Lightweight multi-source high-latitude financial time series data joint prediction system and method thereof

Abstract

The invention discloses a lightweight multi-source high-latitude financial time series data joint prediction system and a method thereof, relates to the technical field of time series data analysis, and solves the problems that models are generally uninterpretable, too complex and high in computing power consumption due to a complex neural network framework in an existing neural network. Comprising a data acquisition module, a data cleaning module, a model training module, a data prediction module and a prediction visualization unit. Comprising the following steps: step 1, acquiring time series data from different financial sources by using a data acquisition module; compared with a traditional LDS model, the aLDS adopted by the invention has a novel parameter averaging process, and in an expectation maximization algorithm, averaging processing is carried out on parameters obtained by training each piece of financial time data, so that the purpose of jointly training multiple pieces of financial time data is achieved, and the method has the advantages of being high in practicability and the like. And the prediction precision of the prediction model on multiple pieces of financial time data is further improved.

Description

Technical Field

[0001] The present invention relates to the technical field of time series data analysis, and in particular to a lightweight multi-source high-dimensional financial time series data joint prediction system and method thereof. Background Art

[0002] In the financial field, the analysis and prediction of time series data are widely used in the stock market, bond market, derivatives and other fields. Existing technologies include ARIMA, deep learning algorithms, etc., which have certain application scenarios in the processing and prediction of time series data. Most financial time series prediction methods rely on historical data to capture market trends and optimize the parameters and hyperparameters of the probability mathematical model to derive the optimal probability model to predict future data and achieve higher test accuracy. In order to improve the prediction accuracy, existing technologies usually require powerful computing power and complex algorithm models, such as complex neural network architectures in neural networks, such as RNN, LSTM, etc.

[0003] However, the complex neural network architecture in existing neural networks usually has disadvantages such as uninterpretability, excessive complexity, and high computing power consumption;

[0004] Traditional financial time series prediction algorithms such as ARMA and ARIMA are only suitable for simple multivariate data prediction and cannot process complex multi-source high-dimensional time series data. In particular, the limitations are obvious when processing multiple high-dimensional time series data of different lengths;

[0005] Deep learning algorithms are often not interpretable, such as LSTM and RNN, which have very complex network structures and are often referred to as black-box algorithms. This is very user-unfriendly, as users cannot intuitively understand the actual meaning of these deep learning algorithms and only regard them as a trained input-output function.

[0006] Deep learning algorithms often require high computing resources, such as GPUs, to train models and find optimal parameters, which makes deep learning algorithms unsuitable for lightweight data applications. Therefore, deep learning algorithms are unlikely to become a means of real-time prediction of financial time series.

[0007] The above algorithms can only process a single high-dimensional time series. They cannot process high-dimensional time series data of different lengths, so they do not meet the existing needs. In this regard, we propose a lightweight multi-source high-dimensional financial time series data joint prediction system and method. Summary of the invention

[0008] The purpose of the present invention is to provide a lightweight multi-source high-dimensional financial time series data joint prediction system and method thereof to solve the problems of complex neural network architecture in the existing neural networks proposed in the above background technology, these models are usually unexplainable, too complex, and consume a lot of computing power.

[0009] To achieve the above-mentioned purpose, the present invention provides the following technical solution: a lightweight multi-source high-dimensional financial time series data joint prediction system, including: a data acquisition module, a data cleaning module, a model training module, a data prediction module and a prediction visualization unit.

[0010] A lightweight multi-source high-dimensional financial time series data joint prediction method includes the following steps:

[0011] Step 1: Use the data collection module to obtain time series data from different financial sources;

[0012] Step 2: Use the data cleaning module to preprocess the collected time series data, remove noise and outliers to obtain the training data set;

[0013] Step 3: Use the model training module to train the financial time series;

[0014] Step 4: Based on model training, predict future time series data;

[0015] Step 5: Use the forecast visualization unit to display the forecast results through a graphical interface to facilitate analysis and decision-making

[0016] Preferably, the financial time series training is to use the aLDS algorithm to train the cleaned data and utilize the multivariate weighted average method to process time series of different lengths.

[0017] Preferably, the aLDS algorithm is an average linear dynamic system algorithm, which can predict multi-source, high-dimensional financial time series data. The specific parameter averaging process is as follows:

[0018]

[0019]

[0020] where x i is the implicit variable, y i is the actual variable, E is the expected value, C new and A new is the matrix parameter in the current aLDS algorithm, T n is the length of the nth financial time series.

[0021] Preferably, the aLDS algorithm has a parameter averaging process, and in the expectation-maximization algorithm, the parameters obtained by training each time series data are averaged.

[0022] Preferably, the training data set is completely segmented, and the training data set includes a training data set and a test data set. 80% of the complete time series are selected from the training data set as the training data set, and the remaining 20% ​​of the complete time series are selected as the test data set.

[0023] Compared with the prior art, the present invention has the following beneficial effects:

[0024] 1. The aLDS algorithm of the present invention is a mathematical model that can accurately predict multi-source, high-dimensional financial time series data, especially when the data set contains multiple high-dimensional time series of different lengths;

[0025] 2. The aLDS algorithm of the present invention is a fast and efficient multi-source, high-dimensional financial time series prediction model that uses as few computing resources as possible. Compared with deep learning algorithms, this algorithm consumes less computing resources and is particularly suitable for environments with limited computing power;

[0026] 3. Under the assumption of a reasonable data model, the aLDS algorithm of the present invention is a multi-source high-dimensional time series prediction algorithm with certain physical interpretation. Compared with the traditional LDS model, aLDS has a novel averaging parameters process. Specifically, the average summation process is used in formula (1), for example:

[0027] in It is a process of finding the average value. new ,Σ new ,A new ,Γ new Can be equivalent form.

