Stock market stability evaluation method and system combined with artificial intelligence

Abstract

The invention relates to the technical field of data processing and analysis, and discloses a stock market stability evaluation method and system combined with artificial intelligence, and the method comprises the steps: generating a state vector of stock entity data through an entity and an entity dimension, and constructing an interaction relation graph between the stock entity data through the state vector; querying a neighborhood node of each entity node in the interaction relation graph, and identifying spatial-temporal characteristics of the entity nodes in the interaction relation graph by using the neighborhood nodes; calculating a synchronization error between every two entity nodes in the entity nodes, identifying real-time key nodes in the interaction relation graph, and analyzing disturbance sensitivity of the real-time key nodes in the interaction relation graph; calculating a Lyapunov index of the interaction relation graph by using the dimension reduction processing vector; and based on the synchronization error, the disturbance sensitivity and the Lyapunov index, carrying out stability analysis on the stock market to obtain a stability analysis result. According to the invention, consideration of time-varying characteristics and chaotic characteristics can be improved.

Description

Technical Field

[0001] The present invention relates to a stock market stability assessment method and system combined with artificial intelligence, belonging to the technical field of data processing and analysis. Background Art

[0002] Stock market stability assessment combined with artificial intelligence is a method that integrates complex system theory. It models the nonlinear coupling relationship of the market through dynamic interactive graphs and quantifies the chaos intensity of the system using the Lyapunov index, thereby achieving cross-scale risk assessment from micro-behavior to macro-stability.

[0003] Currently, most existing patents rely on linear statistical models, such as VAR and GARCH, to perform stability analysis. This approach cannot capture the sudden changes and chaotic characteristics of the stock market. Furthermore, existing technologies rely on fixed network structures, such as correlation coefficient matrices, to perform stability analysis, which ignores the time-varying nature of market structure. Furthermore, traditional indicators, such as volatility and VaR, are insensitive to chaotic signals emerging in the stock market. Finally, existing machine learning models, such as LSTM models, are mostly black-box operations. Consequently, existing stability analysis fails to adequately consider time-varying and chaotic characteristics. Summary of the Invention

[0004] The present invention provides a stock market stability assessment method and system combined with artificial intelligence, the main purpose of which is to enhance the consideration of time-varying and chaotic characteristics.

[0005] To achieve the above objectives, the present invention provides a stock market stability assessment method combined with artificial intelligence, comprising: Collecting stock entity data from the stock market over a historical period, wherein the stock entity data includes the entity itself and entity dimensions, generating a state vector for the stock entity data using the entity itself and the entity dimensions, and constructing an interaction relationship graph between each stock entity data using the state vector; Querying neighboring nodes of each entity node in the interaction relationship graph, and using the neighboring nodes to identify spatiotemporal features of the entity node in the interaction relationship graph; Calculating synchronization errors between each of the entity nodes according to the spatiotemporal characteristics, identifying real-time key nodes in the interaction relationship graph, and analyzing disturbance sensitivity of the real-time key nodes in the interaction relationship graph; Performing data dimensionality reduction processing on the state vector corresponding to each entity node in the interaction relationship graph to obtain a dimensionality reduction processing vector, and calculating the Lyapunov index of the interaction relationship graph using the dimensionality reduction processing vector; Based on the synchronization error, the disturbance sensitivity and the Lyapunov index, a stability analysis is performed on the stock market to obtain a stability analysis result.

[0006] Optionally, constructing an interaction relationship diagram between each stock entity data using the state vector includes: calculating a vector-vector weight between each state vector in the state vector; selecting a high numerical weight from the vector-vector weights; The high numerical weight is used as the edge weight; the state vector is used as the entity node; Based on the edge weights and the entity nodes, construct a weighted undirected graph between each entity itself; The weighted undirected graph is used as an interaction relationship graph.

[0007] Optionally, querying the neighboring nodes of each entity node in the interaction relationship graph includes: Querying all target nodes in the interaction relationship graph that are connected to the entity node; Identifying the number of shortest path edges between the entity node and the target node; According to the number of edges in the shortest path, the target node is divided into nodes to obtain neighboring nodes.

[0008] Optionally, the identifying the spatiotemporal features of the entity node in the interaction relationship graph by using the neighborhood nodes includes: Identifying, based on the neighborhood nodes, spatial structural features of the entity node in the interaction relationship graph; Integrate the time information of the entity node into the spatial structure feature to obtain the fusion feature The fusion feature is used as the spatiotemporal feature of the entity node in the interaction relationship graph.

[0009] Optionally, calculating the synchronization error between each of the entity nodes according to the spatiotemporal features includes: Calculating the cosine similarity between each entity node in the entity nodes according to the spatiotemporal features; Based on the cosine similarity, a synchronization error between each of the entity nodes is calculated.

[0010] Optionally, identifying the real-time key nodes in the interaction relationship diagram includes: Calculating the betweenness centrality of each entity node in the interaction relationship graph; The betweenness centrality is normalized to obtain the standard centrality; Calculating a critical judgment threshold of the standard centrality; When the standard centrality is greater than the key judgment threshold, the central node corresponding to the standard centrality is used as a real-time key node.

[0011] Optionally, analyzing the disturbance sensitivity of the real-time key node in the interaction relationship diagram includes: Obtaining the betweenness centrality corresponding to the real-time key node; Obtaining a global edge weight matrix in the interaction relationship graph; Calculating the centrality gradient of the betweenness centrality with respect to the global edge weight matrix; Based on the centrality gradient, the disturbance sensitivity of the real-time key node in the interaction relationship graph is calculated using the following formula: ; in, represents the sensitivity to disturbance, express About the Global Edge Weight Matrix The centrality gradient of represents the Frobenius norm, Indicates the The betweenness centrality of real-time key nodes, Indicates the number of real-time key nodes in the interaction relationship diagram.

