Futures price prediction method, system, equipment and medium

Abstract

The invention discloses a futures price prediction method, system, device and medium, and relates to the technical field of price prediction.The method comprises the steps that complex futures price time sequence data is decomposed into a high-frequency sequence, a low-frequency sequence and a residual sequence through a CEEMDAN decomposition algorithm, modeling is carried out on the decomposed sequences through a GARCH model, an ARIMA model and an SVR model, and a prediction result is obtained; the GARCH model can accurately capture the high-frequency fluctuation clustering phenomenon of the time sequence data, the ARIMA model can accurately capture the low-frequency trend and periodic fluctuation of the time sequence data, and the SVR model can effectively process the complex relation between nonlinear data and a residual error sequence. According to the process, after high-frequency fluctuation and low-frequency trend in original time sequence data are accurately decomposed, accurate processing prediction is carried out, and then a high-frequency fluctuation prediction value, a low-frequency trend prediction value and a nonlinear residual prediction value are subjected to weighted reconstruction, so that a prediction structure of multiple models is fused, and the prediction accuracy of the futures price is integrally improved.

Description

Technical Field

[0001] The present invention relates to the technical field of price prediction, and particularly to a futures price prediction method, system, device and medium. Background Art

[0002] In recent years, with the increasingly severe global climate change problem, it has promoted the rapid development of the new energy industry. The new energy revolution has become a global consensus, and new energy technologies such as solar energy, wind energy, and electric vehicles have gradually spread globally. In new energy technologies, silver, as a key material, has excellent electrical conductivity and corrosion resistance, especially in the manufacturing processes of solar panels (photovoltaic power generation) and electric vehicle batteries, playing an important role. With the continuous and rapid development of new energy, the demand for silver in various industries has increased rapidly, and the price fluctuations of silver have attracted high attention from investors. The price fluctuations of silver are affected by various factors such as the global economy, supply and demand changes, and financial market fluctuations. Especially in the context of the rising demand for new energy, accurately predicting the silver price is of great significance for enterprises to formulate strategic plans and investment decisions.

[0003] For the price prediction of silver, traditional methods mainly include time series analysis and statistical regression-based analysis methods. Such models mainly perform regression analysis and sequence modeling through historical price data, and have a certain degree of accuracy in short-term prediction. However, as a financial time series, the silver price has characteristics such as high-frequency noise, non-linearity, and non-stationarity. A single time series model is difficult to fully capture the long-term trend and short-term fluctuations in the silver price change.

[0004] With the rapid development of machine learning and deep learning technologies, support vector regression (SVR), neural networks and other technologies are gradually adopted for the prediction of silver prices. Such technologies can identify non-linear data and process complex time series through the training of a large amount of data. However, this type of technology is greatly affected by high-frequency noise and low-frequency trends in financial time series, and it is difficult to accurately decompose and process high-frequency fluctuations and low-frequency trends in financial time series, thus making it difficult to achieve accurate prediction of futures prices. Summary of the Invention

[0005] Embodiments of the present invention provide a futures price prediction method, system, device and medium, which can solve the problem in the prior art that the current technology is greatly affected by high-frequency noise and low-frequency trends in financial time series and is difficult to accurately decompose and process high-frequency fluctuations and low-frequency trends in financial time series.

[0006] Embodiments of the present invention provide a futures price prediction method, including the following steps:

[0007] Obtain the historical closing price data of futures and form time series data;

[0008] Decompose the time series data using CEEMDAN decomposition to obtain the intrinsic mode function (IMF) subsequences with different central frequencies and the residual sequence; use the t-test to decompose the IMF subsequences with different central frequencies into high-frequency sequences and low-frequency sequences.

[0009] Use the generalized autoregressive conditional heteroskedasticity (GARCH) model to model the high-frequency sequences and obtain the high-frequency volatility prediction values of the time series data; use the autoregressive integrated moving average (ARIMA) model to model the low-frequency sequences and obtain the low-frequency trend prediction values of the time series data; use the support vector regression (SVR) model to model the residual sequence and obtain the non-linear residual prediction values of the time series data.

[0010] Reconstruct the price prediction value of the futures according to the high-frequency volatility prediction value, low-frequency trend prediction value, and non-linear residual prediction value of the time series data.

