptive trading signal system can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster.

[0020] Example 1: Reference Figure 1 The diagram shown is a flowchart illustrating a medium-to-long-term adaptive trading signal system based on a large model, according to an embodiment of the present invention. In this embodiment, the medium-to-long-term adaptive trading signal system based on a large model includes: S1. Obtain target signal data, query the latent variable association graph corresponding to the target signal data, parse the structural association elements in the latent variable association graph, and generate the evolution path corresponding to the structural association elements.

[0021] This invention acquires target signal data, queries the latent variable correlation graph corresponding to the target signal data, and mines the hidden relationships between data points, establishing connections between originally isolated data points. This allows for the precise capture of subtle market dynamics, enhancing the depth of market understanding and laying a solid foundation for the subsequent formulation of accurate and effective trading strategies.

[0022] The target signal data refers to a wide range of key information collected from various channels in the financial trading market, including price trends, trading volume fluctuations, macroeconomic indicators, industry dynamics, and even investor sentiment on social media. This data directly or indirectly reflects the real-time state and changing trends of the market. The latent variable correlation graph is a graphical representation generated by leveraging the powerful data analysis and mining capabilities of large models. It depicts the potential variable relationships in the target signal data that are difficult to observe directly but are actually interconnected and influential. Nodes in the graph represent different latent variables, and edges represent the correlations between variables. The thickness and color of the edges can symbolize the strength and direction of the correlation. Optionally, querying the latent variable correlation graph corresponding to the target signal data can be achieved through visualization tools, such as using Graphviz, Gephi, etc., to parse data dependencies and render the topological structure, ultimately obtaining the latent variable correlation graph.

[0023] Furthermore, by analyzing the structural relationships in the latent variable correlation graph, this invention can clarify the connection methods, dependencies, and influence weights between different variables, and accurately locate the core factors that play a key role in market trends. This helps investors penetrate complex information and quickly grasp the root causes of dynamic changes in the market.

[0024] The structural association elements refer to the key components in the latent variable association graph used to present the relationship characteristics between variables. These include node characteristics, such as the attributes of the latent variables represented by the nodes in the financial market, like representing a certain type of risk indicator or a specific industry development factor; edge types, i.e., whether the relationship between variables is linear or non-linear, positive or negative (e.g., interest rates and bond prices are often negatively correlated); and edge weights, quantifying the strength of the relationship between variables, such as the degree of influence of the consumer confidence index on the price fluctuations of consumer stocks. Optionally, the analysis of the structural association elements in the latent variable association graph can be achieved using graph traversal algorithms, such as DFS or BFS, to extract key connections through node access paths, ultimately obtaining the structural association elements. Furthermore, by generating the evolution paths corresponding to the structurally related elements, this invention can clearly present the dynamic trajectory of key market factors on the timeline, helping investors to intuitively grasp market trends. By analyzing changes in path turning points and rates, investors can anticipate market inflection points and adjust their investment strategies in a timely manner.

[0025] The evolution path refers to the complete dynamic change trajectory of the structurally related elements in the time dimension after integrating all branch stage markers. It shows how each related element branch develops and changes from the initial state through a series of connected evolution nodes, and finally forms the current state.

[0026] As an embodiment of the present invention, generating the evolution path corresponding to the structurally related elements includes: parsing the element connection relationship corresponding to the structurally related elements; identifying the connection evolution nodes in the element connection relationship; mapping the related element branches corresponding to the connection evolution nodes; marking the related element branches to obtain branch stage markers; and integrating the branch stage markers to generate the evolution path corresponding to the structurally related elements.

[0027] In this context, the element connectivity relationship refers to the way different structurally related elements are associated in a latent variable association graph. This relationship is represented by edges in the graph, where each edge represents a specific connection between two structurally related elements. The attributes of the edges (such as direction and weight) determine the direction and degree of influence between the elements. The connection evolution node refers to the key position where a significant change in the state of the element connectivity relationship occurs over time. For example, in the financial market, when there is a major adjustment in macroeconomic policies, the original correlation between certain industries and interest rates will change at a certain point in time. The position corresponding to this point in time is the connection evolution node. The term "point" refers to a branch of related elements, which is a subset of structural elements and their connections that extend from a connection evolution node and are associated with that node. Taking the stock market as an example, if a major asset restructuring event of a stock is considered a connection evolution node, then the changes in the upstream and downstream companies related to the company and their business relationships around this event form different branches of related elements. The term "branch stage marker" refers to the characteristic identification of the related element branches at different development stages. For example, "A" indicates that the related element branch is in the initial formation stage and the connection is not yet stable; "B" indicates that the branch has entered a mature and stable connection stage.

[0028] Furthermore, the parsing of the element connection relationships corresponding to the structurally related elements can be achieved through graph relationship mining algorithms, such as gSpan and FSG, which extract high-frequency connection structures through pattern matching to ultimately obtain the element connection relationships; the identification of connection evolution nodes in the element connection relationships can be achieved through time-series graph analysis tools, such as capturing node connection changes through a sliding window to identify connection evolution nodes; the mapping of the associated element branches corresponding to the connection evolution nodes can be achieved through graph partitioning tools, such as using METIS and generating associated element branches through the minimum cut criterion; the labeling of the associated element branches can be achieved through graph neural networks, such as GAT and GCN, which automatically generate semantic labels by learning branch features; the generation of the evolution path corresponding to the structurally related elements can be achieved through time-series path reconstruction methods, such as generating evolution paths by connecting key nodes in chronological order.

[0029] S2. Analyze the path response sequence corresponding to the evolution path, identify the sequence transition nodes in the path response sequence, and calculate the node chaos index corresponding to the sequence transition node.

[0030] By analyzing the path response sequence corresponding to the evolution path, this invention can accurately grasp the intensity and speed of the market's response to changes in structurally related elements at different stages. This helps investors to keenly capture subtle changes in market sentiment and predict market trends in advance.

[0031] The path response sequence refers to an ordered set composed of all path response segments arranged in chronological order. For example, by studying the path response sequence of a certain index over the past few years, the response characteristics of the market under different economic cycles can be analyzed, providing a reference for investment decisions in the current market environment.