[0012] Optionally, the calculating the Lyapunov index of the interaction relationship graph by using the dimensionality reduction processing vector includes: The differential change of the dimension reduction processing vector is calculated using the following formula: ; in, represents the differential change, Represents the first differential change corresponding to the dimensionality reduction processing vector, Represents the second differential change corresponding to the dimensionality reduction processing vector, Represents the third differential change corresponding to the dimensionality reduction processing vector, represents the linear response parameter, represents the driving force parameter, represents the dissipation parameter; Splice the differential changes corresponding to all entity nodes in the interaction relationship graph to obtain the spliced ​​change; The Jacobian matrix of the splicing variation is calculated using the following formula: ; ; in, represents the Jacobian matrix, 、 Indicates the splicing change, represents the time variable, represents a differential equation function; According to the Jacobian matrix, the Lyapunov index of the interaction relationship graph is calculated using the following formula: ; ; ; in, represents the Lyapunov exponent, express The growth rate of time, represents the time window, Express The value of the first row and first column in the upper triangular matrix obtained by orthogonal triangular decomposition, represents the differential time change, Express The upper triangular matrix obtained by orthogonal triangular decomposition is Express The new orthogonal basis obtained by orthogonal triangular decomposition is represents the old orthogonal basis, express The Jacobian matrix of .

[0013] Optionally, performing stability analysis on the stock market based on the synchronization error, the disturbance sensitivity, and the Lyapunov index to obtain a stability analysis result includes: Identifying a positive or negative correlation between the synchronization error, the disturbance sensitivity, the Lyapunov index, and the stability of the stock market; Determining the weight signs of the synchronization error, the disturbance sensitivity, and the Lyapunov exponent according to the positive and negative correlation; Normalizing the synchronization error, the disturbance sensitivity, and the Lyapunov exponent to obtain normalized parameters; Calculating a weight parameter of the normalization parameter according to the positive and negative signs of the weights; Performing weighted sum processing on the normalized parameters using the weight parameters to obtain a weighted sum result; Performing nonlinear mapping on the weighted summation result using a preset hyperbolic tangent function to obtain a nonlinear mapping result; A stability analysis is performed on the stock market based on the stability score interval corresponding to the nonlinear mapping result to obtain a stability analysis result.

[0014] In order to solve the above problems, the present invention further provides a stock market stability assessment system combined with artificial intelligence, the system comprising: A relationship graph construction module is configured to collect stock entity data from the stock market over a historical period, wherein the stock entity data includes the entity itself and entity dimensions, generate a state vector for the stock entity data using the entity itself and the entity dimensions, and use the state vector to construct an interaction relationship graph between each stock entity data; A feature recognition module, configured to query neighboring nodes of each entity node in the interaction relationship graph, and use the neighboring nodes to identify spatiotemporal features of the entity node in the interaction relationship graph; a sensitivity analysis module, configured to calculate the synchronization error between each of the entity nodes according to the spatiotemporal characteristics, identify the real-time key nodes in the interaction relationship graph, and analyze the sensitivity of the real-time key nodes to disturbances in the interaction relationship graph; A Liapunov index calculation module is used to perform data dimensionality reduction processing on the state vector corresponding to each entity node in the interaction relationship graph to obtain a dimensionality reduction processing vector, and use the dimensionality reduction processing vector to calculate the Lyapunov index of the interaction relationship graph; The stability analysis module is used to perform stability analysis on the stock market based on the synchronization error, the disturbance sensitivity and the Lyapunov index to obtain a stability analysis result.

[0015] Compared with the problem described in the background technology, the embodiment of the present invention generates the state vector of the stock entity data through the entity itself and the entity dimension, so as to describe a single stock through the state vector, and concretize the abstract single stock data into a state vector, so as to pave the way for the subsequent establishment of a relationship network structure between each stock. Furthermore, the embodiment of the present invention constructs an interactive relationship graph between each stock entity data by using the state vector, and adaptively adjusts the node coupling weight through the kernel function, so as to reflect the changes in capital flow and information propagation path in real time. The embodiment of the present invention queries the neighborhood nodes of each entity node in the interactive relationship graph to express the weighted undirected graph relationship between the entity node and the neighborhood nodes of this entity node. The graph structure relationship between each individual stock in the stock market is shown, thereby capturing the nonlinear relationship between each individual stock, paving the way for the subsequent analysis of the structural stability of the entire stock market. Furthermore, the embodiment of the present invention utilizes the neighborhood nodes to identify the spatiotemporal characteristics of the entity nodes in the interaction relationship graph, and combines the attention mechanism to visualize the graph nodes with risks, thereby reducing black box operations. The embodiment of the present invention calculates the synchronization error between each entity node in the entity node according to the spatiotemporal characteristics, so as to reflect the instantaneous changes of the market collaborative behavior in real time by calculating the synchronization error at each moment. In addition, an ideal synchronization threshold is introduced to distinguish normal collaboration from over-synchronization and loss of synchronization. Finally, the high error is combined with the interaction relationship graph. The edge is marked as a key propagation path that loses stability. Furthermore, an embodiment of the present invention analyzes the global stability of the stock market by identifying the real-time key nodes in the interaction relationship graph through the influence of parameter changes and parameter disturbances on the real-time key nodes in the interaction relationship graph. This can reduce the amount of data and data complexity. This is because when the real-time key nodes in the interaction relationship graph are not identified, but all entities in the interaction relationship graph are used for subsequent calculations, there will be a problem of excessive data volume and inefficient processing. Furthermore, an embodiment of the present invention analyzes the sensitivity of the real-time key nodes in the interaction relationship graph to disturbances and parameter changes in the real-time key nodes in the interaction relationship graph. The sensitivity presented by the influence is used as the global and overall sensitivity of the entire interaction relationship diagram to the influence of parameter disturbances and parameter changes. Furthermore, the embodiment of the present invention calculates the Lyapunov index of the interaction relationship diagram by using the dimensionality reduction processing vector to judge whether the market is in a stable state or a chaotic state through the Lyapunov index. If the Lyapunov index is positive, it indicates that the market is very sensitive to the initial conditions. Small disturbances may cause large price fluctuations, which may indicate market instability. The stock market is essentially a complex dynamic system. The fluctuations of various values ​​are affected by multiple factors. The Lyapunov index can quantify the sensitivity of this dynamic system and help understand the market's dependence on initial conditions.This embodiment of the present invention performs a stability analysis of the stock market based on the synchronization error, the disturbance sensitivity, and the Lyapunov exponent. This analysis enhances consideration of time variability based on the time information contained in the synchronization error, the disturbance sensitivity, and the Lyapunov exponent, and increases consideration of chaotic characteristics through the Lyapunov exponent. Therefore, the stock market stability assessment method and system combined with artificial intelligence provided by this embodiment of the present invention can enhance consideration of time variability and chaotic characteristics. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic diagram of a flow chart of a method for evaluating stock market stability using artificial intelligence according to an embodiment of the present invention; Figure 2 A schematic diagram of modules for implementing the stock market stability assessment system combined with artificial intelligence, provided in one embodiment of the present invention.