[0011] Preferably, the step of using the t-test to decompose the IMF subsequences with different central frequencies into high-frequency sequences and low-frequency sequences includes:

[0012] Obtain the t-statistic values of the IMF subsequences with different central frequencies, and its formula is:

[0013]

[0014] Where: represents the average of the IMF subsequences with different central frequencies; n represents the number of the IMF subsequences with different central frequencies; s represents the sample standard deviation.

[0015] Set the significance level threshold. When the t-statistic value significantly deviates from zero, set the IMF subsequence at this frequency as the high-frequency sequence, and at the same time enhance the significance of the high-frequency sequence according to the volatility percentage condition; when the t-statistic value is not significant, set the IMF subsequence at this frequency as the low-frequency sequence.

[0016] Preferably, the step of using the GARCH model to model the high-frequency sequences includes:

[0017] After obtaining the high-frequency sequences, use the GARCH model to model the high-frequency sequences. The GARCH model extracts the variance and volatility of the error terms at past moments of the high-frequency sequences to obtain the volatility at the current moment of the high-frequency sequences, so as to capture the volatility clustering phenomenon in the futures time series data.

[0018] The modeling equation of the GARCH model is:

[0019] y t = μ + ε t

[0020] ε t = σ t z t

[0021]

[0022] Where: y t represents the predicted value of the high-frequency part of the futures price; ε t represents the error term; represents the volatility; z t represents an independent and identically distributed random variable; α 0 、α 1 and β 1 represent the parameters of the GARCH model; μ represents the constant term.

[0023] Preferably, the autoregressive integrated moving average model ARIMA is used to model the low-frequency sequence, including:

[0024] After obtaining the low-frequency sequence, the autoregressive integrated moving average model ARIMA is used to model the stationary shuffled sequence;

[0025] First, use the differencing term in the autoregressive integrated moving average model ARIMA to convert the non-stationary low-frequency sequence into a stationary low-frequency sequence; then use the autoregressive AR in the autoregressive integrated moving average model ARIMA to extract the long-term trend between the current stationary low-frequency sequence and the past stationary low-frequency sequence; use the moving average MA in the autoregressive integrated moving average model ARIMA to extract the periodic fluctuations between the current stationary low-frequency sequence and the past stationary low-frequency sequence;

[0026] The modeling equation of the autoregressive integrated moving average model ARIMA is:

[0027] y t = c + φ 1 y t-1 + φ 2 y t-2 + L + φ p y t-p + θ 1 ε t-1 + L + θ q ε t-q + ε t

[0028] Where: y t represents the predicted value of the low-frequency part of the futures price; φ 1, φ 2 , L, φ p represents the autoregressive term coefficient; θ 1 , θ 2 , L, θ q represents the moving average term coefficient; ε t represents the error term; c represents the constant term.

[0029] Preferably, using the support vector regression model SVR to model the residual sequence includes:

[0030] After obtaining the residual sequence, the support vector regression model SVR extracts the non - linear residual between the non - linear sequence data and the residual part in the residual sequence;

[0031] The modeling equation of the support vector regression model SVR is:

[0032] f(x) = z T φ(x) + b

[0033] Where: φ(·): n →n h represents a non - linear function that maps points in the input space to a higher - dimensional output space; z ∈ in h represents the weight vector; b represents the bias.

[0034] Preferably, reconstructing the predicted value of the futures price includes:

[0035] Determining the weighted coefficients ω of the high - frequency volatility prediction value, low - frequency trend prediction value, and non - linear residual prediction value respectively through the cross - validation method cross - validation 1 , ω 2 , ω 3 ;

[0036] According to the weighted coefficients of the high - frequency volatility prediction value, low - frequency trend prediction value, and non - linear residual prediction value, reconstruct the futures price according to the corresponding frequency - band weights, and its reconstruction equation is:

[0037]

[0038] Where: represents the high - frequency volatility prediction value of the GARCH model; represents the low - frequency trend prediction value of the ARIMA model; represents the non - linear residual prediction value of the SVR model; represents the final predicted value of the futures price.

