Dynamic simulation matching method and system under multiple strategies

Abstract

The invention discloses a dynamic simulation matching method and system under multiple strategies, and relates to the field of financial data processing, and the method comprises the steps: responding to the received multi-source heterogeneous data, carrying out the data processing of the multi-source heterogeneous data, and obtaining the processed multi-source heterogeneous data; inputting the processed multi-source heterogeneous data into a dynamic strategy evaluation network to generate a candidate matching strategy set; executing strategy rehearsal on the candidate matching strategy set in a virtual order book mirror image generated by a strategy efficiency simulator so as to calculate an expected matching efficiency index of each candidate matching strategy; a dynamic simulation strategy is adopted, and an optimal candidate matching strategy is determined according to the expected matching efficiency index of each candidate matching strategy; and loading the optimal candidate matching strategy to a self-adaptive matching engine so as to carry out order matching. According to the invention, the adaptability of the matching system to the market can be effectively improved in a complex transaction scene, and the matching efficiency is improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of financial data processing, and in particular to a dynamic simulation matching method and system under multiple strategies. Background Art

[0002] Financial transaction matching is the process of automatically matching orders between buyers and sellers in the financial market, and determining the transaction price, quantity and time through specific rules and algorithms to form an effective transaction. It is the core function of electronic trading systems such as securities, futures and foreign exchange, and directly affects market liquidity, price discovery efficiency and transaction fairness.

[0003] The existing mainstream matching system is based on a static rule engine, which usually adopts a single matching strategy (such as price priority, time priority), which can maintain efficiency in a stable market environment. However, with the popularization of high-frequency trading, the complexity of derivatives markets, and multi-asset linkage trading scenarios, a single matching strategy cannot perceive sudden changes in market conditions, such as a sudden drop in liquidity, the emergence of cross-market arbitrage opportunities, etc., resulting in a large number of orders piling up or invalid transactions. There are cases showing that when a single matching strategy coexists with mixed order types (limit orders / market orders / iceberg orders), the best price achievement rate drops by 37%. This makes the current mainstream matching system less adaptable to the market, which in turn causes a decrease in matching efficiency. Summary of the invention

[0004] The embodiments of the present application provide a dynamic simulation matching method and system under multiple strategies, which are used to effectively improve the adaptability of the matching system to the market and improve the matching efficiency in complex trading scenarios.

[0005] To achieve the above objectives, the embodiments of the present application adopt the following technical solutions: In a first aspect, a dynamic simulation matching method under multiple strategies is provided, which is applied to a financial transaction system, wherein the financial transaction system is deployed with an adaptive matching engine, and the method comprises: In response to receiving multi-source heterogeneous data, respectively processing the multi-source heterogeneous data to obtain processed multi-source heterogeneous data, wherein the multi-source heterogeneous data includes market data, public opinion data, and arbitrage channel detection data; Inputting the processed multi-source heterogeneous data into a dynamic strategy evaluation network to generate a candidate matching strategy set, wherein the dynamic strategy evaluation network is implemented based on a pre-built hybrid prediction model and a strategy effectiveness simulator, and the candidate matching strategy set includes at least one candidate matching strategy; Performing strategy preview on the candidate matching strategy set in the virtual order book image generated by the strategy effectiveness simulator to calculate the expected matching efficiency index of each candidate matching strategy; Adopting a dynamic simulation strategy to determine the optimal candidate matching strategy according to the expected matching efficiency index of each candidate matching strategy; The optimal candidate matching strategy is loaded into the adaptive matching engine for order matching.

[0006] In a possible implementation manner of the first aspect, the financial transaction system includes an FPGA coprocessor, and the data processing is performed on the multi-source heterogeneous data respectively to obtain processed multi-source heterogeneous data, including: Using the FPGA coprocessor to classify the market data, and obtain order book and transaction detail data; The FPGA coprocessor is used to parse the public opinion data to obtain a real-time news stream, and an event impact coefficient matrix is ​​generated based on the real-time news stream; Using the FPGA coprocessor to perform cointegration analysis on the relevant asset price series in the arbitrage channel detection data, and calculate the price difference Z-score value; Among them, the order book, the transaction details data, the event impact coefficient matrix and the price difference Z-score value are all processed multi-source heterogeneous data.

[0007] In another possible implementation manner of the first aspect, the method further includes: When the average value of the price difference Z-score value is greater than a preset threshold, an arbitrage signal is generated.

[0008] In another possible implementation of the first aspect, the hybrid prediction model includes an LSTM neural network, a graph convolutional network, and a reinforcement learning agent module, the LSTM neural network is used to process the transaction detail data and output time series features, the graph convolutional network is used to analyze the topological structure of the order book and output a node feature vector, and the reinforcement learning agent module is used to generate a strategy space based on the joint features; Among them, the joint feature is obtained by tensor splicing between the temporal feature output by the LSTM neural network and the node feature vector output by the graph convolutional network.

[0009] In another possible implementation of the first aspect, inputting the processed multi-source heterogeneous data into a dynamic strategy evaluation network to generate a candidate matching strategy set includes: Inputting the order book and the transaction details data into the hybrid prediction model to obtain a strategy space, wherein the strategy space includes all matching strategies; The strategy effectiveness simulator is used to output the expected profit and loss matrix of each matching strategy in the strategy space based on the event impact coefficient matrix and the price difference Z-score value, and determine the candidate matching strategy set in the strategy space according to the expected profit and loss matrix.

[0010] In another possible implementation of the first aspect, the strategy effectiveness simulator outputs an expected profit and loss matrix of each matching strategy in the strategy space based on the event impact coefficient matrix and the price difference Z-score value, and determines a candidate matching strategy set in the strategy space according to the expected profit and loss matrix, including: The strategy performance simulator creates a virtual order book mirror that is synchronized with the order book; The strategy effectiveness simulator adopts a preset scenario algorithm to generate a ±3σ price fluctuation scenario and a liquidity drop simulation scenario based on the event impact coefficient matrix and the price difference Z-score value; The strategy effectiveness simulator adopts a preset genetic algorithm to simulate each matching strategy in the strategy space in the ±3σ price fluctuation scenario and the liquidity drop simulation scenario, and outputs the expected profit and loss matrix of each matching strategy; According to the expected profit and loss matrix, the expected profit and risk ratio of each matching strategy is calculated using a preset profit and risk ratio calculation formula, and each matching strategy is sorted from high to low according to the profit and risk ratio; The top N matching strategies in terms of return-risk ratio are taken as the candidate matching strategy set, where N is a positive integer greater than 2.

[0011] In another possible implementation manner of the first aspect, performing a strategy preview on the candidate matching strategy set in a virtual order book image generated by the strategy effectiveness simulator to calculate an expected matching efficiency index of each candidate matching strategy includes: In the virtual order book image, each candidate matching strategy is executed respectively according to a preset time window; Recording the matching success rate, matching delay time and matching cost of each candidate matching strategy within the time window; According to the matching success rate, matching delay time and matching cost, a weighted calculation is performed to obtain an expected matching efficiency index for each candidate matching strategy.