[0017] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0018] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0019] The present embodiment provides a method for assessing stock market stability using artificial intelligence. The method can be performed by at least one of a server, a terminal, or other electronic device capable of executing the method provided by the present embodiment. In other words, the method can be executed by software or hardware installed on a terminal or 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 FIG2 is a flow chart of a method for evaluating stock market stability using artificial intelligence according to an embodiment of the present invention. In this embodiment, the method for evaluating stock market stability using artificial intelligence includes: S1. Collect stock entity data from the stock market within a historical period, wherein the stock entity data includes the entity itself and entity dimensions, generate a state vector of the stock entity data using the entity itself and the entity dimensions, and use the state vector to construct an interaction relationship diagram between each stock entity data.

[0021] In an embodiment of the present invention, the stock market refers to a venue or platform where investors trade stocks (financial instruments representing company ownership shares). The stock market has functions such as determining the real-time fair price of stocks through a bidding process between buyers and sellers, allowing investors to quickly buy and sell assets without excessively influencing prices, directing social capital to companies with growth potential, hedging investment risks through instruments such as derivatives, and reflecting macroeconomic expectations and industry prosperity. Therefore, the stock entity data refers to data generated when the stock entity operates within the functions of the stock market, and the stock entity refers to a single stock. The single stock refers to a standardized financial instrument issued by a listed company on a stock exchange that represents its ownership. Each listed company corresponds to only one stock. The historical period refers to the period before the current stability analysis of the stock market. Furthermore, the entity itself refers to the single stock, and the entity dimensions refer to data generated when the entity itself operates within the functions of the stock market, such as price change data, trading volume data, open interest data, and pending order volume data.

[0022] Furthermore, an embodiment of the present invention generates a state vector of the stock entity data through the entity itself and the entity dimension, so as to describe a single stock through the state vector, concretize the abstract single stock data into a state vector, and pave the way for the subsequent establishment of a relationship network structure between each stock.

[0023] In the embodiment of the present invention, the state vector refers to a matrix obtained by explicitly representing the entity itself using the entity dimensions. The state vector is represented by the following matrix: ; ; ; ; ; in, express Moment The state vector of the class entity itself, Indicates the The price percentage yield of the class entity itself, Indicates the time t The price of the class entity itself, Represents the time interval from time t historical moment Time The price of the class entity itself, represents the mean value within the sliding window, represents the standard deviation within the sliding window, Express After dedimensionalization, the value Indicates time period The trading volume volatility within Indicates the The trading volume at the moment, express The trading volume at the previous moment, express The logarithm of the rate of change, Indicates time period within The mean of Represents time t and Moment The difference between When it is greater than 0, it indicates a new position. When it is not greater than 0, the position is closed. Indicates the ask order volume of the jth bid / ask price index at time t, Indicates the total number of bid and ask price indexes, Indicates the quantity of sell orders at time t, Indicates the bid order volume of the jth bid-ask price index at time t, Indicates the buy order volume at time t, Represents the time interval from time t historical moment The amount of ask orders at the time Represents the time interval from time t historical moment The amount of pending orders at the bid price at Indicates the rate of change of pending order volume, It represents the bid-ask spread between different stocks. Indicates the the bid price of the class entity itself, Indicates the The selling price of the class entity itself, Indicates the the bid price of the class entity itself, Indicates the The selling price of the class entity itself, express The variance in consecutive time periods, express The variance in consecutive time periods, , Represents the number of entities themselves, and also the number of entity nodes in the subsequent interaction relationship diagram; It should be noted that the time interval is, for example, 1 minute, 1 day, etc., while the sliding window is, for example, 20 days, etc. The sliding window is longer and is used to measure data changes over a longer period of time, while the time interval is used to measure data changes at an instant or short moment. Secondly, regarding the ask volume of the jth bid-ask index and the buy volume of the jth bid-ask index, the following are examples: The total number of bid-ask indexes is For example, if there are four buying prices of 10, 20, 30, and 40, then j is 1, 2, 3, and 4. The order volume of the first bid price 10 is the bid order volume of the first ask price index. The same is true for the ask price. The ask price index mentioned here is not the total number of ask and bid prices, but the total number of bid prices or the total number of ask prices.

[0024] Furthermore, the embodiment of the present invention utilizes the state vector to construct an interactive relationship diagram between each stock entity data, and adaptively adjusts the node coupling weight through the kernel function, thereby reflecting the changes in capital flow and information propagation path in real time.

[0025] The interaction relationship graph refers to a weighted undirected graph, where each node represents an entity itself and each edge between nodes carries an edge weight.

[0026] In one embodiment of the present invention, constructing an interaction relationship graph between each stock entity data using the state vector includes: calculating a vector-vector weight between each state vector in the state vector using the following formula: ; in, Indicates the The state vector and The vector-vector weights between the state vectors, represents the nonlinear phase space embedding function, Used to The state vector is mapped to a high-dimensional manifold, express Moment The state vector of the class entity itself, Indicates the The state vector of the class entity itself, represents the adaptive kernel width, ; A high numerical weight is selected from the vector-vector weights; the high numerical weight is used as an edge weight; the state vector is used as an entity node; based on the edge weight and the entity node, a weighted undirected graph between each entity itself is constructed; and the weighted undirected graph is used as an interaction relationship graph.

[0027] The nonlinear phase space embedding function refers to a nonlinear mapping function, and the nonlinear mapping function is such as a kernel function, such as a Gaussian kernel function, a polynomial kernel function, and the like.

[0028] It should be noted that for , according to the formula Dynamic Adjustment The size of Represents a function for taking the median. Further, regarding high numerical weights, the high numerical weights refer to weights in the vector-vector weights whose absolute values ​​are greater than a threshold. The threshold is determined based on the quantile of historical data, such as retaining the absolute values ​​of the vector-vector weights in the top 10% quantiles.