[0032] As an embodiment of the present invention, the analysis of the path response sequence corresponding to the evolution path includes: extracting path change features in the evolution path; dividing the path response stages corresponding to the evolution path based on the dynamic change features; marking stage markers in the path response stages; analyzing the path response segments corresponding to the stage markers; and analyzing the path response sequence corresponding to the evolution path based on the path response segments.

[0033] The path change characteristics refer to various attributes that can reflect the dynamic changes in the evolution path, such as the speed at which stock prices rise or fall over a period of time; directional changes, such as from an upward trend to a downward trend, which means a reversal of market conditions; and fluctuation range, that is, the degree to which the path fluctuates around a certain benchmark, such as the price oscillation range of commodity prices within a specific period. The path response stage refers to different time periods divided based on the path change characteristics. The evolution path exhibits a relatively consistent change pattern in each time period, which is the path response stage. For example, in the financial market, when the market is in a one-sided upward trend, the evolution path is characterized by a continuously rising slope and a stable direction, which can be divided into a bull market path response phase. When the market enters a consolidation period, the path change is characterized by smaller fluctuations and frequent minor directional adjustments, which constitutes a consolidation path response phase. The stage markers refer to key time points or data points with special significance in the path response phase. These points usually mark the beginning, end, or important turning points of the path response phase. In the stock market, the point in time when a stock changes from a long-term downtrend to a sideways consolidation is a stage marker. The path response segment refers to the part of the evolution path between two stage markers, which contains complete path change information within that time period. Each path response segment corresponds to the specific performance of the market in a specific phase. For example, on the daily chart of a stock, the price trend curve between the appearance of a limit-up (stage marker) and the next new high (another stage marker) is a path response segment.

[0034] Furthermore, the extraction of path change features in the evolutionary path can be achieved using feature engineering tools, such as tsfresh and Tsfel, which automatically extract statistical features (such as slope and volatility) to obtain path change features; the division of the path response stages corresponding to the evolutionary path can be achieved using stage detection models, such as HMM models, which automatically divide the path response stages by learning state transition probabilities; the marking of stage markers in the path response stages can be achieved using rule engines, such as the Drools engine, which marks stage boundaries based on predefined rules to obtain stage markers; the analysis of path response fragments corresponding to the stage markers can be achieved using similarity algorithms, such as DTW, which analyze response features by matching fragment patterns to obtain path response fragments; and the analysis of the path response sequence corresponding to the evolutionary path can be achieved using time series prediction models, such as LSTM and Transformer models, which reconstruct the response sequence by learning sequence dependencies to obtain the path response sequence.

[0035] This invention can accurately capture sudden market changes by identifying sequence transition nodes in the path response sequence. It can also help investors quickly perceive market trend reversals or accelerations, adjust trading strategies in a timely manner, and make early arrangements to avoid risks during market changes.

[0036] The sequence transition node refers to a key position in the path response sequence where a significant, discontinuous change occurs. In the context of financial markets, it manifests as a sharp change in market conditions. For example, a stock price that was originally rising steadily may suddenly start to fall sharply at a certain moment due to a major policy announcement or negative news about a company. This point in time when the stock price trend suddenly changes is the sequence transition node. Optionally, the identification of the sequence transition node in the path response sequence can be achieved through a mutation point detection algorithm, such as CUSUM, Bayesian change point detection, etc. The transition node is located by analyzing the sudden changes in the statistical characteristics of the sequence, and finally the sequence transition node is obtained.

[0037] Furthermore, by calculating the node chaos index corresponding to the sequence transition node, this invention accurately measures the degree of market disorder at key turning points. The numerical data intuitively demonstrates the disorder and complexity of market changes, clearly indicating the risk level during market upheavals, and thus allowing for flexible adjustments to trading strategies.

[0038] The node chaos index is used to measure the degree of disorder or chaos in the market or system at the sequence transition node. By comprehensively considering factors such as the number of nodes, the acceleration of change of state variables, and the rate of change, it can help understand the level of complexity and uncertainty at key turning points.

[0039] As an embodiment of the present invention, calculating the node chaos index corresponding to the sequence transition node includes: The node chaos index corresponding to the sequence transition node is calculated using the following formula: ; in, This represents the node chaos index corresponding to the sequence transition node. This indicates the number of nodes corresponding to the sequence transition nodes. This represents the index of the number of nodes corresponding to the sequence transition. Indicates the first The state variables corresponding to each sequence transition node and This represents the start and end times corresponding to the chaotic time interval of a node. Indicates the first The acceleration corresponding to the change of state variables at each sequence transition node Indicates the first The rate of change of the state variables at each sequence transition node. This represents the average speed value.

[0040] Specifically, the state variable refers to a variable describing the characteristics of a sequence transition node. It represents a certain attribute or state of the system or market at that node. For example, in the financial market, it could be stock price, trading volume, or rate of return; in the physical system, it could be temperature, pressure, etc. Different fields have different specific meanings, used to characterize the actual situation at the node. The node chaos time interval refers to a time range selected around the sequence transition node, which defines the time boundary for analyzing the changing characteristics of the state variable. The acceleration of change refers to the rate of change of the state variable at the sequence transition node. For example, in stock price changes, if the change... A positive and large acceleration indicates that the speed of stock price increase or decrease is accelerating; a negative acceleration indicates that the rate of change is decreasing, reflecting the dynamic trend characteristics of the state variable change. The rate of change value refers to how quickly the state variable changes over time at the sequence transition node. It measures the instantaneous rate of change of the state variable at a certain moment, which can be understood as the instantaneous rise and fall speed of asset prices in the financial market. The average rate value is the value obtained by averaging the rate of change values ​​of the state variable at multiple sequence transition nodes. It comprehensively reflects the average level of the rate of change of the state variable within the research scope, and plays a role in smoothing the data and providing an overall reference benchmark for the rate of change.