[0012] In another possible implementation manner of the first aspect, the dynamic simulation strategy includes: Based on the expected matching efficiency index of each candidate matching strategy, construct an objective function, wherein the objective function includes maximizing matching success rate, minimizing matching delay time, and minimizing matching cost; The Pareto optimal solution algorithm is used to solve the objective function and obtain the optimal candidate matching strategy.

[0013] In another possible implementation manner of the first aspect, loading the optimal candidate matching strategy into the adaptive matching engine to perform order matching includes: Loading the optimal candidate matching strategy into the adaptive matching engine; The adaptive matching engine is used to monitor market data and order book changes in real time according to the optimal candidate matching strategy; The adaptive matching engine is used to monitor the changes in market data in real time according to the optimal candidate matching strategy, and automatically perform matching operations to complete order matching when an order that meets the optimal candidate matching strategy is detected.

[0014] In a second aspect, the present application provides a financial transaction system, comprising: A memory is configured to store instructions; and a processor is configured to call the instructions from the memory and implement the above-mentioned dynamic simulation matching method under multiple strategies when executing the instructions.

[0015] Through the above technical solution, multi-source heterogeneous data are processed separately to generate processed multi-source heterogeneous data. The processed multi-source heterogeneous data are analyzed by the dynamic strategy evaluation network to generate a set of candidate matching strategies. The dynamic strategy evaluation network is based on a pre-built hybrid prediction model and a strategy effectiveness simulator, which can comprehensively consider the influence of market status, trading behavior and external factors to generate a variety of candidate matching strategies; the candidate matching strategies are previewed in the virtual order book image generated by the strategy effectiveness simulator, and the expected matching efficiency index of each candidate matching strategy is calculated by simulating the actual trading environment, which can not only evaluate the strategy under the current market conditions, but also It can not only predict the performance of a strategy under certain conditions, but also predict its adaptability in future market changes; adopt a dynamic simulation strategy to determine the optimal candidate matching strategy based on the expected matching efficiency index of each candidate matching strategy. It can ensure that the system can respond quickly to market changes by adjusting the strategy selection in real time to avoid the decline in efficiency caused by strategy rigidity; finally, load the optimal candidate matching strategy into the adaptive matching engine for order matching. The adaptive matching engine can dynamically adjust the matching rules according to the real-time market data and the feedback from the strategy effectiveness simulator to ensure the efficiency and fairness of the matching process, and effectively solve the adaptability problem of a single matching strategy in the event of market mutations.

[0016] The combination of the dynamic strategy evaluation network and the strategy effectiveness simulator has significantly improved the matching efficiency, especially in the case of mixed order types, the optimal price achievement rate has been significantly improved, avoiding the problems of order accumulation and invalid transactions. The comprehensive processing of multi-source heterogeneous data has enhanced the perception of market status. The fusion analysis of market data, public opinion data and arbitrage channel detection data can more accurately predict market changes and adjust matching strategies in advance, so as to maintain efficient matching capabilities when liquidity drops sharply or arbitrage opportunities emerge. In addition, the application of the strategy effectiveness simulator can rehearse matching strategies in a virtual environment, discover potential problems in advance and optimize strategy selection, effectively reducing the risks in actual transactions. Finally, the dynamic adjustment capability of the adaptive matching engine ensures continuous and efficient operation under different market conditions. Whether it is a stable market or a volatile market, it can quickly adjust the matching rules based on real-time data and strategy effectiveness feedback to ensure the fairness and efficiency of the matching process.

[0017] Other features and advantages of the embodiments of the present application will be described in detail in the subsequent specific implementation section. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A schematic diagram of a flow chart of a dynamic simulation matchmaking method under multiple strategies provided in an embodiment of the present application; Figure 2 A schematic diagram of the structure of a financial transaction system provided in an embodiment of the present application; Figure 3 A schematic diagram of the structure of a hybrid prediction model provided in an embodiment of the present application. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. It should be understood that the specific implementation methods described herein are only used to illustrate and explain the embodiments of the present application, and are not used to limit the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0020] It should be noted that if the embodiments of the present application involve directional indications (such as up, down, left, right, front, back...), such directional indications are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0021] In addition, if there are descriptions involving "first", "second", etc. in the embodiments of the present application, the descriptions of "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or suggesting their relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in the field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such combination of technical solutions does not exist and is not within the scope of protection required by this application.

[0022] Figure 1 The following schematically shows a flow chart of a dynamic simulation matchmaking method under multiple strategies according to an embodiment of the present application. Figure 1 As shown, an embodiment of the present application provides a dynamic simulation matching method under multiple strategies, which is applied to a financial transaction system. The financial transaction system is deployed with an adaptive matching engine. The method may include the following steps.

[0023] S110, in response to receiving multi-source heterogeneous data, respectively processing the multi-source heterogeneous data to obtain processed multi-source heterogeneous data, where the multi-source heterogeneous data includes market data, public opinion data, and arbitrage channel detection data; S120, inputting the processed multi-source heterogeneous data into a dynamic strategy evaluation network to generate a candidate matching strategy set, wherein the dynamic strategy evaluation network is implemented based on a pre-built hybrid prediction model and a strategy effectiveness simulator, and the candidate matching strategy set includes at least one candidate matching strategy; S130, performing strategy preview on the candidate matching strategy set in the virtual order book image generated by the strategy effectiveness simulator to calculate the expected matching efficiency index of each candidate matching strategy; S140, using a dynamic simulation strategy to determine the optimal candidate matching strategy based on the expected matching efficiency index of each candidate matching strategy; S150: Load the optimal candidate matching strategy into the adaptive matching engine for order matching.

[0024] Figure 2 A schematic diagram of the structure of a financial transaction system provided by an embodiment of the present application is shown. Figure 2 In the financial trading system shown, multi-source heterogeneous data include market data, public opinion data, and arbitrage channel detection data. Market data usually includes order books, transaction details, price fluctuations and other information, which are used to reflect the real-time status and trading behavior of the market. Public opinion data can include news, social media comments, analyst reports, etc., which are used to reflect the emotions and expectations of market participants. Arbitrage channel detection data involves price differences and correlations between different markets or assets, which are used to identify potential arbitrage opportunities.

[0025] Specifically, market data is classified by an FPGA coprocessor, and order books and transaction details are extracted to reflect the supply and demand relationship and trading activity of the market. Public opinion data is parsed by natural language processing technology to generate real-time news streams, and an event impact coefficient matrix is ​​generated based on sentiment analysis and event recognition technology, which can quantify the impact of different events on the market. Arbitrage channel detection data is processed by cointegration analysis technology to calculate the price difference Z-score value of the relevant asset price series, which can reflect the price deviation between assets and the significance of arbitrage opportunities. In this embodiment, during the data processing process, the data needs to be cleaned and standardized to remove noise and outliers to ensure the quality and consistency of the data. Finally, the processed multi-source heterogeneous data includes order books, transaction details, event impact coefficient matrix and price difference Z-score value.