[0029] S2. Query the neighboring nodes of each entity node in the interaction relationship graph, and use the neighboring nodes to identify the spatiotemporal features of the entity node in the interaction relationship graph.

[0030] The embodiment of the present invention queries the neighborhood nodes of each entity node in the interactive relationship graph to represent the graph structure relationship between individual stocks in the stock market through the weighted undirected graph relationship between the entity node and the neighborhood nodes of this entity node, thereby capturing the nonlinear relationship between individual stocks and laying the foundation for subsequent analysis of the structural stability of the entire stock market.

[0031] The domain node refers to a node that has a path with the entity node.

[0032] In one embodiment of the present invention, querying the neighborhood nodes of each entity node in the interaction relationship graph includes: querying all target nodes connected to the entity node in the interaction relationship graph; identifying the number of shortest path edges between the entity node and the target node; and classifying the target node according to the number of shortest path edges to obtain neighborhood nodes.

[0033] Among them, the target node refers to a node that is k-order connected to the entity node. For example, if the number of edges of the shortest path between the target node and the entity node is 1 edge, then the k-order is 1 order; if the number of edges of the shortest path between the target node and the entity node is 2 edges, then the k-order is 2 order; if the number of edges of the shortest path between the target node and the entity node is 3 edges, then the k-order is 3 order. Therefore, the number of edges of the shortest path refers to the number of edges of the shortest path between the target node and the entity node. Furthermore, the process of classifying the target node into nodes of order 1, order 2, and order k is also the process of classifying the target node into order 1, order 2, and order k.

[0034] Furthermore, an embodiment of the present invention utilizes the neighborhood nodes to identify the spatiotemporal features of the entity nodes in the interaction relationship graph, and combines the attention mechanism to visualize the graph nodes with risks, thereby reducing black box operations.

[0035] In one embodiment of the present invention, the step of using the neighborhood nodes to identify the spatiotemporal features of the entity node in the interaction relationship graph includes: identifying the spatial structural features of the entity node in the interaction relationship graph using the following formula based on the neighborhood nodes: ; in, Represents the spatial structure characteristics, Indicates the entity nodes The set of order neighbor nodes, Indicates the The linear transformation matrix of the order neighborhood nodes, represents the number of shortest path edges between the neighborhood node and the target node, Indicates the maximum number of shortest path edges among the shortest path edges. represents the feature concatenation operation, express The index of any neighbor node in the set of order neighbor nodes, express The corresponding neighboring nodes, represents the activation function, represents a multilayer perceptron, Indicates the The entity node and Hyperbolic attention weights between neighboring nodes; The time information of the entity node is integrated into the spatial structure feature using the following formula to obtain the fusion feature: ; ; ; in, Indicates the The fusion features of entity nodes, Indicates the length of the time window, represents the basic time step, represents the causal dilated convolution group, represents the dynamic gating vector, and Consistent, Represents element-by-element multiplication, LayerNorm represents layer standardization operation, represents the time characteristics of the entity node, Represents the spatial structure characteristics, express The moment entity nodes, express The moment entity nodes, express The linear transformation matrix of The fusion feature is used as the spatiotemporal feature of the entity node in the interaction relationship graph.

[0036] The causal dilated convolution group consists of dilated convolution and causal convolution.

[0037] S3. Calculate the synchronization error between each entity node in the entity nodes according to the spatiotemporal characteristics, identify the real-time key nodes in the interaction relationship graph, and analyze the disturbance sensitivity of the real-time key nodes in the interaction relationship graph.

[0038] The embodiment of the present invention calculates the synchronization error between each entity node in the entity nodes according to the spatiotemporal characteristics, so as to reflect the instantaneous changes of market collaborative behavior in real time by calculating the synchronization error at each moment. In addition, an ideal synchronization threshold is introduced to distinguish normal collaboration from excessive synchronization and loss of synchronization. Finally, the high-error edge is marked as a critical propagation path that loses stability in combination with the interaction relationship diagram.

[0039] The synchronization error refers to an error value indicating whether two different physical nodes are coordinated.

[0040] In one embodiment of the present invention, calculating the synchronization error between each entity node in the entity nodes according to the spatiotemporal features includes: calculating the cosine similarity between each entity node in the entity nodes according to the spatiotemporal features; and calculating the synchronization error between each entity node in the entity nodes based on the cosine similarity using the following formula: ; in, represents the synchronization error, represents the ideal synchronization threshold, , express and The cosine similarity between Indicates the The fusion features of entity nodes, Indicates the The fusion features of entity nodes, Indicates the number of entity nodes in the interaction graph.

[0041] Optionally, the process of calculating the cosine similarity between each entity node in the entity nodes according to the spatiotemporal features refers to calculating the cosine similarity between the spatiotemporal features corresponding to each entity node.

[0042] Furthermore, an embodiment of the present invention analyzes the global stability of the stock market by identifying real-time key nodes in the interaction relationship diagram through the influence of parameter changes and parameter disturbances on the real-time key nodes in the interaction relationship diagram. This can reduce the amount of data and data complexity. This is because when the real-time key nodes in the interaction relationship diagram are not identified, but all entities in the interaction relationship diagram are used to perform subsequent calculations, the problem of excessive data volume and inefficient processing will arise.

[0043] The real-time key nodes refer to important nodes in the interaction relationship diagram detected in real time, and the real-time key nodes may be different at different times.

[0044] In one embodiment of the present invention, identifying the real-time key nodes in the interaction relationship graph includes: calculating the betweenness centrality of each entity node in the interaction relationship graph; normalizing the betweenness centrality to obtain a standard centrality; and calculating a key judgment threshold of the standard centrality using the following formula: ; in, Indicates the critical judgment threshold, express The mean of express The variance of Indicates that the sample size exceeds the threshold, represents the total sample size, represents the shape parameter, Indicates the The betweenness centrality of entity nodes; When the standard centrality is greater than the key judgment threshold, the central node corresponding to the standard centrality is used as a real-time key node.