[0041] Furthermore, in the numerator of the above formula The absolute values ​​of the acceleration of the state variables at each node are accumulated to reflect the changes in the rate of change of the state variables. The greater the acceleration, the more drastic the change of the state variables. This integral term incorporates the rate of change. For time intervals The integral value is measured by the combined effect of the rate of change of the internal state variables. The greater the rate of change, the smaller the relative integral value. The combination of these two factors reflects the dynamic characteristics of the state variables at the nodes. The denominator of the formula... In this context, n represents the number of nodes. It is the average rate value, which plays a normalization role, eliminating the influence of the number of nodes and the average rate of change on the results. This allows the chaos index to measure the degree of chaos of sequence transition nodes in different systems or time periods under a unified standard. Finally, through such calculation, the node chaos index CI can quantify the degree of disorder and chaos at the sequence transition nodes.

[0042] S3. Based on the node chaos index, perform mode decomposition on the target signal data to obtain signal sub-data clusters, perform label encoding on the signal sub-data clusters to obtain a basis function label set, query the sparse representation coefficients corresponding to the basis function label set, and label the representation weight vectors corresponding to the sparse representation coefficients.

[0043] Based on the node chaos index, this invention performs mode decomposition on the target signal data to obtain signal sub-data clusters, which helps to accurately mine hidden information in the signal and provides a clear and targeted data foundation for subsequent signal processing, feature extraction and prediction, thereby improving analysis efficiency and accuracy.

[0044] The signal sub-data cluster refers to multiple data sets obtained after modal decomposition of the target signal data based on signal alignment labels. Each signal sub-data cluster contains signal parts with similar oscillation modes and chaotic characteristics, realizing the orderly division of the original complex target signal.

[0045] As an embodiment of the present invention, the step of performing mode decomposition on the target signal data based on the node chaos index to obtain signal sub-data clusters includes: dividing the oscillation mode interval corresponding to the node chaos index; extracting chaotic signal segments within the oscillation mode interval; performing frequency domain alignment on the chaotic signal segments to obtain signal alignment labels; and performing mode decomposition on the target signal data based on the signal alignment labels to obtain signal sub-data clusters.

[0046] The oscillation mode interval refers to an interval with specific oscillation characteristics defined by the node chaos index. Within this interval, the signal oscillation exhibits a relatively consistent pattern, such as similar oscillation frequency and amplitude. This interval is used to define the scope of subsequent signal analysis and serves as the basic unit for classifying complex signals based on chaotic characteristics. The chaotic signal segment refers to the signal portion extracted from the oscillation mode interval. These segments contain the signal characteristics of the system in a chaotic state, reflecting the irregular and complex changes of the signal within this interval. They are the raw data material for subsequent frequency domain analysis and mode decomposition. The signal alignment label refers to the identifier obtained after frequency domain alignment processing of the chaotic signal segments. It records the correspondence and adjustment parameters of different chaotic signal segments in the frequency domain, ensuring that each signal segment can be processed according to a unified frequency reference during mode decomposition. This is a key identifier for achieving accurate mode decomposition.

[0047] Furthermore, the division of the oscillation mode intervals corresponding to the chaotic index of the nodes can be achieved using recursive graph analysis tools, such as PyRQA, which divides the chaotic state intervals by calculating recursive quantitative indices; the extraction of chaotic signal segments within the oscillation mode intervals can be achieved using empirical mode decomposition, such as extracting chaotic feature segments through intrinsic mode functions to finally obtain chaotic signal segments; the frequency domain alignment of the chaotic signal segments can be achieved using time-frequency analysis methods, such as wavelet coherence analysis, which achieves frequency band alignment through time-frequency energy matching to finally obtain signal alignment labels; the mode decomposition of the target signal data can be achieved using multi-scale decomposition algorithms, such as singular spectrum analysis, which reconstructs sub-signals using trajectory matrix eigenvalue decomposition to finally obtain signal sub-data clusters.

[0048] This invention obtains a basis function label set by label encoding the signal sub-data clusters, which helps simplify signal analysis and provides a clear foundation for signal reconstruction. This facilitates the accurate recovery of the original signal using these orthogonal bases, thereby improving the accuracy and reliability of signal processing.

[0049] The basis function label set refers to the result of label encoding the data in the standard basis function set. The encoding process assigns a unique identifier label to each standard basis function, which facilitates the identification, management and invocation of different basis functions.

[0050] As an embodiment of the present invention, the step of labeling the signal sub-data cluster to obtain a basis function label set includes: analyzing the time-frequency components of the data corresponding to the signal sub-data cluster; constructing the correlation component matrix corresponding to the time-frequency components of the data; filtering the data vector groups in the correlation component matrix; normalizing the vector parameters in the data vector groups to obtain a standard basis function set; and labeling the data in the standard basis function set to obtain a basis function label set.

[0051] The time-frequency components of the data refer to the components of the signal sub-data cluster in the time and frequency dimensions. They reflect the frequency information contained in the signal at different times. For example, in financial market trading signals, different time-frequency components can correspond to price fluctuations or trading volume change patterns of different periods. The correlation component matrix is ​​a matrix constructed based on the time-frequency components of the data. The elements in the matrix reflect the correlation relationship between the various time-frequency components, with rows and columns corresponding to different time-frequency components. The data vector group is a set of vectors selected from the correlation component matrix. These vectors represent combinations of time-frequency components with specific correlation characteristics. Each vector contains relevant information of multiple time-frequency components. By studying these vector groups, we can further focus on the key correlation features in the signal. The standard basis function set is a set of functions obtained after normalizing the vector parameters in the data vector group. In this set, each basis function is obtained by processing the original data vector group. They are independent of each other and can effectively characterize the features of the signal sub-data cluster. They can be used as basic units for constructing signal models or performing signal decomposition and reconstruction.