[0026] The processed multi-source heterogeneous data is input into the dynamic strategy evaluation network to generate a set of candidate matching strategies. The dynamic strategy evaluation network is implemented based on a pre-built hybrid prediction model and a strategy effectiveness simulator, which can comprehensively consider the influence of market status, trading behavior and external factors to generate a variety of candidate matching strategies. The hybrid prediction model includes LSTM neural network, graph convolution network and reinforcement learning agent module, which can process different types of multi-source heterogeneous data and generate joint features respectively. LSTM neural network is used to process transaction details data and extract time series features to reflect the dynamic changes of the market and the trend of trading behavior. Graph convolution network is used to analyze the topological structure of the order book and generate node feature vectors to reflect the supply and demand relationship and trading activity of the market. The reinforcement learning agent module generates a strategy space based on the joint features, which includes all possible matching strategies.

[0027] The strategy performance simulator can output the expected profit and loss matrix of each matching strategy in the strategy space based on the event impact coefficient matrix and the spread Z-score value, and determine the candidate matching strategy set based on the matrix. The strategy performance simulator first creates a virtual order book mirror synchronized with the order book, and then uses the preset scenario algorithm to generate ±3σ price fluctuation scenarios and liquidity drop simulation scenarios, which can simulate the performance of the market under different conditions. In the simulation scenario, the strategy performance simulator uses a preset genetic algorithm to simulate the performance of each matching strategy and outputs the expected profit and loss matrix. According to the expected profit and loss matrix, the preset return-risk ratio calculation formula is used to calculate the expected return-risk ratio of each matching strategy, and the matching strategies are sorted from high to low according to the return-risk ratio. Finally, the matching strategies with the top N return-risk ratios are selected as the candidate matching strategy set.

[0028] The candidate matching strategy set performs strategy preview in the virtual order book image generated by the strategy performance simulator to calculate the expected matching efficiency index of each candidate matching strategy. The virtual order book image is synchronized with the real order book and can simulate the real-time status and trading behavior of the market.

[0029] During the strategy preview, each candidate matching strategy is executed according to the preset time window, and the matching success rate, matching delay time and matching cost of each strategy within the time window are recorded. Matching success rate refers to the proportion of orders successfully matched by the strategy within the time window, reflecting the effectiveness of the strategy. Matching delay time refers to the time from receiving an order to completing matching, reflecting the response speed of the strategy. Matching cost refers to the transaction cost incurred during the execution of the strategy, reflecting the economy of the strategy. According to the matching success rate, matching delay time and matching cost, the weighted calculation method is used to obtain the expected matching efficiency index of each candidate matching strategy. The weighted calculation method can comprehensively consider the importance of different indicators to ensure the comprehensiveness and accuracy of the evaluation results. Through strategy preview, the performance of each candidate matching strategy under current market conditions can be evaluated, and its adaptability to future market changes can be predicted.

[0030] After obtaining the expected matching efficiency index of each candidate matching strategy, the dynamic simulation strategy is used to determine the optimal candidate matching strategy. The dynamic simulation strategy constructs an objective function based on the expected matching efficiency index of each candidate matching strategy, which includes maximizing the matching success rate, minimizing the matching delay time, and minimizing the matching cost. The objective function can comprehensively consider the importance of different indicators to ensure the balance of the optimal matching strategy in terms of effectiveness, response speed, and economy. The Pareto optimal solution algorithm is used to solve the objective function and obtain the optimal candidate matching strategy. The Pareto optimal solution algorithm can find the optimal solution in multi-objective optimization problems to ensure that the optimal matching strategy can achieve the best performance in all aspects. Through the dynamic simulation strategy, the strategy selection can be adjusted in real time to ensure that the system can respond quickly to market changes and avoid the decline in efficiency caused by strategy rigidity.

[0031] The best candidate matching strategy is loaded into the adaptive matching engine for order matching. The adaptive matching engine can monitor market data and order book changes in real time according to the best candidate matching strategy, and automatically perform matching operations and complete order matching when an order that meets the best candidate matching strategy is detected. The dynamic adjustment capability of the adaptive matching engine ensures continuous and efficient operation under different market conditions. Whether it is a stable market or a volatile market, it can quickly adjust the matching rules based on real-time data and strategy performance feedback to ensure the fairness and efficiency of the matching process.

[0032] In this embodiment, data processing is performed on multi-source heterogeneous data respectively to generate processed multi-source heterogeneous data. The processed multi-source heterogeneous data is analyzed by a dynamic strategy evaluation network to generate a set of candidate matching strategies. The dynamic strategy evaluation network is based on a pre-built hybrid prediction model and a strategy effectiveness simulator, which can comprehensively consider the influence of market status, trading behavior and external factors to generate a variety of candidate matching strategies; the candidate matching strategies are previewed in a virtual order book image generated by the strategy effectiveness simulator, and the expected matching efficiency index of each candidate matching strategy is calculated by simulating the actual trading environment, which can not only evaluate the strategy under the current market conditions The system can not only predict the performance of the future market, but also predict its adaptability in future market changes; adopt a dynamic simulation strategy, determine the optimal candidate matching strategy according to the expected matching efficiency index of each candidate matching strategy, and adjust the strategy selection in real time to ensure that the system can respond quickly to market changes and avoid the decline in efficiency caused by strategy rigidity; finally, load the optimal candidate matching strategy into the adaptive matching engine for order matching. The adaptive matching engine can dynamically adjust the matching rules according to the real-time market data and the feedback from the strategy performance simulator to ensure the efficiency and fairness of the matching process, and effectively solve the adaptability problem of a single matching strategy in the event of market mutations.

[0033] The combination of the dynamic strategy evaluation network and the strategy effectiveness simulator has significantly improved the matching efficiency, especially in the case of mixed order types, the optimal price achievement rate has been significantly improved, avoiding the problems of order accumulation and invalid transactions. The comprehensive processing of multi-source heterogeneous data has enhanced the perception of market status. The fusion analysis of market data, public opinion data and arbitrage channel detection data can more accurately predict market changes and adjust matching strategies in advance, so as to maintain efficient matching capabilities when liquidity drops sharply or arbitrage opportunities emerge. In addition, the application of the strategy effectiveness simulator can rehearse matching strategies in a virtual environment, discover potential problems in advance and optimize strategy selection, effectively reducing the risks in actual transactions. Finally, the dynamic adjustment capability of the adaptive matching engine ensures continuous and efficient operation under different market conditions. Whether it is a stable market or a volatile market, it can quickly adjust the matching rules based on real-time data and strategy effectiveness feedback to ensure the fairness and efficiency of the matching process.

[0034] In one implementation of this embodiment, the financial transaction system includes an FPGA coprocessor, which processes multi-source heterogeneous data respectively to obtain processed multi-source heterogeneous data, including the following steps: S210, using the FPGA coprocessor to classify market data and obtain order book and transaction details data; S220, using an FPGA coprocessor to parse public opinion data to obtain a real-time news stream, and generating an event impact coefficient matrix based on the real-time news stream; S230, using an FPGA coprocessor to perform cointegration analysis on the price series of related assets in the arbitrage channel detection data, and calculate the price difference Z-score value; Among them, the order book, transaction details data, event impact coefficient matrix and price difference Z-score value are all processed multi-source heterogeneous data.