[0045] The betweenness centrality is also called betweenness centrality. It is a measure of graph centrality based on the shortest path. For each pair of nodes in a connected graph, there is at least one shortest path between the nodes, such that the number of edges passed by the path (for unweighted graphs) or the sum of edge weights (for weighted graphs) is minimized. The specific calculation formula is: ; in, Representation node The betweenness centrality of Represents a slave node To Node The total number of shortest paths, Represents a slave node To Node The shortest path through the node The number of paths, Furthermore, the key judgment threshold refers to the threshold for judging whether a node is a key node. The formula for calculating the key judgment threshold is based on the extreme value theory and uses the tail model of the generalized Pareto distribution model. The central node is the node when calculating the betweenness centrality. , express The mean here refers to the mean within a period of time or within a time window. Indicates that the sample size exceeds the threshold, represents the total sample size, represents the shape parameter, and the total sample size refers to the number of observations in the entire data set. For example, , then the number of days in the past ten years is the total sample size, and the sample size exceeding the threshold is The number of extreme values ​​of The reason is the balance deviation ( and If the deviation between If it is too small, it may lead to large deviations in the estimation and unstable results. If it is too large, it may introduce non-extreme values ​​and cause estimation bias. The shape parameter determines the tail behavior of the distribution. A positive shape parameter indicates a fat-tailed distribution, which is common in financial data. A negative shape parameter indicates a short-tailed distribution. The estimation of the shape parameter is usually based on the sample size exceeding the threshold and is obtained by maximum likelihood estimation or moment estimation method. When setting the shape parameter, the actual tail characteristics of the data must be considered. For example, the risk of financial market collapse usually presents a fat tail, and the shape parameter is positive at this time.

[0046] Optionally, the process of normalizing the betweenness centrality to obtain the standard centrality is the process of subtracting the mean from the betweenness centrality and dividing it by the variance, which is similar to the above-mentioned The principle is similar and will not be further elaborated here.

[0047] Furthermore, an embodiment of the present invention analyzes the sensitivity of the real-time key nodes to disturbances in the interaction relationship diagram, and uses the sensitivity of the real-time key nodes in the interaction relationship diagram to parameter disturbances and parameter changes as the global and overall sensitivity of the entire interaction relationship diagram to parameter disturbances and parameter changes.

[0048] The disturbance sensitivity refers to the sensitivity of the real-time key node to the influence of parameter disturbance and parameter change in the interaction relationship diagram.

[0049] In one embodiment of the present invention, analyzing the perturbation sensitivity of the real-time key node in the interaction relationship graph includes: obtaining the betweenness centrality corresponding to the real-time key node; obtaining the global edge weight matrix in the interaction relationship graph; calculating the centrality gradient of the betweenness centrality with respect to the global edge weight matrix; and calculating the perturbation sensitivity of the real-time key node in the interaction relationship graph based on the centrality gradient using the following formula: ; in, represents the sensitivity to disturbance, express About the Global Edge Weight Matrix The centrality gradient of represents the Frobenius norm, Indicates the The betweenness centrality of real-time key nodes, Indicates the number of real-time key nodes in the interaction relationship diagram.

[0050] Among them, the global edge weight matrix refers to the matrix composed of all edge weights in the interaction relationship graph. For example, the edge weight of the first row and first column in the global edge weight matrix is ​​the edge weight between the first node and the first node in the interaction relationship graph. The centrality gradient refers to the partial derivative and gradient of the betweenness centrality with respect to the global edge weight matrix. It should be noted that the Frobenius norm can be intuitively understood as the "energy" or "size" of the matrix. It reflects the overall degree of change of all elements in the matrix. In the interaction relationship graph, by calculating the Frobenius norm corresponding to the key node, the overall change of the node after the disturbance can be directly quantified, thereby evaluating its sensitivity.

[0051] S4. Perform data dimensionality reduction processing on the state vector corresponding to each entity node in the interaction relationship graph to obtain a dimensionality reduction processing vector, and use the dimensionality reduction processing vector to calculate the Lyapunov index of the interaction relationship graph.

[0052] Optionally, the process of performing data dimensionality reduction processing on the state vector corresponding to each entity node in the interaction relationship diagram to obtain the dimensionality reduction processing vector is achieved through tensor principal component analysis (TPCA). The tensor principal component analysis (TPCA) refers to a dimensionality reduction method for processing high-dimensional tensor data, which aims to characterize high-dimensional tensor data through low-dimensional subspaces. It extends traditional principal component analysis (PCA) and can better process multidimensional data structures. TPCA extracts the main components in the data through tensor decomposition technology, thereby achieving data dimensionality reduction and feature extraction.

[0053] Furthermore, an embodiment of the present invention calculates the Lyapunov index of the interaction relationship diagram by using the dimensionality reduction processing vector to determine whether the market is in a stable state or a chaotic state through the Lyapunov index. If the Lyapunov index is positive, it indicates that the market is very sensitive to the initial conditions, and a small disturbance may cause large price fluctuations, which may indicate market instability. The stock market is essentially a complex dynamic system, and the fluctuations of various numerical values ​​are affected by multiple factors. The Lyapunov index can quantify the sensitivity of this dynamic system and help understand the market's dependence on initial conditions.

[0054] The Lyapunov Exponent is an indicator used to measure the speed at which small differences in initial conditions in a dynamic system diverge or converge over time.

[0055] In one embodiment of the present invention, the calculating of the Lyapunov index of the interaction relationship graph using the dimensionality reduction processing vector includes: calculating the differential change of the dimensionality reduction processing vector using the following formula: ; in, represents the differential change, Represents the first differential change corresponding to the dimensionality reduction processing vector, Represents the second differential change corresponding to the dimensionality reduction processing vector, Represents the third differential change corresponding to the dimensionality reduction processing vector, represents the linear response parameter, represents the driving force parameter, represents the dissipation parameter; The differential changes corresponding to all entity nodes in the interaction relationship graph are concatenated to obtain the concatenated change; the Jacobian matrix of the concatenated change is calculated using the following formula: ; ; in, represents the Jacobian matrix, 、 Indicates the splicing change, represents the time variable, represents a differential equation function; According to the Jacobian matrix, the Lyapunov index of the interaction relationship graph is calculated using the following formula: ; ; ; in, represents the Lyapunov exponent, express The growth rate of time, represents the time window, Express The value of the first row and first column in the upper triangular matrix obtained by orthogonal triangular decomposition, represents the differential time change, Express The upper triangular matrix obtained by orthogonal triangular decomposition is Express The new orthogonal basis obtained by orthogonal triangular decomposition is represents the old orthogonal basis, express The Jacobian matrix of .