[0039] The embodiment of the present invention also provides a futures price prediction system, including:

[0040] A data acquisition module, configured to acquire historical closing price data of futures and form time series data;

[0041] A signal decomposition module, configured to decompose the time series data by using CEEMDAN decomposition to obtain intrinsic mode function (IMF) subsequences with different central frequencies and a residual sequence; and decompose the IMF subsequences with different central frequencies into high-frequency sequences and low-frequency sequences by using a t-test;

[0042] A multi-model modeling module, configured to model the high-frequency sequences by using a generalized autoregressive conditional heteroskedasticity (GARCH) model to obtain high-frequency volatility prediction values of the time series data; model the low-frequency sequences by using an autoregressive integrated moving average (ARIMA) model to obtain low-frequency trend prediction values of the time series data; and model the residual sequence by using a support vector regression (SVR) model to obtain non-linear residual prediction values of the time series data;

[0043] A prediction reconstruction module, configured to reconstruct the price prediction value of futures according to the high-frequency volatility prediction values, low-frequency trend prediction values and non-linear residual prediction values of the time series data.

[0044] An embodiment of the present invention further provides an electronic device, including a memory and a processor;

[0045] The memory is configured to store a computer program;

[0046] The processor is configured to implement the steps of a futures price prediction method as described above when executing the computer program stored in the memory.

[0047] An embodiment of the present invention further provides a computer-readable storage medium, configured to store a computer program, and the computer program implements the steps of a futures price prediction method as described above when executed by a processor.

[0048] An embodiment of the present invention provides a futures price prediction method, system, device and medium. Compared with the prior art, the beneficial effects are as follows:

[0049] Based on the principle of adding adaptive noise to the signal in each iteration by the CEEMDAN decomposition algorithm, the present invention decomposes the complex futures price time series data into high-frequency series, low-frequency series and residual series. For the decomposed high-frequency series, low-frequency series and residual series, the GARCH model, ARIMA model and SVR model are used for modeling respectively. The GARCH model can accurately capture the high-frequency volatility clustering phenomenon of time series data, the ARIMA model can accurately capture the low-frequency trend and periodic fluctuations of time series data, and the SVR model can effectively handle the complex relationship between non-linear data and residual series. After the high-frequency fluctuations and low-frequency trends in the original time series data are accurately decomposed and precisely processed and predicted, the predicted values of high-frequency fluctuations, low-frequency trends and non-linear residuals are weighted and reconstructed to fuse the prediction structures of multiple models, thereby overall improving the prediction accuracy of futures prices.

[0050] Moreover, when processing the high-frequency series, the present invention uses the t-test method to judge whether the mean value of the high-frequency series is zero, and combines the variable percentage condition to significantly process the high-frequency fluctuations, thereby improving the ability of the GARCH model to capture short-term fluctuations and the fitting accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a schematic diagram of the overall process of a futures price prediction method provided by an embodiment of the present invention;

[0052] Figure 2 It is a schematic diagram of the decomposed series obtained by decomposing the original price series by the CEEMDAN module in a futures price prediction method provided by an embodiment of the present invention;

[0053] Figure 3 It is a schematic diagram of the volatility curves of the original series, high-frequency series, low-frequency series and residual series after decomposing the original price series in a futures price prediction method provided by an embodiment of the present invention;

[0054] Figure 4 It is a schematic diagram of the prediction results of using the CEEMDAN-ARIMA-GARCH-SVR model in a futures price prediction method provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0055] In order to make the above objects, features and advantages of the present invention more obvious and understandable, the following detailed description of the specific embodiments of the present invention will be given with reference to the accompanying drawings. Many specific details are set forth in the following description in order to fully understand the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0056] See Figure 1 , an embodiment of the present invention provides a futures price prediction method and system, including: a data acquisition and preprocessing module, a signal decomposition module, a fine processing module for high-frequency sequences, a multi-model modeling module, and a prediction reconstruction module.

[0057] 1. Data acquisition and preprocessing module.

[0058] This module is mainly responsible for obtaining the historical data of silver prices from external data sources and preprocessing the data to ensure the smooth progress of subsequent modeling processing; the specific execution steps are as follows:

[0059] Step S1: Data collection.

[0060] Obtain the historical closing price data y(t) of silver futures from the futures market. The data time span is from December 16, 2014 to April 30, 2024. The time granularity of the data is the daily closing price, with a total of 2,279 data. The data set is divided into a training set and a test set according to a ratio of 9:1. Table 1 shows the descriptive statistics of the sample data.

[0061] Table 1 Descriptive statistics of sample data

[0062] Count Mean Variance Range Skewness Kurtosis Full sample sequence 2279 4379.796 767757.1 2913-7442 0.6818020 -0.4219893 Training set 2051 4206.006 530778.4 2913-6658 0.6602619 -0.5796548 Test set 228 5943.145 183538.1 5289-7442 1.8428704 3.5093138

[0063] Step S2: Data preprocessing.