[0035] In this embodiment, in order to process data efficiently, an FPGA coprocessor is used for classification and analysis. FPGA (Field Programmable Gate Array) is a high-performance hardware accelerator that can process a large amount of data in parallel, significantly improving the speed and efficiency of data processing.

[0036] Market data is first classified by the FPGA coprocessor to extract order book and transaction details. Order book data includes the quotes and quantities of buyers and sellers, reflecting the supply and demand relationship and trading activity in the market. Transaction details include the price, quantity and time of each transaction, reflecting the market's trading behavior and price fluctuations. Through parallel computing and pipeline technology, the FPGA coprocessor can quickly process large amounts of market data to ensure the real-time and accuracy of the data.

[0037] Public opinion data is an important external data source in financial trading systems, including news, social media comments, analyst reports, etc., which are used to reflect the emotions and expectations of market participants and have a significant impact on market status and trading behavior. In order to efficiently process this data, FPGA coprocessors are also used for parsing and classification. Public opinion data is first parsed by the FPGA coprocessor to extract real-time news streams. Real-time news streams include news titles, text, release time and other information, which can reflect the latest market trends and changes in participants' emotions. Through natural language processing technology, FPGA coprocessors can quickly parse large amounts of public opinion data to ensure the real-time and accuracy of the data.

[0038] Based on the real-time news stream, the FPGA coprocessor can generate an event impact coefficient matrix. The event impact coefficient matrix can quantify the impact of different events on the market and is an important reference for matching strategy generation and optimization. The generation process of the event impact coefficient matrix includes event identification, sentiment analysis, and impact assessment. Event identification uses keyword matching and pattern recognition technology to identify important events in the news stream. Sentiment analysis uses sentiment dictionaries and machine learning algorithms to evaluate the emotional impact of events on market participants. Impact assessment quantifies the impact of events on market status and trading behavior through historical data analysis and statistical models.

[0039] Arbitrage channel detection data involves price differences and correlations between different markets or assets. This embodiment also uses an FPGA coprocessor for cointegration analysis. The arbitrage channel detection data is first preprocessed by the FPGA coprocessor to extract the relevant asset price series. The relevant asset price series includes price data of different markets or assets, reflecting the price differences and correlations between assets. The FPGA coprocessor can quickly process a large amount of arbitrage channel detection data through parallel computing and pipeline technology to ensure the real-time and accuracy of the data.

[0040] Based on the price series of related assets, the FPGA coprocessor performs cointegration analysis and calculates the Z-score value of the price difference. Cointegration analysis is a statistical method used to detect the long-term equilibrium relationship between asset price series. The Z-score value of the price difference can reflect the degree of price deviation between assets and the significance of arbitrage opportunities. The process of cointegration analysis includes cointegration test, price difference calculation and Z-score value calculation. The cointegration test detects the cointegration relationship between asset price series through unit root test and cointegration regression. The price difference calculation calculates the price difference between asset price series through linear combination and difference operation. The Z-score value calculation calculates the Z-score value of the price difference through standardization and statistical model.

[0041] This implementation method uses an FPGA coprocessor to classify, parse and cointegrate multi-source heterogeneous data, which significantly improves the speed and efficiency of data processing, ensures the real-time and accuracy of multi-source heterogeneous data, and provides reliable data support for the generation and optimization of matching strategies, thereby significantly improving the matching efficiency, ensuring the efficiency and fairness of the matching process, and effectively solving the adaptability problem of a single matching strategy when the market changes suddenly.

[0042] In one implementation of this embodiment, the following steps are also included: S310: When the average value of the price difference Z-score value is greater than a preset threshold, generate an arbitrage signal.

[0043] The spread Z-score is an important indicator for identifying arbitrage opportunities in financial trading systems. It reflects the degree of deviation between related asset price series and the significance of arbitrage opportunities. In order to effectively identify and utilize arbitrage opportunities, it is necessary to generate arbitrage signals when the average value of the spread Z-score is greater than the preset threshold. The calculation of the spread Z-score is based on cointegration analysis. By detecting the long-term equilibrium relationship between asset price series, calculating the spread between asset price series, and normalizing the spread, the Z-score is obtained.

[0044] The calculation formula of Z-score value is: Z = (X - μ) / σ; Where X is the spread, μ is the mean of the spread, and σ is the standard deviation of the spread. The Z-score value can reflect the degree of deviation of the spread from its historical mean and standard deviation. The larger the Z-score value, the greater the deviation of the spread from the historical mean, and the higher the significance of the arbitrage opportunity. In the process of calculating the Z-score value of the spread, it is necessary to first perform a cointegration test on the relevant asset price series to ensure that there is a long-term equilibrium relationship between the asset price series. The cointegration test detects the cointegration relationship between the asset price series through unit root test and cointegration regression. If there is a cointegration relationship between the asset price series, the spread between the asset price series is calculated through linear combination and difference operation. Then, the spread is standardized and the Z-score value is calculated. After obtaining the Z-score value of the spread, its average value needs to be calculated and compared with the preset threshold. The preset threshold is set based on historical data and market experience to determine the significance of arbitrage opportunities. If the average value of the Z-score value of the spread is greater than the preset threshold, an arbitrage signal is generated. The signal release of arbitrage signals is sent to traders and automatic trading systems through trading systems and communication networks so that arbitrage transactions can be executed in a timely manner.

[0045] This implementation can effectively identify and utilize arbitrage opportunities, improve the profitability and risk control capabilities of the trading system, significantly improve the efficiency of identifying and utilizing arbitrage opportunities, ensure the reliability and significance of arbitrage signals, and provide important support for the profitability and risk control capabilities of the trading system.

[0046] In one implementation of this embodiment, the hybrid prediction model includes an LSTM neural network, a graph convolutional network, and a reinforcement learning agent module, wherein the LSTM neural network is used to process transaction detail data and output time series features, the graph convolutional network is used to analyze the topological structure of the order book and output a node feature vector, and the reinforcement learning agent module is used to generate a strategy space based on the joint features; Among them, the joint features are obtained by tensor splicing between the temporal features output by the LSTM neural network and the node feature vectors output by the graph convolutional network.

[0047] Figure 3 A structural schematic diagram of a hybrid prediction model provided by an embodiment of the present application is shown, wherein the hybrid prediction model is the core component of the dynamic strategy evaluation network, which can comprehensively consider the influence of market status, trading behavior and external factors to generate a variety of candidate matching strategies. The hybrid prediction model includes an LSTM neural network, a graph convolutional network and a reinforcement learning agent module, which respectively process different types of multi-source heterogeneous data and generate joint features to provide a basis for the generation of the strategy space.

[0048] LSTM neural network (Long Short-Term Memory Network) is a deep learning model specially designed for processing time series data, which can capture long-term dependencies in data. In the hybrid prediction model, LSTM neural network is used to process transaction details data and extract time series features. Transaction details data includes the price, quantity and time of each transaction, reflecting the market's trading behavior and price fluctuations. Through its unique memory unit and gating mechanism, LSTM neural network can effectively capture the time series patterns and trends in transaction details data and output time series features. Time series features can reflect the dynamic changes of the market and the trend of trading behavior, providing an important reference for subsequent strategy generation and optimization.