[0052] Furthermore, the analysis of the time-frequency components corresponding to the signal sub-data clusters can be achieved through time-frequency analysis algorithms, such as short-time Fourier transform and continuous wavelet transform, to analyze signal features through joint time-frequency distribution and ultimately obtain the time-frequency components. The construction of the correlation matrix corresponding to the time-frequency components can be achieved through matrix operation tools, such as the cov function in NumPy, to construct the matrix by quantifying the correlation between components and ultimately obtain the correlation matrix. The selection of data vector groups in the correlation matrix can be achieved through feature selection methods, such as principal component analysis, to extract key feature vectors through dimensionality reduction and ultimately obtain the data vector group. The normalization of vector parameters in the data vector group can be achieved through normalization algorithms, such as Z-score and Min-Max algorithms, to eliminate parameter bias by unifying dimensions and ultimately obtain the standard basis function set. The label encoding of the data in the standard basis function set can be achieved through classification encoding methods, such as one-hot encoding and label encoding.

[0053] This invention can accurately extract key features of a signal by querying the sparse representation coefficients corresponding to the basis function label set. It uses a small number of non-zero coefficients to represent the main information of the signal, which greatly reduces data redundancy. This not only reduces storage and computing costs, but also helps to efficiently identify important patterns in the signal and improve the overall efficiency of signal processing and analysis.

[0054] The sparse representation coefficients refer to a set of coefficients used to describe the representation of a signal under a specific basis function label set. For example, in image compression, the image signal is decomposed by basis functions (such as wavelet basis), and the sparse representation coefficients can indicate the key image features. By retaining a small number of non-zero coefficients and their corresponding basis functions, the key information of the image can be preserved while significantly reducing the amount of data. In the scenario of financial trading signals, the signal components that have a core impact on market trends can be highlighted. Optionally, the query of the sparse representation coefficients corresponding to the basis function label set can be implemented by sparse coding algorithms, such as K-SVD, OMP orthogonal matching pursuit, etc.

[0055] This invention, by labeling the representation weight vectors corresponding to the sparse representation coefficients, can quickly locate the sparse coefficients that play a dominant role in signal description, clearly distinguish the differences in the influence of different signal components, and help focus on core information in signal processing, thereby improving the accuracy of signal analysis.

[0056] The representation weight vector refers to a vector obtained by labeling non-zero feature terms. Each element in the vector corresponds to the weight of a sparse representation coefficient, reflecting the importance of the coefficient in describing signal features. Through the representation weight vector, the contribution of each sparse representation coefficient to the overall signal features can be seen intuitively.

[0057] As an embodiment of the present invention, the step of marking the representation weight vector corresponding to the sparse representation coefficient includes: parsing the representation coefficient index corresponding to the sparse representation coefficient; dividing the weight activation region corresponding to the sparse representation coefficient based on the representation coefficient index; locating the active weight component in the weight activation region; extracting the non-zero feature terms corresponding to the active weight component; and marking the representation weight vector corresponding to the sparse representation coefficient based on the non-zero feature terms.

[0058] The representation coefficient index refers to a unique identifier assigned to each sparse representation coefficient, similar to an "address" for data. This index allows for accurate location and differentiation of different sparse representation coefficients. In signal processing, a signal is decomposed into a series of sparse representation coefficients, each representing the projection of the signal onto a specific basis function. The weighted activation region refers to the region obtained after dividing the sparse representation coefficients based on the representation coefficient index. For example, in image processing, coefficients in some regions are related to edge features, while others are related to texture features. The active weight components refer to the components in the weighted activation region that have a significant impact on signal features. These components correspond to larger sparse representation coefficients, indicating that they play a crucial role in describing signal features and dominate signal construction and feature expression. By locating these active weight components, the core features of the signal can be extracted. The non-zero feature terms refer to the non-zero feature parts extracted from the active weight components. These non-zero feature terms are the carriers of meaningful information, representing the significant features of the signal under a specific basis function and are a concentrated manifestation of the signal features.

[0059] Furthermore, parsing the representation coefficient index corresponding to the sparse representation coefficient can be achieved using sparse matrix analysis tools, such as sparse matrix compressed storage formats (e.g., CSR / CSC format). Key coefficient positions are extracted through non-zero element indices to ultimately obtain the representation coefficient index. Dividing the weight activation region corresponding to the sparse representation coefficient can be achieved using threshold segmentation methods, such as the Otsu algorithm. An adaptive threshold is used to determine the activation boundary, ultimately obtaining the weight activation region. Locating the active weight components within the weight activation region can be achieved using saliency analysis methods, such as the SHAP value calculation method. The importance of each component is evaluated, ultimately obtaining the active weight components. Extracting the non-zero feature terms corresponding to the active weight components can be achieved using sparse feature selection methods, such as using NumPy's nonzero function to directly extract effective feature terms. Labeling the representation weight vector corresponding to the sparse representation coefficient can be achieved using visualization tools, such as Matplotlib's annotate tool, which interactively annotates the weight distribution, ultimately obtaining the representation weight vector.

[0060] S4. Encode the representation weight vector into a preset historical memory module, monitor the convergence status of the data in the preset historical memory module, dynamically adjust the hyperparameter combination of the target signal data in the memory encoding process based on the convergence status, and calculate the generalization efficiency ratio corresponding to the hyperparameter combination.

[0061] This invention encodes the representation weight vector into a preset historical memory module and monitors the convergence status of the data in the preset historical memory module. It can detect whether the data change trend is stable in a timely manner, which helps to make accurate judgments based on stable and reliable historical data in signal processing, decision making and other work, avoid misjudgments caused by data fluctuations, and improve the overall work efficiency.

[0062] The preset historical memory module refers to a pre-defined data storage and management unit specifically designed to retain and organize past data information related to a specific task or analysis. It functions like a meticulously constructed "data archive," possessing a specific data structure and storage rules. In the field of signal processing, this module can store signal characteristic data from different time periods, such as the previously mentioned representation weight vector. The convergence state, within the preset historical memory module, describes a situation where data gradually stabilizes over time or during processing. For example, when using the historical memory module to predict market trends, if, with the continuous input of new data, the data characteristics and output results relied upon by the prediction model no longer fluctuate significantly but gradually stabilize within a small range of change, this indicates that the data has reached a convergence state. Optionally, encoding the representation weight vector into the preset historical memory module can be achieved through a memory network architecture. Monitoring the convergence state of data in the preset historical memory module can be achieved through dynamic evaluation methods, such as sliding window statistics or the KL divergence detection algorithm.