[0028] In the EM (Expectation Maximization) algorithm, the parameters obtained from the training of each financial time data are averaged to achieve the purpose of joint training of multiple financial time data, further improving the prediction accuracy of the prediction model on multiple financial time data. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 The aLDS algorithm flow chart of the present invention: lightweight multi-source high-dimensional financial time series data joint prediction system and method;

[0030] Figure 2This is the aLDS algorithm diagram of the present invention: a multi-source high-dimensional financial time series data joint prediction system, and a parameter forward propagation diagram on each time series; wherein x represents the implicit function layer in the aLDS model, and y represents the actual data layer in the aLDS model.

[0031] Figure 3 This is the aLDS algorithm diagram of the present invention: a joint prediction system for multi-source high-dimensional financial time series data, with parameters propagating backward on each time series. DETAILED DESCRIPTION

[0032] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0033] See also Figures 1 to 3 An embodiment of the present invention provides a lightweight multi-source high-dimensional financial time series data joint prediction system, including: a data acquisition module, a data cleaning module, a model training module, a data prediction module and a prediction visualization unit.

[0034] A lightweight multi-source high-dimensional financial time series data joint prediction method, characterized by comprising the following steps:

[0035] Step 1: Use the data collection module to obtain time series data from different financial sources;

[0036] Step 2: Use the data cleaning module to preprocess the collected time series data, remove noise and outliers to obtain the training data set. The training data set is completely segmented. The training data set includes the training data set and the test data set. 80% of the complete time series is selected from the training data set as the training data set, and the remaining 20% ​​of the complete time series is the test data set;

[0037] Step 3: Use the model training module to train the financial time series;

[0038] Step 4: Based on model training, predict future time series data;

[0039] Step 5: Use the forecast visualization unit to display the forecast results through a graphical interface to facilitate analysis and decision-making

[0040] The above-mentioned financial time series training uses the aLDS algorithm to train the cleaned data and uses the multivariate weighted average method to process time series of different lengths. The aLDS algorithm is an average linear dynamic system algorithm that can accurately predict multi-source, high-dimensional financial time series data. The specific parameter averaging process and formula are as follows:

[0041]

[0042] where x i is the implicit variable, y i is the actual variable, E is the expected value, C new and A new is the matrix parameter in the current aLDS algorithm, T n is the length of the nth financial time series. The aLDS algorithm has a parameter averaging process. In the expectation maximization algorithm, the parameters obtained by training each time series data are averaged.

[0043] The multivariate financial time series weighted averaging method of aLDS algorithm: Compared with the traditional LDS model, aLDS adds a novel averaging parameters process on the basis of the LDS model. Specifically, the average summation process is used in formula (1), for example:

[0044] in It is a process of finding the average value. new ,Σ new ,A new ,Γ new Can be equivalent form;

[0045] In the EM (Expectation Maximization) algorithm, the parameters obtained from the training of each piece of financial time data are averaged, so as to achieve the purpose of joint training of multiple pieces of financial time data and further improve the prediction accuracy of the prediction model on multiple pieces of financial time data;

[0046] Training and prediction process for multi-source, high-dimensional financial time series prediction model: Different from traditional financial time series prediction methods such as ARIMA (Autoregressive Integrated Moving Average model), LSTM (Long Short-Term Memory), and RNN (Recurrent Neural Network), our training data set and test data set are completely split, that is, we select 80% of the complete time series from the data as the training data set, and the remaining 20% ​​of the complete time series as the test data set. When our model is trained on the training data set, these model parameters and hyperparameters will be directly substituted into the test data set for further evaluation and data visualization.

[0047] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. A lightweight multi-source high-dimensional financial time series data joint prediction system, characterized by: include: Data collection module, data cleaning module, model training module, data prediction module and prediction visualization unit.

2. A lightweight multi-source high-dimensional financial time series data joint prediction method, characterized in that: The steps include: Step 1: Use the data collection module to obtain time series data from different financial sources; Step 2: Use the data cleaning module to preprocess the collected time series data, remove noise and outliers to obtain the training data set; Step 3: Use the model training module to train the financial time series; Step 4: Based on model training, predict future time series data; Step 5: Use the prediction visualization unit to display the prediction results through a graphical interface to facilitate analysis and decision-making.

3. The lightweight multi-source high-dimensional financial time series data joint prediction system and method according to claim 1 is characterized by: The financial time series training uses the aLDS algorithm to train the cleaned data and uses the multivariate weighted average method to process time series of different lengths.

4. The lightweight multi-source high-dimensional financial time series data joint prediction system and method according to claim 1 is characterized by: The aLDS algorithm is an average linear dynamic system algorithm that can predict multi-source, high-dimensional financial time series data. The specific parameter averaging process and formula are as follows: where x i is the implicit variable, y i is the actual variable, E is the expected value, C new and A new is the matrix parameter in the current aLDS algorithm, T n is the length of the nth financial time series.

5. The lightweight multi-source high-dimensional financial time series data joint prediction system and method according to claim 1 is characterized by: The aLDS algorithm has a parameter averaging process, and in the expectation maximization algorithm, the parameters obtained by training each time series data are averaged.

6. The lightweight multi-source high-dimensional financial time series data joint prediction system and method according to claim 1 is characterized by: The training data set is completely segmented, and the training data set includes a training data set and a test data set. 80% of the complete time series are selected from the training data set as the training data set, and the remaining 20% ​​of the complete time series are selected as the test data set.

Images