[0049] Graph convolutional network is a deep learning model specially designed for processing graph structure data, which can capture the topological relationship and node features in the data. In the hybrid prediction model, the graph convolutional network is used to analyze the topological structure of the order book and generate node feature vectors. The order book data includes the quotes and quantities of buyers and sellers, reflecting the supply and demand relationship and trading activity in the market. Through its unique convolution and pooling operations, the graph convolutional network can effectively analyze the topological structure of the order book and extract the node feature vector. The node feature vector can reflect the supply and demand relationship and trading activity in the market, providing an important reference for subsequent strategy generation and optimization. The reinforcement learning agent module is a machine learning model based on a reward mechanism, which can learn the optimal strategy by interacting with the environment.

[0050] In the hybrid prediction model, the reinforcement learning agent module generates a policy space based on joint features. Joint features are obtained by tensor splicing between the time series features output by the LSTM neural network and the node feature vectors output by the graph convolutional network. Tensor splicing is an operation that connects multiple feature vectors in a specific dimension, which can integrate features from different sources into a unified feature representation. Joint features can comprehensively reflect the dynamic changes, trading behaviors and supply and demand relationships of the market, and provide comprehensive information for the generation of policy space. The reinforcement learning agent module learns the optimal strategy and generates a policy space by interacting with the environment. The policy space includes all possible matching strategies, providing rich choices for subsequent strategy preview and optimization.

[0051] This implementation method, through the LSTM neural network, graph convolutional network and reinforcement learning agent module in the hybrid prediction model, can comprehensively consider the influence of market status, trading behavior and external factors, generate a variety of candidate matching strategies, significantly improve the diversity and accuracy of strategy generation, ensure the comprehensiveness and reliability of the strategy space, effectively improve the generation and optimization efficiency of matching strategies, and ensure the efficiency and fairness of the matching process.

[0052] In one implementation of this embodiment, the processed multi-source heterogeneous data is input into a dynamic strategy evaluation network to generate a candidate matching strategy set, including the following steps: S510, inputting the order book and transaction details data into the hybrid prediction model to obtain a strategy space, which includes all matching strategies; S520, the strategy effectiveness simulator is used to output the expected profit and loss matrix of each matching strategy in the strategy space based on the event impact coefficient matrix and the price difference Z-score value, and determine the candidate matching strategy set in the strategy space according to the expected profit and loss matrix.

[0053] In this embodiment, the LSTM neural network in the hybrid prediction model is used to process the transaction details data and extract the time series features. The time series features can reflect the dynamic changes of the market and the trend of trading behavior, and provide an important reference for strategy generation. The graph convolutional network is used to analyze the topological structure of the order book and generate node feature vectors. The node feature vector can reflect the supply and demand relationship and trading activity in the market, and provide an important reference for strategy generation. The reinforcement learning agent module generates a strategy space based on the joint features. The joint features are obtained by integrating the time series features and the node feature vectors through tensor splicing, which can comprehensively reflect the dynamic changes of the market, trading behavior and supply and demand. The strategy space includes all possible matching strategies.

[0054] The strategy effectiveness simulator is an important part of the dynamic strategy evaluation network, which is used to evaluate the performance of each matching strategy and determine the candidate matching strategy set. The strategy effectiveness simulator outputs the expected profit and loss matrix of each matching strategy in the strategy space based on the event impact coefficient matrix and the spread Z-score value.

[0055] Among them, the event impact coefficient matrix can quantify the impact of different events on the market, and is an important reference for the evaluation of matching strategies. The spread Z-score value can reflect the price deviation between assets and the significance of arbitrage opportunities, and is an important reference for the evaluation of matching strategies.

[0056] Specifically, the strategy effectiveness simulator first creates a virtual order book mirror that is synchronized with the order book. The virtual order book mirror can simulate the real-time status and trading behavior of the market. Then, the strategy effectiveness simulator uses a preset scenario algorithm to generate a ±3σ price fluctuation scenario and a liquidity drop simulation scenario based on the event impact coefficient matrix and the spread Z-score value. The ±3σ price fluctuation scenario can simulate the market's performance under extreme price fluctuations, and the liquidity drop simulation scenario can simulate the market's performance under conditions of a sharp drop in liquidity.

[0057] In the simulation scenario, the strategy performance simulator uses a preset genetic algorithm to simulate the performance of each matching strategy and output an expected profit and loss matrix. The expected profit and loss matrix can reflect the expected return and risk of each matching strategy under different market conditions and is an important reference for matching strategy evaluation. According to the expected profit and loss matrix, the preset return-risk ratio calculation formula is used to calculate the expected return-risk ratio of each matching strategy, and the matching strategies are sorted from high to low according to the return-risk ratio.

[0058] The formula for calculating the risk-return ratio is: Return-risk ratio = expected return / expected risk; The expected return and expected risk are extracted from the expected profit and loss matrix respectively. Finally, the top N matching strategies with the highest return-risk ratio are selected as the candidate matching strategy set, where N is a positive integer greater than 2.

[0059] This implementation method generates a strategy space by inputting the order book and transaction details data into a hybrid prediction model, which significantly improves the diversity and accuracy of strategy generation and ensures the comprehensiveness and reliability of the strategy space. The strategy space includes all possible matching strategies, providing a wealth of options for subsequent strategy previews and optimizations. Through the strategy effectiveness simulator, based on the event impact coefficient matrix and the spread Z-score value, the expected profit and loss matrix of each matching strategy is output, and the candidate matching strategy set is determined based on the expected profit and loss matrix, which significantly improves the accuracy and reliability of strategy evaluation and ensures the diversity and effectiveness of the candidate matching strategy set. Through this technical solution, it is possible to comprehensively consider the influence of market conditions, trading behaviors and external factors, generate diversified matching strategies, evaluate the performance of each matching strategy, determine the candidate matching strategy set, and provide important references for subsequent strategy previews and optimizations, significantly improve the generation and optimization efficiency of matching strategies, and ensure the efficiency and fairness of the matching process.

[0060] In one implementation of this embodiment, the strategy effectiveness simulator outputs the expected profit and loss matrix of each matching strategy in the strategy space based on the event impact coefficient matrix and the spread Z-score value, and determines the candidate matching strategy set in the strategy space according to the expected profit and loss matrix, including the following steps: S610, the strategy effectiveness simulator creates a virtual order book image synchronized with the order book; S620, the strategy effectiveness simulator uses a preset scenario algorithm to generate ±3σ price fluctuation scenarios and liquidity drop simulation scenarios based on the event impact coefficient matrix and the spread Z-score value; S630, the strategy effectiveness simulator uses a preset genetic algorithm to simulate each matching strategy in the strategy space in the ±3σ price fluctuation scenario and the liquidity drop simulation scenario, and outputs the expected profit and loss matrix of each matching strategy; S640. Calculate the expected return-risk ratio of each matching strategy based on the expected profit-loss matrix and the preset return-risk ratio calculation formula, and sort each matching strategy from high to low according to the return-risk ratio; S650. The top N matching strategies in terms of return-risk ratio are selected as a candidate matching strategy set, where N is a positive integer greater than 2.