[0063] Furthermore, based on the convergence state, the present invention dynamically adjusts the hyperparameter combination of the target signal data in the memory encoding process, which can improve the accuracy of the data in memory encoding, reduce information loss, enhance the system's adaptability to different data modes, and ensure the continuous and efficient operation of memory encoding.

[0064] The hyperparameter combination refers to a set of multiple hyperparameters that are used together in the memory coding process. By dynamically adjusting the hyperparameter combination based on the effective parameter items, the memory coding process can better adapt to the convergence state of the target signal data.

[0065] As an embodiment of the present invention, the step of dynamically adjusting the hyperparameter combination of the target signal data in the memory encoding process based on the convergence state includes: extracting state feature parameters corresponding to the convergence state; generating a parameter adjustment list corresponding to the state feature parameters; calculating the corrected hyperparameters corresponding to the target signal data item by item based on the parameter adjustment list; performing segmented filtering on the corrected hyperparameters to obtain effective parameter items; and dynamically adjusting the hyperparameter combination of the target signal data in the memory encoding process based on the effective parameter items.

[0066] The state characteristic parameters refer to a set of parameters used to describe the specific characteristics of the convergence state. The convergence state reflects the stability and trend of the target signal data during the memory encoding process, while the state characteristic parameters are a quantitative representation of these conditions. For example, they may include indicators such as data fluctuation amplitude, rate of change, and correlation. The parameter adjustment list refers to a list generated based on the state characteristic parameters that contains a series of parameter adjustment suggestions. It clarifies which hyperparameters need to be adjusted and how to adjust them for the current convergence state. Each item in the list corresponds to the adjustment direction (increase or decrease) and adjustment magnitude of a hyperparameter. The modified hyperparameters refer to the new hyperparameters obtained by calculating and modifying the original hyperparameters of the target signal data based on the parameter adjustment list. For example, if the parameter adjustment list suggests increasing a certain hyperparameter by a certain percentage, then the original hyperparameter will be increased according to this suggestion to obtain the modified hyperparameters. The effective parameter items refer to the parameter items that meet specific conditions after segmenting and filtering the modified hyperparameters. During the filtering process, modified hyperparameters that do not meet the requirements will be excluded according to some preset rules or standards, such as the range of parameter values ​​and the degree of influence on the coding effect, and only those parameter items that can effectively improve the memory coding effect will be retained.

[0067] Furthermore, the extraction of state feature parameters corresponding to the convergence state can be achieved using time-series feature extraction tools, such as the minimal parameter set in the tsfresh library. Key features are obtained by automatically calculating statistics (mean, variance, etc.), ultimately yielding the state feature parameters. The generation of the parameter adjustment list corresponding to the state feature parameters can be achieved using a Bayesian optimizer, such as HyperOpt, which recommends the optimal parameter adjustment direction based on a probability model, ultimately yielding the parameter adjustment list. The item-by-item calculation of the corrected hyperparameters corresponding to the target signal data can be achieved using parameter optimization methods, such as meta-learning frameworks, which quickly adapt to new signal features using a small number of samples, ultimately yielding the corrected hyperparameters. The segmented screening of the corrected hyperparameters can be achieved using significance testing methods, such as t-tests, which screen statistically significant parameter items, ultimately yielding effective parameter items. The dynamic adjustment of the hyperparameter combination of the target signal data during the memory encoding process can be achieved using dynamic memory networks, such as DMN, which adaptively adjust the encoding strength through a gating mechanism, ultimately yielding the hyperparameter combination.

[0068] This invention calculates the generalization efficiency ratio corresponding to the hyperparameter combination, which can clearly compare the impact of different hyperparameter settings on the model's generality, and quickly determine which combination can enable the model to break out of the limitations of specific training data and be accurately applied in new data.

[0069] The generalization efficiency ratio refers to a quantitative indicator that measures the general performance of hyperparameter combinations in different data scenarios. It comprehensively considers factors such as the characteristics of hyperparameters, training process and sample conditions. The higher the value, the better the model performs on new data, that is, the stronger the model's generalization ability, the better it can adapt to different datasets, and the lower the risk of overfitting or underfitting.

[0070] As an embodiment of the present invention, calculating the generalization efficiency ratio corresponding to the hyperparameter combination includes: The generalization efficiency ratio corresponding to the hyperparameter combination is calculated using the following formula: ; in, This represents the generalization performance ratio corresponding to the hyperparameter combination. This represents the total number of types included in the hyperparameter combination. This represents the category index corresponding to the hyperparameter combination. Indicates the first The weight coefficients corresponding to the hyperparameters Indicates the first Standardized values ​​of hyperparameters This indicates the time period corresponding to training the hyperparameter combination. This represents the parameter performance index corresponding to training the hyperparameter combination at time t. This represents the predicted sample corresponding to the hyperparameter combination. This represents the verification sample corresponding to the hyperparameter combination. This represents the correction factor.

[0071] In detail, the weight coefficient refers to the value reflecting the relative importance of the k-th hyperparameter in the hyperparameter combination. Different hyperparameters have varying degrees of influence on model performance, and hyperparameters with large weight coefficients play a key role in determining the generalization efficiency ratio. Adjusting the weight coefficients can reflect the different influences of each hyperparameter. The standardized value refers to the value obtained after standardizing the k-th hyperparameter. Standardization can eliminate the influence of different units on the original hyperparameter values, making different hyperparameters on the same comparable scale, which facilitates the calculation and analysis of the contribution of hyperparameters to the generalization efficiency ratio under a unified standard. The time period refers to the duration of training the hyperparameter combination. During this period, the model continuously adjusts the hyperparameter combination to optimize performance by learning from the data. The parameter performance index refers to the index used to measure the model performance at time t when training the hyperparameter combination. For example, in machine learning, it can be accuracy, etc. Recall rate, mean squared error, etc.; the predicted samples refer to the sample set used to test the performance of the hyperparameter combination in the model prediction task. By comparing the model's prediction results on these samples with the actual values, the predictive ability of the model due to the hyperparameter combination can be evaluated, which is a key data for measuring the model's prediction performance in the generalization efficiency ratio; the validation samples refer to the sample set used to verify the effect of the hyperparameter combination during the model training process. They are independent of the training samples and their main function is to prevent the model from overfitting. The hyperparameter combination is evaluated and adjusted during the training process, and it is an important reference for evaluating the model's generalization performance in the calculation of the generalization efficiency ratio; the correction factor refers to the adjustment parameter introduced to make the calculation of the generalization efficiency ratio more reasonable and accurate. It can correct the calculation results according to specific application scenarios, data characteristics, and other factors, balance the weights of different parts in the formula, and make the generalization efficiency ratio more accurately reflect the generalization performance of the hyperparameter combination.

