Search

CN-122023007-A - Intelligent period present fusion market analysis system and method

CN122023007ACN 122023007 ACN122023007 ACN 122023007ACN-122023007-A

Abstract

The invention relates to an intelligent period present fusion quotation analysis system and method, wherein the system comprises a data source layer, a data access and processing layer, a core calculation layer and an application display layer, wherein the data source layer is configured to acquire data information of a futures exchange, a spot exchange platform and a macroscopic/industrial database in real time, the data access and processing layer is configured to receive, buffer and preprocess the data acquired by the data source layer, the core calculation layer is configured to perform reasoning calculation through loading a pre-trained Transformer model and perform data storage, and the application display layer is configured to convert a calculation result into a transaction strategy and visual content of user interaction. The invention solves the problem of data island, realizes automatic fusion and standardization of the periodical data, introduces an AI deep learning model, remarkably improves the accuracy of market forecast, realizes a full-automatic analysis flow, greatly improves the transaction efficiency and avoids human errors.

Inventors

  • Yao Aijia
  • YANG JINHONG

Assignees

  • 普杉科技发展(四川)有限公司

Dates

Publication Date
20260512
Application Date
20260316

Claims (9)

  1. 1. The intelligent period present fusion quotation analysis system is characterized by comprising a data source layer, a data access and processing layer, a core calculation layer and an application display layer; the data source layer is configured to acquire data information of a futures exchange, a spot exchange platform and a macroscopic/industrial database in real time; the data access and processing layer is configured to receive, buffer and preprocess the data acquired by the data source layer; The core calculation layer is configured to perform reasoning calculation by loading a pre-trained transducer model and perform data storage; The application presentation layer is configured to convert the calculation result into a transaction strategy and visual content of user interaction.
  2. 2. The intelligent period present fusion quotation analysis system of claim 1, wherein the data access and processing layer comprises a data access module, a message queue cluster, a data alignment and cleaning module and a characteristic engineering module; the data access module is deployed on a two-way server, operates a multithreading program and analyzes binary stream data of a futures/spot interface through a protocol; the message queue cluster adopts a distributed message queue, configures 10 partitions and 3 copy factors and is used for buffering flow fluctuation caused by a data source structure; The data alignment and cleaning module fills spot discrete quotation forward to futures time granularity through a linear interpolation algorithm, and simultaneously executes abnormal value detection to remove noise generated by network jitter or input errors; the characteristic engineering module calculates various characteristics in real time based on the cleaned data and stores the calculation result into the characteristic engineering database.
  3. 3. The intelligent period present fusion quotation analysis system of claim 1, wherein the performing inference calculation by loading a pre-trained transducer model comprises: a1, an input layer of a transducer model receives time sequence data of futures, spot goods and macroscopic/price auxiliary channels; A2, extracting local morphological features of each channel by the feature extraction layer through parallel 1D-CNN; A3, calculating dynamic weights of futures and spot channels by the attention fusion layer to realize feature weighted fusion; a4, capturing long-term dependency relationship by using LSTM by the time sequence reasoning layer, and outputting base difference trend prediction probability by the output layer; And A5, the output layer judges the transaction signal according to the probability value and the threshold value, and pushes the transaction signal to the front-end visual terminal.
  4. 4. The intelligent period presentation fusion quotation analysis system of claim 3, wherein the output layer judges the trading signal according to the probability value and the threshold value comprises: Setting a dynamic threshold value theta, judging that the high confidence degree looks at a plurality of signals if the probability value P > theta, judging that the high confidence degree looks at a null signal if P < (1-theta), and judging that the shock signal if theta < P < (1-theta); If the high-confidence-level multi-signal watching is triggered, a JSON format transaction signal packet comprising a variety code, an operation type, an expected anti-interference point position, an anti-loss point position and a signal generation time stamp is automatically generated.
  5. 5. The intelligent period presentation fusion quotation analysis system of claim 3, wherein the output layer judges the trading signal according to the probability value and the threshold value comprises: Setting a dynamic threshold value theta, judging that the high confidence degree looks at a plurality of signals if the probability value P > theta, judging that the high confidence degree looks at a null signal if P < (1-theta), and judging that the shock signal if theta < P < (1-theta); If the high-confidence-level multi-signal watching is triggered, a JSON format transaction signal packet comprising a variety code, an operation type, an expected anti-interference point position, an anti-loss point position and a signal generation time stamp is automatically generated.
  6. 6. The intelligent period present fusion quotation analysis method is characterized by comprising the following steps of: s1, monitoring ThostFtdcMdApi protocol data flow of futures and RESTful API request of spot in real time, triggering event interruption when new Tick data arrives, storing input into a memory buffer area, and cleaning and time aligning monitored data; S2, calculating various characteristics of the aligned data in real time, and normalizing the characteristics; And S3, inputting the normalized feature vector into a pre-trained transducer model to obtain the base difference rising probability, and generating a transaction signal through threshold judgment and pushing the transaction signal to a front-end visual terminal.
  7. 7. The method for analyzing intelligent period present fusion quotation of claim 6, wherein S3 comprises the following specific contents: a1, an input layer of a transducer model receives time sequence data of futures, spot goods and macroscopic/price auxiliary channels; A2, extracting local morphological features of each channel by the feature extraction layer through the parallel one-dimensional convolution layer; A3, calculating dynamic weights of futures and spot channels by the attention fusion layer to realize feature weighted fusion; a4, capturing long-term dependency relationship by using LSTM by the time sequence reasoning layer, and outputting base difference trend prediction probability by the output layer; And A5, the output layer judges the transaction signal according to the probability value and the threshold value, and pushes the transaction signal to the front-end visual terminal.
  8. 8. The method for intelligent period present fusion quotation analysis according to claim 7, wherein the outputting layer judges the trading signal according to the probability value and the threshold value comprises: Setting a dynamic threshold value theta, judging that the high confidence degree looks at a plurality of signals if the probability value P > theta, judging that the high confidence degree looks at a null signal if P < (1-theta), and judging that the shock signal if theta < P < (1-theta); If the high-confidence-level multi-signal watching is triggered, a JSON format transaction signal packet comprising a variety code, an operation type, an expected anti-interference point position, an anti-loss point position and a signal generation time stamp is automatically generated.
  9. 9. The intelligent period present fusion quotation analysis method according to claim 7, wherein A1 comprises setting a time window T=60, constructing a futures price channel matrix and a spot price channel matrix, and inputting the futures price channel matrix and the spot price channel matrix into a feature extraction layer; using two parallel one-dimensional convolution layers to respectively process futures and spot channel matrixes and extracting local morphological characteristics; Splicing feature matrixes of futures and spot channels, and calculating an attention weight vector alpha through a full-connection layer and a Sigmoid activation function; The A4 comprises the steps of carrying out weighted fusion on feature matrixes of futures and spot channels according to an attention weight vector alpha to obtain a fused feature vector Xfused, inputting the feature vector Xfused into a double-layer LSTM network, extracting long-term dependency relations in a time sequence, and finally mapping a result output by the LSTM to a prediction result by a full-connection layer.