[0061] The virtual order book image is used to simulate the real-time status and trading behavior of the market. The virtual order book image is synchronized with the real order book, which can accurately reflect the supply and demand relationship and trading activity of the market. The process of creating a virtual order book image includes data synchronization, image construction and status update. Data synchronization ensures the consistency of the virtual order book image with the real order book by obtaining the data of the real order book in real time, including the quotes and quantities of buyers and sellers. Image construction generates a virtual order book image by copying the data structure of the real order book, ensuring the integrity and accuracy of the virtual order book image. Status update monitors the changes of the real order book in real time and dynamically updates the virtual order book image to ensure the real-time and reliability of the virtual order book image.

[0062] ±3σ price fluctuation scenarios and liquidity drop simulation scenarios are important scenarios for the strategy effectiveness simulator to evaluate the performance of matching strategies. ±3σ price fluctuation scenarios can simulate the performance of the market under extreme price fluctuation conditions, and liquidity drop simulation scenarios can simulate the performance of the market under conditions of a sharp drop in liquidity.

[0063] The process of generating the above scenarios includes scenario definition, parameter setting and scenario generation. Scenario definition determines the characteristics and scope of the ±3σ price fluctuation scenario and the liquidity drop simulation scenario by analyzing historical data and market experience. Parameter setting determines the key parameters of the scenario, such as price fluctuation amplitude, liquidity change speed, etc., through the event impact coefficient matrix and the spread Z-score value. The event impact coefficient matrix can quantify the impact of different events on the market, and the spread Z-score value can reflect the price deviation between assets and the significance of arbitrage opportunities. Scenario generation generates ±3σ price fluctuation scenarios and liquidity drop simulation scenarios based on the event impact coefficient matrix and the spread Z-score value through the preset scenario algorithm. The preset scenario algorithm can be Monte Carlo simulation, random process model, etc., which can accurately simulate the performance of the market under different conditions.

[0064] Genetic algorithms can optimize the performance of matching strategies by simulating the process of natural selection and genetic mutation. The process of simulating matching strategies in the ±3σ price fluctuation scenario and the liquidity drop simulation scenario includes strategy initialization, fitness evaluation, selection, crossover and mutation.

[0065] Specifically, strategy initialization generates the initial matching strategy by randomly generating or based on historical data to ensure the diversity and comprehensiveness of the strategy. Fitness evaluation calculates the fitness value of each matching strategy by simulating the performance of the matching strategy in the ±3σ price fluctuation scenario and the liquidity drop simulation scenario. The fitness value can reflect the performance of the matching strategy under different market conditions and is an important reference for strategy optimization. Selection ensures the retention and inheritance of excellent strategies by selecting matching strategies with excellent performance according to the fitness value. Crossover generates a new matching strategy by combining the characteristics of two matching strategies to ensure the diversity and innovation of the strategy. Mutation generates a new matching strategy by randomly changing certain characteristics of the matching strategy to ensure the diversity and innovation of the strategy.

[0066] The return-risk ratio is an important indicator for evaluating the performance of a matching strategy, and can reflect the balance between the expected return and risk of the matching strategy. According to the expected profit and loss matrix, the expected return-risk ratio of each matching strategy is calculated using a preset return-risk ratio calculation formula. The return-risk ratio calculation formula can refer to S520, and will not be repeated in this embodiment.

[0067] Expected returns can reflect the expected returns of the matching strategy under different market conditions, and expected risks can reflect the expected risks of the matching strategy under different market conditions. Expected returns and expected risks need to be standardized to eliminate dimensional differences under different market conditions.

[0068] After calculating the return-risk ratio, each matching strategy is sorted from high to low according to the return-risk ratio to ensure that matching strategies with outstanding performance are selected first.

[0069] The process of selecting the top N matching strategies in terms of return-risk ratio as candidate matching strategy sets includes strategy selection and set construction. Strategy selection selects the top N matching strategies based on the return-risk ratio sorting results to ensure that the matching strategies with excellent performance are selected first. Set construction ensures the diversity and comprehensiveness of the strategies by integrating the selected matching strategies into the candidate matching strategy set.

[0070] This implementation method creates a virtual order book mirror synchronized with the order book through a strategy performance simulator, which significantly improves the accuracy and reliability of the simulation environment and ensures the authenticity and effectiveness of the strategy preview. By adopting a preset scenario algorithm, based on the event impact coefficient matrix and the price difference Z-score value, ±3σ price fluctuation scenarios and liquidity drop simulation scenarios are generated, which significantly improves the comprehensiveness and reliability of the simulation environment and ensures the diversity and effectiveness of the strategy preview. By adopting a preset genetic algorithm, the matching strategy is simulated in the ±3σ price fluctuation scenario and liquidity drop simulation scenario, and the expected profit and loss matrix is ​​output, which significantly improves the accuracy and reliability of the strategy simulation and ensures the comprehensiveness and effectiveness of the strategy evaluation. By using the expected profit and loss matrix and the preset return-risk ratio calculation formula to calculate the expected return-risk ratio of each matching strategy, and sorting each matching strategy from high to low according to the return-risk ratio, the accuracy and reliability of the strategy evaluation are significantly improved, ensuring the scientificity and effectiveness of the strategy selection. By selecting the top N matching strategies in terms of risk-return ratio as the candidate matching strategy set, the accuracy and reliability of strategy selection are significantly improved, ensuring the diversity and effectiveness of the candidate matching strategy set. In summary, the market status, trading behavior and the influence of external factors can be comprehensively considered to generate diversified matching strategies, evaluate the performance of each matching strategy, and determine the candidate matching strategy set, which provides an important reference for subsequent strategy preview and optimization, significantly improves the generation and optimization efficiency of matching strategies, and ensures the efficiency and fairness of the matching process.

[0071] In one implementation of this embodiment, a candidate matching strategy set is subjected to strategy preview in a virtual order book image generated by a strategy performance simulator to calculate an expected matching efficiency index of each candidate matching strategy, including the following steps: S710. In the virtual order book image, each candidate matching strategy is executed according to a preset time window; S720, recording the matching success rate, matching delay time and matching cost of each candidate matching strategy within the time window; S730: According to the matching success rate, matching delay time and matching cost, a weighted calculation is performed to obtain an expected matching efficiency index of each candidate matching strategy.

[0072] In order to evaluate the performance of candidate matching strategies, it is necessary to execute each candidate matching strategy separately in the virtual order book mirror according to the preset time window. The time window is set according to market characteristics and trading needs, and can reflect the market status and trading behavior in different time periods. The process of executing candidate matching strategies includes strategy loading, strategy execution and result recording. Strategy loading ensures that the strategy can run in a simulation environment by loading the candidate matching strategy into the virtual order book mirror. Strategy execution ensures the comprehensiveness and effectiveness of the strategy by simulating the running process of the matching strategy in the virtual order book mirror, including order matching, transaction execution and result feedback. Result recording ensures the accuracy and completeness of the data by recording the performance of the matching strategy within the time window, including matching success rate, matching delay time and matching cost.