[0072] S5. Based on the generalization efficiency ratio, generate the distribution feature key corresponding to the target signal data, query the stable identifier, differentiable identifier, and invariant identifier in the distribution feature key, and construct the adaptive trading signal mechanism corresponding to the target signal data based on the stable identifier, differentiable identifier, and invariant identifier.

[0073] Based on the generalization efficiency ratio, this invention generates a distribution feature key corresponding to the target signal data, which can accurately extract the key distribution characteristics of the signal and effectively characterize the distribution pattern of the signal in different scenarios. This helps to quickly identify signal features in data storage, transmission and analysis, and ensure data security and accuracy.

[0074] The distribution feature key refers to the key generated based on the core key seed to characterize the distribution features of the target signal data. It is the final product of the entire process and condenses information such as the distribution characteristics and generalization efficiency of the target signal data. The distribution features of the target signal data can be quickly and accurately identified using this key.

[0075] As an embodiment of the present invention, generating the distribution feature key corresponding to the target signal data based on the generalization performance ratio includes: analyzing the performance interval threshold corresponding to the generalization performance ratio; extracting the dynamic distribution entropy value corresponding to the target signal data according to the performance interval threshold; constructing a dynamic key sequence corresponding to the target signal data based on the dynamic distribution entropy value; iteratively encrypting the dynamic key sequence to obtain a core key seed; and generating the distribution feature key corresponding to the target signal data based on the core key seed.

[0076] The performance interval threshold refers to several interval boundary values ​​divided according to the generalization performance ratio. By determining the performance interval threshold, the generalization performance ratio can be classified and evaluated, providing a basis for subsequent extraction of relevant features of the target signal data and determining the performance state of the hyperparameter combination. The dynamic distribution entropy value refers to the value extracted from the target signal data based on the performance interval threshold, reflecting the uncertainty of the data distribution. The larger the entropy value, the more complex and disordered the distribution of the target signal data; the smaller the entropy value, the simpler and more ordered the distribution. The dynamic key sequence refers to a series of ordered key element combinations constructed based on the dynamic distribution entropy value. The elements in this sequence are closely related to the distribution characteristics of the target signal data and will be dynamically adjusted according to the changes in the dynamic distribution entropy value. It is the basis for generating the final distribution feature key and carries the distribution information of the target signal data. The core key seed refers to the key starting element obtained after iterative encryption of the dynamic key sequence. The iterative encryption process continuously strengthens the security and uniqueness of the key. As the core starting point for generating the final distribution feature key, the core key seed contains key information after multiple encryption processes and is the core hub of the entire key generation process.

[0077] Furthermore, the analysis of the performance interval threshold corresponding to the generalization performance ratio can be achieved using kernel density estimation tools, such as the KDEpy library, which automatically identifies performance inflection points through probability density function fitting, ultimately obtaining the performance interval threshold; the extraction of the dynamic distribution entropy value corresponding to the target signal data can be achieved using sample entropy algorithms, such as quantizing signal complexity through pattern repetition rate, ultimately obtaining the dynamic distribution entropy value; the construction of the dynamic key sequence corresponding to the target signal data can be achieved using SHA-3 hash chain, such as generating a key stream through iterative hashing of signal feature values, ultimately obtaining the dynamic key sequence; the iterative encryption of the dynamic key sequence can be achieved using lattice cryptography algorithms, such as the NTRUEncrypt algorithm, which constructs a quantum-resistant encryption layer based on mathematical problems, ultimately obtaining the core key seed; the generation of the distribution feature key corresponding to the target signal data can be achieved using feature fusion encryption methods, such as generating a unique key based on signal feature negotiation, ultimately obtaining the distribution feature key.

[0078] This invention, by querying the stable identifier, differentiable identifier, and invariant identifier in the distributed feature key, can comprehensively grasp the signal characteristics by identifying these identifiers, providing accurate insights for data processing, analysis, and signal-based decision-making, and helping to optimize various signal-related applications and operations.

[0079] Among them, the stable identifier refers to the relatively fixed and unchanging feature information in the target signal data in the distributed feature key. It reflects the constant property of the signal over a long time scale or under different conditions, similar to the "inherent label" of the signal. For example, in the trading signals of the financial market, certain basic trading rules or market mechanism-related features exist in the distributed feature key in the form of stable identifiers, helping to identify the stable and reliable parts of the signal. The differentiable identifier refers to the feature information in the target signal data that has a changing trend and can be finely adjusted and analyzed. In signal processing, differentiable identifiers can help track the dynamic evolution of the signal and understand its changing patterns. For example, in speech signals, features such as pitch and volume that change continuously over time can be represented by differentiable identifiers. The invariant identifier refers to those essential attribute identifiers in the distributed feature key that remain unchanged no matter how the signal changes. In image signals, the basic geometric structure of an image (such as the aspect ratio of a rectangular image) may be invariant identifiers. These are unique and constant features that distinguish a signal from other signals, helping to identify and differentiate different target signal data from their essence. Optionally, querying stable identifiers, differentiable identifiers, and invariant identifiers in the distributed feature key can be achieved through deep learning models, such as convolutional neural networks (CNNs). By training the model, the feature patterns in the target signal data can be automatically learned, thereby extracting stable identifiers, differentiable identifiers, and invariant identifiers.