[0056] It should be noted that represents the linear response parameter, represents the driving force parameter, Represents dissipation parameters. These parameters are obtained through maximum likelihood estimation, Bayesian optimization, etc. The linear response parameters are used to control the linear coupling strength between variables. The driving force parameters are used to adjust the energy input of the system. The dissipation parameters are used to describe the energy dissipation or attenuation rate of the system. 、 、 are all parameters that already exist in the original Lorenz system, but they are given different meanings in the stock market scenario of the present invention. The splicing variation is the result of splicing the vectors of all nodes. Each node has three vectors, namely 、 、 , concatenate the three vectors of each node, for example, concatenate the vector of node 1 and the vector of node 2 to get 、 、 、 、 , that is, a matrix vector with one row and six columns, 、 、 represents the vector of node 1, 、 、 represents the vector of node 2. Further, 、 、 For the general The result of dimensionality reduction, citing the principle of differential calculus, dx represents a small change in x, and the differential change corresponding to the above dimensionality reduction vector is the same as dx. ,exist is the initial moment, for example, when the initial moment is moment 1, is the unit orthogonal basis matrix, for , calculated using the Runge-Kutta method The Runge-Kutta method is a high-precision single-step algorithm widely used in engineering, mainly used for numerical solution of differential equations. The principle of calculation and The calculation principle is similar, but they use different moments to calculate. Here we use the value of the first row and first column in the upper triangular matrix to calculate (If the value of the second row and second column in the upper triangular matrix is ​​selected, the calculation is ) is because is the maximum Li index, each Li index (these Li indices include 、 、 etc.) independently reflect the dynamic characteristics of the system in different orthogonal directions, among which only the largest Li index It reflects the chaotic characteristics of the system.

[0057] S5. Based on the synchronization error, the disturbance sensitivity, and the Lyapunov index, perform stability analysis on the stock market to obtain a stability analysis result.

[0058] The embodiment of the present invention performs a stability analysis on the stock market based on the synchronization error, the disturbance sensitivity and the Lyapunov index, so as to enhance the consideration of time variability based on the time information contained in the synchronization error, the disturbance sensitivity and the Lyapunov index, and increase the consideration of chaotic characteristics through the Lyapunov index.

[0059] In one embodiment of the present invention, the stability analysis of the stock market is performed based on the synchronization error, the disturbance sensitivity and the Lyapunov index to obtain a stability analysis result, including: identifying the positive and negative correlation between the synchronization error, the disturbance sensitivity and the Lyapunov index and the stability of the stock market; determining the positive and negative signs of the weights of the synchronization error, the disturbance sensitivity and the Lyapunov index based on the positive and negative correlation; normalizing the synchronization error, the disturbance sensitivity and the Lyapunov index to obtain normalization parameters; calculating weight parameters of the normalization parameters based on the positive and negative signs of the weights; performing weighted summation on the normalization parameters using the weight parameters to obtain a weighted summation result; performing nonlinear mapping on the weighted summation result using a preset hyperbolic tangent function to obtain a nonlinear mapping result; performing stability analysis on the stock market based on the stability score interval corresponding to the nonlinear mapping result to obtain a stability analysis result.

[0060] Optionally, the process of identifying the positive and negative correlation between the synchronization error, the disturbance sensitivity and the Lyapunov index and the stability of the stock market is a manual identification process. For example, the larger the synchronization error, the more unstable the market, and the two are negatively correlated. The larger the disturbance sensitivity, the more unstable the market, and the two are also negatively correlated. The Lyapunov index is a positive number, and the more chaotic the market, that is, the more unstable, the two are negatively correlated. The process of determining the positive and negative signs of the weights of the synchronization error, the disturbance sensitivity and the Lyapunov index according to the positive and negative correlation means that the weight corresponding to the positive correlation is a positive number, and the weight corresponding to the negative correlation is a negative number. The principle of normalizing the synchronization error, the disturbance sensitivity and the Lyapunov index to obtain the normalization parameter is the same as the aforementioned The principle is similar and will not be further elaborated here. Furthermore, the process of calculating the weight parameter of the normalization parameter according to the positive and negative signs of the weights refers to calculating a large number of historical normalization parameter samples through historical data, and fitting the weights of the historical normalization parameter samples through the least squares method. The historical normalization parameter samples here are different from the normalization parameters of this embodiment. The normalization parameters of this embodiment refer to the current parameters, while the historical normalization parameter samples refer to historical parameters. Furthermore, the weighted summation result is nonlinearly mapped using a preset hyperbolic tangent function to obtain a nonlinear mapping The process of obtaining the result refers to the process of using the weighted summation result as the input data of the hyperbolic tangent function, the hyperbolic tangent function outputs a value, and using the value output by the hyperbolic tangent function as the nonlinear mapping result. Furthermore, the stability analysis of the stock market is performed based on the stability score interval corresponding to the nonlinear mapping result. The process of obtaining the stability analysis result refers to pre-dividing the intervals corresponding to different hyperbolic tangent function output values ​​based on historical experience. For example, a hyperbolic tangent function output value greater than 0.7 indicates stability, between 0.4 and 0.7 indicates instability, and less than 0.4 indicates extreme instability.