[0073] Matching success, matching delay time and matching cost are important indicators for evaluating the performance of candidate matching strategies, which can reflect the effectiveness, response speed and economy of matching strategies. In order to calculate the expected matching efficiency indicators of each candidate matching strategy, it is necessary to record the matching success, matching delay time and matching cost of each candidate matching strategy within the time window. Matching success refers to the proportion of orders successfully matched by the strategy within the time window, reflecting the effectiveness of the strategy. Matching delay time refers to the time from receiving an order to completing matching, reflecting the response speed of the strategy. Matching cost refers to the transaction cost incurred by the strategy during execution, reflecting the economy of the strategy. The original data of matching success, matching delay time and matching cost can be obtained by real-time monitoring of the running process of the matching strategy in the virtual order book mirror.

[0074] The expected matching efficiency index is a comprehensive index for evaluating the performance of candidate matching strategies, which can reflect the balance between the effectiveness, response speed and economy of matching strategies. In order to calculate the expected matching efficiency index of each candidate matching strategy, a weighted calculation method is required based on the matching success rate, matching delay time and matching cost. The weighted calculation method can comprehensively consider the importance of different indicators and ensure the comprehensiveness and accuracy of the evaluation results.

[0075] The calculation formula of the expected matching efficiency index is: Expected matching efficiency index = w1×matching success rate+w2×(1-matching delay time)+w3×(1-matching cost); Among them, w1, w2 and w3 are the weights of matching success rate, matching delay time and matching cost respectively, and w1+w2+w3=1. The weights are set according to market characteristics and transaction needs, and can reflect the importance of different indicators in the matching efficiency evaluation.

[0076] This implementation method significantly improves the accuracy and reliability of strategy execution by executing each candidate matching strategy separately according to a preset time window in the virtual order book mirror, thereby ensuring the comprehensiveness and effectiveness of strategy evaluation. By recording the matching success rate, matching delay time, and matching cost of each candidate matching strategy within the time window, the accuracy and reliability of data recording are significantly improved, ensuring the comprehensiveness and effectiveness of strategy evaluation. By weighted calculation based on the matching success rate, matching delay time, and matching cost, the expected matching efficiency index of each candidate matching strategy is obtained, which significantly improves the accuracy and reliability of strategy evaluation and ensures the scientificity and effectiveness of strategy selection.

[0077] In one implementation of this embodiment, the dynamic simulation strategy includes the following steps: S810. Based on the expected matching efficiency index of each candidate matching strategy, construct an objective function, where the objective function includes maximizing matching success rate, minimizing matching delay time, and minimizing matching cost; S820. Use the Pareto optimal solution algorithm to solve the objective function and obtain the optimal candidate matching strategy.

[0078] The construction of the objective function is based on the expected matching efficiency indicators of each candidate matching strategy, including maximizing matching success, minimizing matching delay time, and minimizing matching cost. Matching success refers to the proportion of orders successfully matched by the strategy within the time window, reflecting the effectiveness of the strategy. Matching delay time refers to the time from receiving an order to completing matching, reflecting the response speed of the strategy. Matching cost refers to the transaction cost incurred during the execution of the strategy, reflecting the economic efficiency of the strategy. The construction process of the objective function includes indicator selection, weight allocation, and function definition.

[0079] Specifically, indicator selection determines the matching success rate, matching delay time and matching cost as the key indicators of the objective function by analyzing market characteristics and transaction needs. Weight allocation assigns corresponding weights to each indicator according to market characteristics and transaction needs to reflect its importance in the objective function. For example, the weight of matching success rate may be higher to reflect its important impact on matching efficiency. Function definition ensures that the objective function can fully reflect the performance of the matching strategy by integrating indicators and weights into the objective function. The formula of the objective function is: objective function = q1×matching success rate + q2×(1-matching delay time) + q3×(1-matching cost), where q1, q2 and q3 are the weights of matching success rate, matching delay time and matching cost, respectively, and q1+q2+q3=1.

[0080] The Pareto optimal solution algorithm can determine the optimal candidate matching strategy through multi-objective optimization. The solution process of the Pareto optimal solution algorithm includes initial solution generation, solution evaluation, solution selection and solution optimization. Initial solution generation ensures the diversity and comprehensiveness of solutions by randomly generating or generating initial matching strategies based on historical data. Solution evaluation ensures the accuracy and reliability of solutions by calculating the objective function value of each initial solution and evaluating its matching success rate, matching delay time and matching cost. Solution selection ensures the retention and inheritance of excellent solutions by selecting solutions with excellent performance according to the objective function value. Solution optimization ensures the diversity and innovation of solutions by combining the features of two solutions to generate new solutions.

[0081] This embodiment constructs an objective function based on the expected matching efficiency index of each candidate matching strategy, which significantly improves the accuracy and reliability of the objective function and ensures the comprehensiveness and effectiveness of the strategy evaluation. The objective function includes maximizing the matching success rate, minimizing the matching delay time and minimizing the matching cost, which can fully reflect the performance of the matching strategy. By adopting the Pareto optimal solution algorithm to solve the objective function and obtain the optimal candidate matching strategy, the accuracy and reliability of the strategy solution are significantly improved, ensuring the scientificity and effectiveness of the optimal candidate matching strategy. The optimal candidate matching strategy refers to a matching strategy that achieves the optimal balance between the matching success rate, the matching delay time and the matching cost, which can perform well under different market conditions, significantly improves the generation and optimization efficiency of the matching strategy, and ensures the efficiency and fairness of the matching process.

[0082] In one implementation of this embodiment, the optimal candidate matching strategy is loaded into the adaptive matching engine to perform order matching, including the following steps: S910, loading the optimal candidate matching strategy into the adaptive matching engine; Among them, the adaptive matching engine is used to monitor the changes in market data in real time according to the optimal candidate matching strategy, and automatically execute matching operations to complete order matching when an order that meets the optimal candidate matching strategy is detected.

[0083] In order to apply the optimal candidate matching strategy to actual transactions, it needs to be loaded into the adaptive matching engine. The adaptive matching engine is an important part of the financial trading system. It can monitor the changes of market data in real time according to the optimal candidate matching strategy, and automatically perform matching operations and complete order matching when an order that meets the optimal candidate matching strategy is detected.

[0084] The process of loading the best candidate matching strategy includes strategy parsing, strategy configuration, and strategy activation. Strategy parsing ensures that the strategy can be correctly understood and executed by the adaptive matching engine by parsing the parameters and rules of the best candidate matching strategy. Strategy configuration ensures that the strategy can be run in actual transactions by configuring the parsed strategy parameters and rules into the adaptive matching engine. Strategy activation ensures that the strategy can monitor changes in market data in real time and automatically perform matching operations when orders that meet the strategy are detected by activating the configured strategy.