Description

Intelligent period present fusion market analysis system and method Technical Field The invention relates to the field of big data processing, in particular to an intelligent period and current fusion market analysis system and method. Background With the penetration of global economy integration, price fluctuation of bulk commodities (such as black series, chemical industry, agricultural products and the like) is increasingly severe, the market trading scale of commodity futures in China is huge as the largest global bulk commodity consumption state, the commodity futures in China is counted by relevant industry associations, the commodity futures in China continuously occupy the first place of the world for many years, the 2023 success rate breaks through 50 hundred million-hand customs, and the daily holding capacity is maintained above 4000 thousand-hand. In order to avoid the risk caused by severe price fluctuation, more and more entity enterprises and investment institutions adopt a 'period-now combined' trading mode, namely, the operation is performed on the futures market and the spot market at the same time, and the benefit is locked by using the period guarantee value or the period-now benefit. Under such a background, the application of the financial science and technology in the trade field is continuously deepened, and extremely high requirements are put on the speed and depth of data processing. The current date (futures and spot) trading data mainly originate from two channels, namely 1, futures quotation, namely, from the quotation interfaces of authorities of various futures exchanges (such as the institute of the upper period, the institute of the large business, the institute of the Zheng Shang and the like), the data standardization degree is high, the quotation data packet refreshing frequency of the main stream variety in the trading activity period reaches 500 milliseconds, and the grade of Tick data generated every day can reach tens of millions. 2. Spot quotation is derived from spot e-commerce platforms (such as steel network searching and steel silver e-commerce), information portal sites (such as my steel network and android information) or off-line manual quotation. The data formats are extremely non-uniform, the update frequency difference is huge, the discretization characteristic is often presented, and the quotation update frequency of partial non-mainstream varieties is even lower than 10 minutes. The existing market analysis technology mainly stays in the following stages of 1, manual statistics and Excel calculation, wherein a trader obtains spot quotation through a telephone or an instant messaging tool, manually inputs an Excel form, and calculates price difference by using a simple linear formula (such as: base difference = spot price-futures price). 2. And the independent quotation software is used for checking futures trends by using traditional futures disc-watching software (such as Wen Hua financial finance and blogs Yi Da, and checking spot information by using a web browser), and part of high-end terminals provide cross-market arbitrage monitoring, but are mostly based on fixed rules and lack deep fusion capability on heterogeneous data. 3. Simple threshold pre-warning, setting a static threshold based on linear logic, such as "alarm when price difference is greater than 500 points", but such static rules tend to fail in the face of high frequency fluctuations and complex market emotions. Aiming at the scene of the prior art combining at present, the prior art has the obvious defects and defects that 1, the phenomenon of data islanding is serious, heterogeneous data is difficult to automatically correlate in real time, futures market high-density time sequence data, spot market is mostly unstructured text or discrete data, the prior art lacks an effective data fusion middleware, and high-precision alignment of the two types of data in the time dimension is difficult to realize. In practice, due to the lack of automated data cleansing piping, data engineers and traders typically spend 30% -40% of their working time for data cleansing and reconciliation. In addition, the average acquisition delay of spot data is usually 1 to 5 minutes, and futures data is in millisecond level, so that the calculated base difference is often delayed from the real market value due to serious 'dislocation' (Time Lag) on a Time axis, and transaction decisions based on error information are extremely easy to be caused. 2. The analysis means is lagged, and market fluctuation of microsecond level cannot be captured in real time, the existing analysis method mainly depends on static historical statistics or manual experience review, is basically post-analysis, is fatal when facing programmed High Frequency Transactions (HFT), and can lead to price fluctuation rate of active varieties (such as screw steel and crude oil) to rise in a few seconds under extreme market conditions (such as seri