[0061] Compared with the problem described in the background technology, the embodiment of the present invention generates the state vector of the stock entity data through the entity itself and the entity dimension, so as to describe a single stock through the state vector, and concretize the abstract single stock data into a state vector, so as to pave the way for the subsequent establishment of a relationship network structure between each stock. Furthermore, the embodiment of the present invention constructs an interactive relationship graph between each stock entity data by using the state vector, and adaptively adjusts the node coupling weight through the kernel function, so as to reflect the changes in capital flow and information propagation path in real time. The embodiment of the present invention queries the neighborhood nodes of each entity node in the interactive relationship graph to express the weighted undirected graph relationship between the entity node and the neighborhood nodes of this entity node. The graph structure relationship between each individual stock in the stock market is shown, thereby capturing the nonlinear relationship between each individual stock, paving the way for the subsequent analysis of the structural stability of the entire stock market. Furthermore, the embodiment of the present invention utilizes the neighborhood nodes to identify the spatiotemporal characteristics of the entity nodes in the interaction relationship graph, and combines the attention mechanism to visualize the graph nodes with risks, thereby reducing black box operations. The embodiment of the present invention calculates the synchronization error between each entity node in the entity node according to the spatiotemporal characteristics, so as to reflect the instantaneous changes of the market collaborative behavior in real time by calculating the synchronization error at each moment. In addition, an ideal synchronization threshold is introduced to distinguish normal collaboration from over-synchronization and loss of synchronization. Finally, the high error is combined with the interaction relationship graph. The edge is marked as a key propagation path that loses stability. Furthermore, an embodiment of the present invention analyzes the global stability of the stock market by identifying the real-time key nodes in the interaction relationship graph through the influence of parameter changes and parameter disturbances on the real-time key nodes in the interaction relationship graph. This can reduce the amount of data and data complexity. This is because when the real-time key nodes in the interaction relationship graph are not identified, but all entities in the interaction relationship graph are used for subsequent calculations, there will be a problem of excessive data volume and inefficient processing. Furthermore, an embodiment of the present invention analyzes the sensitivity of the real-time key nodes in the interaction relationship graph to disturbances and parameter changes in the real-time key nodes in the interaction relationship graph. The sensitivity presented by the influence is used as the global and overall sensitivity of the entire interaction relationship diagram to the influence of parameter disturbances and parameter changes. Furthermore, the embodiment of the present invention calculates the Lyapunov index of the interaction relationship diagram by using the dimensionality reduction processing vector to judge whether the market is in a stable state or a chaotic state through the Lyapunov index. If the Lyapunov index is positive, it indicates that the market is very sensitive to the initial conditions. Small disturbances may cause large price fluctuations, which may indicate market instability. The stock market is essentially a complex dynamic system. The fluctuations of various values ​​are affected by multiple factors. The Lyapunov index can quantify the sensitivity of this dynamic system and help understand the market's dependence on initial conditions.This embodiment of the present invention performs a stability analysis of the stock market based on the synchronization error, the disturbance sensitivity, and the Lyapunov exponent. This analysis enhances consideration of time variability based on the time information contained in the synchronization error, the disturbance sensitivity, and the Lyapunov exponent, and increases consideration of chaotic characteristics through the Lyapunov exponent. Therefore, the stock market stability assessment method and system combined with artificial intelligence provided by this embodiment of the present invention can enhance consideration of time variability and chaotic characteristics.

[0062] Example 2: like Figure 2 1 is a functional module diagram of a stock market stability assessment system combined with artificial intelligence according to the present invention.

[0063] The AI-integrated stock market stability assessment system 200 described in the present invention can be installed in an electronic device. Depending on the functionality implemented, the AI-integrated stock market stability assessment system may include a relationship graph construction module 201, a feature recognition module 202, a sensitivity analysis module 203, a Lie index calculation module 204, and a stability analysis module 205. A module, also referred to as a unit, refers to a series of computer program segments that can be executed by an electronic device processor and perform a fixed function, and is stored in the electronic device's memory.

[0064] In the embodiment of the present invention, the functions of each module / unit are as follows: The relationship graph construction module 201 is configured to collect stock entity data from the stock market within a historical period, wherein the stock entity data includes the entity itself and entity dimensions, generate a state vector of the stock entity data using the entity itself and the entity dimensions, and construct an interaction relationship graph between each stock entity data using the state vector; The feature recognition module 202 is configured to query the neighboring nodes of each entity node in the interaction relationship graph, and use the neighboring nodes to recognize the spatiotemporal features of the entity node in the interaction relationship graph; The sensitivity analysis module 203 is configured to calculate the synchronization error between each of the entity nodes according to the spatiotemporal characteristics, identify the real-time key nodes in the interaction relationship graph, and analyze the sensitivity of the real-time key nodes to disturbances in the interaction relationship graph; The Liapunov index calculation module 204 is configured to perform data dimensionality reduction processing on the state vector corresponding to each entity node in the interaction relationship graph to obtain a dimensionality reduction processing vector, and calculate the Lyapunov index of the interaction relationship graph using the dimensionality reduction processing vector; The stability analysis module 205 is configured to perform a stability analysis on the stock market based on the synchronization error, the disturbance sensitivity, and the Lyapunov index to obtain a stability analysis result.

[0065] In detail, each module in the stock market stability assessment system 200 combined with artificial intelligence in the embodiment of the present invention adopts the same Figure 1 The same technical means as the stock market stability assessment method combined with artificial intelligence described in , and can produce the same technical effects, will not be repeated here.

[0066] 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.

[0067] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for evaluating stock market stability using artificial intelligence, characterized in that: The method comprises: Collecting stock entity data from the stock market over a historical period, wherein the stock entity data includes the entity itself and entity dimensions, generating a state vector for the stock entity data using the entity itself and the entity dimensions, and constructing an interaction relationship graph between each stock entity data using the state vector; Querying neighboring nodes of each entity node in the interaction relationship graph, and using the neighboring nodes to identify spatiotemporal features of the entity node in the interaction relationship graph; Calculating synchronization errors between each of the entity nodes according to the spatiotemporal characteristics, identifying real-time key nodes in the interaction relationship graph, and analyzing disturbance sensitivity of the real-time key nodes in the interaction relationship graph; Performing data dimensionality reduction processing on the state vector corresponding to each entity node in the interaction relationship graph to obtain a dimensionality reduction processing vector, and calculating the Lyapunov index of the interaction relationship graph using the dimensionality reduction processing vector; Based on the synchronization error, the disturbance sensitivity and the Lyapunov index, a stability analysis is performed on the stock market to obtain a stability analysis result.

2. The method for evaluating stock market stability using artificial intelligence according to claim 1, wherein: The step of constructing an interaction relationship diagram between each stock entity data using the state vector includes: calculating a vector-vector weight between each state vector in the state vector; selecting a high numerical weight from the vector-vector weights; The high numerical weight is used as the edge weight; the state vector is used as the entity node; Based on the edge weights and the entity nodes, construct a weighted undirected graph between each entity itself; The weighted undirected graph is used as an interaction relationship graph.