[0085] The adaptive matching engine can monitor the changes in market data in real time according to the optimal candidate matching strategy, and automatically perform matching operations and complete order matching when an order that meets the optimal candidate matching strategy is detected. Real-time monitoring of changes in market data is the core function of the adaptive matching engine, which can ensure the real-time and accuracy of the matching process. Market data includes information such as order books, transaction details, price fluctuations, etc., which reflects the real-time status and trading behavior of the market.

[0086] When an order that meets the optimal candidate matching strategy is detected, the adaptive matching engine automatically performs the matching operation and completes the order matching. Order matching ensures the efficiency and fairness of the matching process by matching the orders of buyers and sellers according to the optimal candidate matching strategy.

[0087] This implementation significantly improves the accuracy and reliability of strategy loading by loading the optimal candidate matching strategy into the adaptive matching engine, ensuring the efficiency and fairness of the matching process. The adaptive matching engine monitors changes in market data in real time, and automatically performs matching operations and completes order matching when an order that meets the optimal candidate matching strategy is detected, significantly improving the real-time and accuracy of the matching process, ensuring the efficiency and fairness of the matching process, and significantly improving the real-time and accuracy of the matching process, ensuring the efficiency and fairness of the matching process.

[0088] The present application also provides a financial transaction system. Figure 2 As shown, including: A memory is configured to store instructions; and a processor is configured to call instructions from the memory and implement the above-mentioned dynamic simulation matching method under multiple strategies when executing the instructions.

[0089] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0090] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram and the combination of the processes and / or blocks in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0091] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0092] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0093] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0094] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0095] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0096] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0097] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A dynamic simulation matching method under multiple strategies, characterized in that: Applied to a financial transaction system, the financial transaction system is deployed with an adaptive matching engine, and the method includes: In response to receiving multi-source heterogeneous data, respectively processing the multi-source heterogeneous data to obtain processed multi-source heterogeneous data, wherein the multi-source heterogeneous data includes market data, public opinion data, and arbitrage channel detection data; Inputting the processed multi-source heterogeneous data into a dynamic strategy evaluation network to generate a candidate matching strategy set, wherein the dynamic strategy evaluation network is implemented based on a pre-built hybrid prediction model and a strategy effectiveness simulator, and the candidate matching strategy set includes at least one candidate matching strategy; Performing strategy preview on the candidate matching strategy set in the virtual order book image generated by the strategy effectiveness simulator to calculate the expected matching efficiency index of each candidate matching strategy; Adopting a dynamic simulation strategy to determine the optimal candidate matching strategy according to the expected matching efficiency index of each candidate matching strategy; The optimal candidate matching strategy is loaded into the adaptive matching engine for order matching.

2. The method according to claim 1, characterized in that: The financial transaction system includes an FPGA coprocessor, and the multi-source heterogeneous data are processed separately to obtain processed multi-source heterogeneous data, including: Using the FPGA coprocessor to classify the market data, and obtain order book and transaction detail data; The FPGA coprocessor is used to parse the public opinion data to obtain a real-time news stream, and an event impact coefficient matrix is ​​generated based on the real-time news stream; Using the FPGA coprocessor to perform cointegration analysis on the relevant asset price series in the arbitrage channel detection data, and calculate the price difference Z-score value; Among them, the order book, the transaction details data, the event impact coefficient matrix and the price difference Z-score value are all processed multi-source heterogeneous data.

3. The method according to claim 2, characterized in that The method further comprises: When the average value of the price difference Z-score value is greater than a preset threshold, an arbitrage signal is generated.

4. The method according to claim 2, characterized in that: The hybrid prediction model includes an LSTM neural network, a graph convolutional network and a reinforcement learning agent module, wherein the LSTM neural network is used to process the transaction detail data and output time series features, the graph convolutional network is used to analyze the topological structure of the order book and output a node feature vector, and the reinforcement learning agent module is used to generate a strategy space based on the joint features; Among them, the joint feature is obtained by tensor splicing between the temporal feature output by the LSTM neural network and the node feature vector output by the graph convolutional network.

5. The method according to claim 4, characterized in that The processed multi-source heterogeneous data is input into the dynamic strategy evaluation network to generate a candidate matching strategy set, including: Inputting the order book and the transaction details data into the hybrid prediction model to obtain a strategy space, wherein the strategy space includes all matching strategies; The strategy effectiveness simulator is used to output the expected profit and loss matrix of each matching strategy in the strategy space based on the event impact coefficient matrix and the price difference Z-score value, and determine the candidate matching strategy set in the strategy space according to the expected profit and loss matrix.

6. The method according to claim 5, characterized in that The strategy effectiveness simulator outputs the expected profit and loss matrix of each matching strategy in the strategy space based on the event impact coefficient matrix and the price difference Z-score value, and determines the candidate matching strategy set in the strategy space according to the expected profit and loss matrix, including: The strategy performance simulator creates a virtual order book mirror that is synchronized with the order book; The strategy effectiveness simulator adopts a preset scenario algorithm to generate a ±3σ price fluctuation scenario and a liquidity drop simulation scenario based on the event impact coefficient matrix and the price difference Z-score value; The strategy effectiveness simulator adopts a preset genetic algorithm to simulate each matching strategy in the strategy space in the ±3σ price fluctuation scenario and the liquidity drop simulation scenario, and outputs the expected profit and loss matrix of each matching strategy; According to the expected profit and loss matrix, the expected profit and risk ratio of each matching strategy is calculated using a preset profit and risk ratio calculation formula, and each matching strategy is sorted from high to low according to the profit and risk ratio; The top N matching strategies in terms of return-risk ratio are taken as the candidate matching strategy set, where N is a positive integer greater than 2.

7. The method according to claim 1, characterized in that The step of performing strategy preview on the candidate matching strategy set in the virtual order book image generated by the strategy performance simulator to calculate the expected matching efficiency index of each candidate matching strategy includes: In the virtual order book image, each candidate matching strategy is executed respectively according to a preset time window; Recording the matching success rate, matching delay time and matching cost of each candidate matching strategy within the time window; According to the matching success rate, matching delay time and matching cost, a weighted calculation is performed to obtain an expected matching efficiency index for each candidate matching strategy.

8. The method according to claim 7, characterized in that The dynamic simulation strategy includes: Based on the expected matching efficiency index of each candidate matching strategy, construct an objective function, wherein the objective function includes maximizing matching success rate, minimizing matching delay time, and minimizing matching cost; The Pareto optimal solution algorithm is used to solve the objective function and obtain the optimal candidate matching strategy.

9. The method according to claim 1, characterized in that: The loading of the optimal candidate matching strategy into the adaptive matching engine for order matching includes: Loading the optimal candidate matching strategy into the adaptive matching engine; The adaptive matching engine is used to monitor market data and order book changes in real time according to the optimal candidate matching strategy; The adaptive matching engine is used to monitor the changes in market data in real time according to the optimal candidate matching strategy, and automatically perform matching operations to complete order matching when an order that meets the optimal candidate matching strategy is detected.

10. A financial transaction system, characterized in that: include: a memory configured to store instructions; and a processor configured to call the instructions from the memory and to implement the dynamic simulation matching method under multiple strategies according to any one of claims 1 to 9 when executing the instructions.

Images