[0080] Furthermore, based on the stable identifier, the differentiable identifier, and the invariant identifier, this invention constructs an adaptive trading signal mechanism corresponding to the target signal data. The stable identifier reveals long-term market patterns, ensuring a solid foundation for the trading strategy; the differentiable identifier captures dynamic market changes, helping to adjust trading behavior in a timely manner; and the invariant identifier grasps the essential characteristics of the market, preventing trading from deviating from the core logic. The three work together to allow the trading mechanism to flexibly adapt to the market, effectively improving the success rate and profitability of trading.

[0081] The adaptive trading signal mechanism refers to an intelligent trading decision generation system that dynamically adjusts trading strategies based on market changes. This mechanism is based on stable identifiers that reflect long-term, constant market characteristics, such as cyclical patterns in certain industries or the long-term impact of macroeconomics on specific markets, providing a stable foundation for trading strategies and ensuring that trading directions align with major market trends. Differentiable identifiers are used to track real-time market changes, such as short-term volatility in asset prices and instantaneous changes in trading volume. Through continuous monitoring and analysis of these characteristics, the mechanism can keenly perceive market dynamics and promptly make fine adjustments to trading parameters such as buying and selling opportunities and position sizing. The variable identifier represents the essential attributes of the market that will not change due to short-term fluctuations, such as the basic supply and demand relationship and transaction cost structure. This ensures that trading decisions are always anchored to the core logic of the market and avoids making incorrect judgments due to market noise. In actual trading, when the market environment is stable, the mechanism executes the predetermined strategy based on stable and unchanging identifiers. Once the market fluctuates, the variable identifier prompts the mechanism to respond quickly, adjust the trading details, and achieve precise adaptation between the trading strategy and the real-time market state to improve trading efficiency and returns. Optionally, the adaptive trading signal mechanism corresponding to the target signal data can be implemented using mechanism construction tools such as Python and TensorFlow.

[0082] Compared to the problems described in the background art, this invention, by acquiring target signal data and querying the latent variable correlation graph corresponding to the target signal data, uncovers hidden relationships between data points, establishing connections between previously isolated data points. This allows for the precise capture of subtle market dynamics, enhancing the depth of market understanding and laying a solid foundation for developing accurate and effective trading strategies. Furthermore, by analyzing the path response sequence corresponding to the evolution path, this invention can accurately grasp the intensity and speed of the market's response to changes in structural correlation elements at different stages. This helps investors keenly capture subtle changes in market sentiment and predict market trends in advance. Moreover, based on the node chaos index, this invention performs modal decomposition on the target signal data to obtain signal sub-data clusters, which helps to accurately uncover hidden information in the signal, providing a basis for subsequent signal processing. This invention provides a clear and targeted data foundation for signal processing, feature extraction, and prediction, improving analytical efficiency and accuracy. Furthermore, the invention encodes the representation weight vector into a preset historical memory module and monitors the convergence status of the data in this module. This allows for timely detection of data trend stability, facilitating accurate judgments based on stable and reliable historical data in signal processing and decision-making. It avoids misjudgments caused by data fluctuations, improving overall work efficiency. Finally, based on the generalization performance ratio, the invention generates a distribution feature key corresponding to the target signal data, accurately extracting key signal distribution characteristics and effectively characterizing the signal's distribution patterns in different scenarios. This helps in quickly identifying signal features during data storage, transmission, and analysis, ensuring data security and accuracy. Therefore, the medium-to-long-term adaptive trading signal system based on a large model provided by this invention can improve the adaptive accuracy of trading signals.

[0083] Example 2: like Figure 2 The diagram shown is a functional block diagram of a medium-to-long-term adaptive trading signal system based on a large model according to the present invention.

[0084] The medium-to-long-term adaptive trading signal system 200 based on a large model described in this invention can be installed in an electronic device. Depending on the functions implemented, the medium-to-long-term adaptive trading signal system based on a large model may include a path generation module 201, an index calculation module 202, a vector labeling module 203, an efficiency ratio calculation module 204, and a mechanism construction module 205. The modules described in this invention can also be referred to as units, which are a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, and are stored in the memory of the electronic device.

[0085] In this embodiment of the invention, the functions of each module / unit are as follows: The path generation module 201 is used to acquire target signal data, query the latent variable association graph corresponding to the target signal data, parse the structural association elements in the latent variable association graph, and generate the evolution path corresponding to the structural association elements. The index calculation module 202 is used to analyze the path response sequence corresponding to the evolution path, identify the sequence transition nodes in the path response sequence, and calculate the node chaos index corresponding to the sequence transition node. The vector labeling module 203 is used to perform mode decomposition on the target signal data based on the node chaos index to obtain signal sub-data clusters, perform label encoding on the signal sub-data clusters to obtain a basis function label set, query the sparse representation coefficients corresponding to the basis function label set, and label the representation weight vectors corresponding to the sparse representation coefficients. The efficiency ratio calculation module 204 is used to encode the representation weight vector into a preset historical memory module, monitor the convergence state of the data in the preset historical memory module, dynamically adjust the hyperparameter combination of the target signal data in the memory encoding process based on the convergence state, and calculate the generalization efficiency ratio corresponding to the hyperparameter combination. The mechanism construction module 205 is used to generate a distribution feature key corresponding to the target signal data based on the generalization efficiency ratio, query the stable identifier, differentiable identifier, and invariant identifier in the distribution feature key, and construct an adaptive trading signal mechanism corresponding to the target signal data based on the stable identifier, differentiable identifier, and invariant identifier.

[0086] In detail, the modules in the medium-to-long-term adaptive trading signal system 200 based on a large model described in this embodiment of the invention employ the same methods as described above. Figure 1 The technique is the same as that of a medium- to long-term adaptive trading signal system based on a large model, and can produce the same technical effect, so it will not be elaborated here.