3. The method for evaluating stock market stability using artificial intelligence according to claim 1, wherein: The querying of the neighboring nodes of each entity node in the interaction relationship graph includes: Querying all target nodes in the interaction relationship graph that are connected to the entity node; Identifying the number of shortest path edges between the entity node and the target node; According to the number of edges in the shortest path, the target node is divided into nodes to obtain neighboring nodes.

4. The method for evaluating stock market stability using artificial intelligence according to claim 1, wherein: The using the neighborhood nodes to identify the spatiotemporal features of the entity node in the interaction relationship graph includes: Identifying, based on the neighborhood nodes, spatial structural features of the entity node in the interaction relationship graph; Integrating the time information of the entity node into the spatial structure feature to obtain a fusion feature; The fusion feature is used as the spatiotemporal feature of the entity node in the interaction relationship graph.

5. The method for evaluating stock market stability in combination with artificial intelligence according to claim 1, wherein: The calculating the synchronization error between each entity node in the entity nodes according to the spatiotemporal characteristics includes: Calculating the cosine similarity between each entity node in the entity nodes according to the spatiotemporal features; Based on the cosine similarity, a synchronization error between each of the entity nodes is calculated.

6. The method for evaluating stock market stability in combination with artificial intelligence according to claim 1, wherein: The identifying of real-time key nodes in the interaction relationship diagram includes: Calculating the betweenness centrality of each entity node in the interaction relationship graph; The betweenness centrality is normalized to obtain the standard centrality; Calculating a critical judgment threshold of the standard centrality; When the standard centrality is greater than the key judgment threshold, the central node corresponding to the standard centrality is used as a real-time key node.

7. The method for evaluating stock market stability using artificial intelligence according to claim 1, wherein: The analyzing the disturbance sensitivity of the real-time key node in the interaction relationship diagram includes: Obtaining the betweenness centrality corresponding to the real-time key node; Obtaining a global edge weight matrix in the interaction relationship graph; Calculating the centrality gradient of the betweenness centrality with respect to the global edge weight matrix; Based on the centrality gradient, the disturbance sensitivity of the real-time key node in the interaction relationship graph is calculated using the following formula: ； in, represents the sensitivity to disturbance, express About the Global Edge Weight Matrix The centrality gradient of represents the Frobenius norm, Indicates the The betweenness centrality of real-time key nodes, Indicates the number of real-time key nodes in the interaction relationship diagram.

8. The method for evaluating stock market stability in combination with artificial intelligence according to claim 1, wherein: The calculating the Lyapunov index of the interaction relationship graph by using the dimension reduction processing vector includes: The differential change of the dimension reduction processing vector is calculated using the following formula: ； in, represents the differential change, Represents the first differential change corresponding to the dimensionality reduction processing vector, Represents the second differential change corresponding to the dimensionality reduction processing vector, Represents the third differential change corresponding to the dimensionality reduction processing vector, represents the linear response parameter, represents the driving force parameter, represents the dissipation parameter; Splice the differential changes corresponding to all entity nodes in the interaction relationship graph to obtain the spliced ​​change; The Jacobian matrix of the splicing variation is calculated using the following formula: ； 1 ； in, represents the Jacobian matrix, 、 Indicates the splicing change, represents the time variable, represents a differential equation function; According to the Jacobian matrix, the Lyapunov index of the interaction relationship graph is calculated using the following formula: ； ； ； in, represents the Lyapunov exponent, express The growth rate of time, represents the time window, Express The value of the first row and first column in the upper triangular matrix obtained by orthogonal triangular decomposition, represents the differential time change, Express The upper triangular matrix obtained by orthogonal triangular decomposition is Express The new orthogonal basis obtained by orthogonal triangular decomposition is represents the old orthogonal basis, express The Jacobian matrix of .

9. The method for evaluating stock market stability in combination with artificial intelligence according to claim 1, wherein: The performing of stability analysis on the stock market based on the synchronization error, the disturbance sensitivity, and the Lyapunov index to obtain a stability analysis result includes: Identifying a positive or negative correlation between the synchronization error, the disturbance sensitivity, the Lyapunov index, and the stability of the stock market; Determining the weight signs of the synchronization error, the disturbance sensitivity, and the Lyapunov exponent according to the positive and negative correlation; Normalizing the synchronization error, the disturbance sensitivity, and the Lyapunov exponent to obtain normalized parameters; Calculating a weight parameter of the normalization parameter according to the positive and negative signs of the weights; Performing weighted sum processing on the normalized parameters using the weight parameters to obtain a weighted sum result; Performing nonlinear mapping on the weighted summation result using a preset hyperbolic tangent function to obtain a nonlinear mapping result; A stability analysis is performed on the stock market based on the stability score interval corresponding to the nonlinear mapping result to obtain a stability analysis result.

10. A stock market stability assessment system combined with artificial intelligence, characterized in that: The system comprises: A relationship graph construction module is configured to collect stock entity data from the stock market over a historical period, wherein the stock entity data includes the entity itself and entity dimensions, generate a state vector for the stock entity data using the entity itself and the entity dimensions, and use the state vector to construct an interaction relationship graph between each stock entity data; A feature recognition module, configured to query neighboring nodes of each entity node in the interaction relationship graph, and use the neighboring nodes to identify spatiotemporal features of the entity node in the interaction relationship graph; a sensitivity analysis module, configured to calculate the synchronization error between each of the entity nodes according to the spatiotemporal characteristics, identify the real-time key nodes in the interaction relationship graph, and analyze the sensitivity of the real-time key nodes to disturbances in the interaction relationship graph; A Liapunov index calculation module is used to perform data dimensionality reduction processing on the state vector corresponding to each entity node in the interaction relationship graph to obtain a dimensionality reduction processing vector, and use the dimensionality reduction processing vector to calculate the Lyapunov index of the interaction relationship graph; The stability analysis module is used to perform stability analysis on the stock market based on the synchronization error, the disturbance sensitivity and the Lyapunov index to obtain a stability analysis result.

Images