[0087] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0088] Finally, it should be noted that the above 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 preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A medium- to long-term adaptive trading signal system based on a large model, characterized in that, The system includes: Obtain target signal data, query the latent variable correlation graph corresponding to the target signal data, parse the structural correlation elements in the latent variable correlation graph, and generate the evolution path corresponding to the structural correlation elements; Analyze the path response sequence corresponding to the evolution path, identify the sequence transition nodes in the path response sequence, and calculate the node chaos index corresponding to the sequence transition node; Based on the node chaos index, the target signal data is decomposed into mode to obtain signal sub-data clusters. The signal sub-data clusters are then labeled to obtain a basis function label set. The sparse representation coefficients corresponding to the basis function label set are queried, and the representation weight vectors corresponding to the sparse representation coefficients are labeled. The representation weight vector is encoded into a preset historical memory module, and the convergence status of the data in the preset historical memory module is monitored. Based on the convergence status, the hyperparameter combination of the target signal data in the memory encoding process is dynamically adjusted, and the generalization efficiency ratio corresponding to the hyperparameter combination is calculated. Based on the generalization efficiency ratio, a distribution feature key corresponding to the target signal data is generated. The stable identifier, differentiable identifier, and invariant identifier in the distribution feature key are queried. Based on the stable identifier, differentiable identifier, and invariant identifier, an adaptive trading signal mechanism corresponding to the target signal data is constructed.

2. The medium-to-long-term adaptive trading signal system based on a large model as described in claim 1, characterized in that, The generation of the evolution path corresponding to the structural association element includes: Analyze the element connection relationships corresponding to the structural associated elements; Identify the connection evolution nodes in the element connection relationships; Map the associated element branches corresponding to the connection evolution nodes; The associated element branches are marked to obtain branch stage marks; By integrating the branch stage markers, the evolution path corresponding to the structural association elements is generated.

3. The medium-to-long-term adaptive trading signal system based on a large model as described in claim 1, characterized in that, The analysis of the path response sequence corresponding to the evolution path includes: Extract path change features from the evolutionary path; Based on the dynamic change characteristics, the path response stages corresponding to the evolution path are divided; Mark the stage marker points in the path response stage; Analyze the path response fragments corresponding to the stage markers; Based on the path response fragments, the path response sequences corresponding to the evolutionary path are analyzed.

4. The medium-to-long-term adaptive trading signal system based on a large model as described in claim 1, characterized in that, The calculation of the node chaos index corresponding to the sequence transition node includes: The node chaos index corresponding to the sequence transition node is calculated using the following formula: ； in, This represents the node chaos index corresponding to the sequence transition node. This indicates the number of nodes corresponding to the sequence transition nodes. This represents the index of the number of nodes corresponding to the sequence transition. Indicates the first The state variables corresponding to each sequence transition node and This represents the start and end times corresponding to the chaotic time interval of a node. Indicates the first The acceleration corresponding to the change of state variables at each sequence transition node Indicates the first The rate of change of the state variables at each sequence transition node. This represents the average speed value.

5. The medium-to-long-term adaptive trading signal system based on a large model as described in claim 1, characterized in that, The modal decomposition of the target signal data based on the node chaos index yields signal sub-data clusters, including: Divide the oscillation mode intervals corresponding to the chaotic index of the nodes; Extract chaotic signal segments within the oscillation mode range; The chaotic signal segment is frequency-domain aligned to obtain a signal alignment label; Based on the signal alignment label, the target signal data is subjected to mode decomposition to obtain signal sub-data clusters.

6. The medium-to-long-term adaptive trading signal system based on a large model as described in claim 1, characterized in that, The step of label encoding the signal sub-data clusters to obtain a basis function label set includes: Analyze the time-frequency components of the data corresponding to the signal sub-data clusters; Construct the correlation component matrix corresponding to the time-frequency components of the data; Filter the data vector groups in the relevant component matrix; The vector parameters in the data vector group are normalized to obtain a standard basis function set; The data in the standard basis function set is labeled to obtain the basis function label set.

7. The medium-to-long-term adaptive trading signal system based on a large model as described in claim 1, characterized in that, The representation weight vector corresponding to the sparse representation coefficients includes: Parse the index of the sparse characterization coefficients; Based on the characterization coefficient index, the weight activation region corresponding to the sparse characterization coefficient is divided; Locate the active weight components in the weight activation region; Extract the non-zero feature terms corresponding to the active weight components; Based on the non-zero feature terms, label the representation weight vector corresponding to the sparse representation coefficients.

8. The medium-to-long-term adaptive trading signal system based on a large model as described in claim 1, characterized in that, The step of dynamically adjusting the hyperparameter combination of the target signal data during the memory encoding process based on the convergence state includes: Extract the state feature parameters corresponding to the convergence state; Generate a parameter adjustment list corresponding to the state feature parameters; Based on the parameter adjustment list, the correction hyperparameters corresponding to the target signal data are calculated item by item; The modified hyperparameters are segmented and filtered to obtain valid parameter items; Based on the effective parameter items, the hyperparameter combination of the target signal data during the memory encoding process is dynamically adjusted.

9. A medium- to long-term adaptive trading signal system based on a large model as described in claim 1, characterized in that, The calculation of the generalization efficiency ratio corresponding to the hyperparameter combination includes: The generalization efficiency ratio corresponding to the hyperparameter combination is calculated using the following formula: ； in, This represents the generalization performance ratio corresponding to the hyperparameter combination. This represents the total number of types included in the hyperparameter combination. This represents the category index corresponding to the hyperparameter combination. Indicates the first The weight coefficients corresponding to the hyperparameters Indicates the first Standardized values ​​of hyperparameters This indicates the time period corresponding to training the hyperparameter combination. This represents the parameter performance index corresponding to training the hyperparameter combination at time t. This represents the predicted sample corresponding to the hyperparameter combination. This represents the verification sample corresponding to the hyperparameter combination. This represents the correction factor.

10. A medium- to long-term adaptive trading signal system based on a large model as described in claim 1, characterized in that, The step of generating the distribution feature key corresponding to the target signal data based on the generalization efficiency ratio includes: Analyze the performance range threshold corresponding to the generalization performance ratio; Based on the performance interval threshold, extract the dynamic distribution entropy value corresponding to the target signal data; Based on the dynamic distribution entropy value, a dynamic key sequence corresponding to the target signal data is constructed; The dynamic key sequence is iteratively encrypted to obtain the core key seed; Based on the core key seed, a distribution feature key corresponding to the target signal data is generated.

Images