[0064] Denoise, remove outliers, and standardize the obtained silver price data, and convert the price data into a standardized value between 0 and 1; the main purpose is to eliminate the interference of noise on the prediction model and ensure the consistency of the data; the standardization processing formula is:

[0065]

[0066] Among them: X represents the original data; X min and X max represent the minimum and maximum values of the data respectively; X′ represents the standardized data.

[0067] 2. Signal decomposition module.

[0068] This module is mainly responsible for decomposing the silver price data sequence, aiming to decompose the complex price sequence into components of different frequencies, so as to process high-frequency noise and low-frequency trend information separately; the specific execution steps are as follows:

[0069] Step S3: CEEMDAN decomposition.

[0070] The standardized silver price data is decomposed using the CEEMDAN algorithm. By adding adaptive noise, this algorithm decomposes the original time series into several mode functions (IMFs) with different frequencies. Each mode function represents an independent frequency component, containing short-term fluctuation information, high-frequency noise information, and long-term trend information respectively.

[0071] The principle of the CEEMDAN algorithm decomposition is as follows: The CEEMDAN decomposition is based on the EMD (Empirical Mode Decomposition) technique. By adding noise multiple times and decomposing step by step, signals with different frequencies are extracted. The decomposed sequence can be expressed as:

[0072]

[0073] where: IMF i (t) represents the i-th mode function; r n (t) represents the residual term.

[0074] Using the Rlibeemd package in R, the noise added to the CEEMDAN module is set to 100, and the standard deviation of the noise is 0.2. The decomposition sequence obtained by decomposing the original price sequence is as Figure 2 shown.

[0075] 3. Fine processing module for high-frequency sequences.

[0076] After signal decomposition, the high-frequency sequence of the silver price represents short-term price fluctuations, usually caused by speculative behaviors, emergencies, etc. in the market; these high-frequency fluctuation signals may have a greater impact on the prediction results. Therefore, this module is responsible for further fine processing of them; the specific implementation steps are as follows:

[0077] Step S4: Use the T-test to determine whether the mean of the high-frequency sequence is zero, specifically including:

[0078] ① Hypothesis testing:

[0079] Null hypothesis: The mean of the high-frequency sequence is zero (no significant fluctuation).

[0080] Alternative hypothesis: The mean of the high-frequency sequence is not zero (there is significant fluctuation).

[0081] ② T-test: Calculate the t statistic according to the high-frequency sequence data, and the formula is:

[0082]

[0083] ③ Significance test:

[0084] Select an appropriate significance level (such as 0.05), calculate the t value, and compare it with the critical value.

[0085] If the t-value significantly deviates from zero, it is considered that the high-frequency sequence contains significant fluctuations.

[0086] If the t-value is not significant, it is considered that the fluctuations of the sequence are small or random.

[0087] Step S5: Add variable percentage conditions to adjust high-frequency fluctuations; specifically including:

[0088] ① Set the fluctuation percentage condition: Set 5% as the threshold. Then, only when the fluctuation amplitude of the high-frequency sequence exceeds 5%, it is considered that the fluctuation has a significant impact; when the fluctuation amplitude of the high-frequency sequence exceeds this threshold, further enhance its fluctuation amplitude to amplify the influence of the high-frequency sequence in the model.

[0089] ② Application of the adjustment coefficient: According to the results of the t-test and the fluctuation percentage condition, apply an amplification coefficient to the high-frequency sequence to increase the impact of high-frequency fluctuations on the overall data; if the t-test shows that the mean of the high-frequency sequence is not zero and the fluctuation amplitude exceeds the set threshold, then apply the amplification coefficient α to the high-frequency sequence to enhance its influence in the final data reconstruction; its amplification formula is: Adjusted high-frequency sequence = α × Original high-frequency sequence, where α > 1 is the enhancement coefficient used to improve the significance of the high-frequency sequence.

[0090] Step S6: Data reconstruction.

[0091] After comprehensive analysis, it is determined that IMF1 - IMF6 are high-frequency components and IMF7 - IMF9 are low-frequency components. IMF1 - IMF6 are summarized as the high-frequency sequence IMF_H, and IMF7 - IMF9 are summarized as the low-frequency sequence IMF_L. The decomposed sequence curves are as Figure 3 shown.

[0092] 4. Multi-model modeling module.

[0093] This module models the sequences (IMFs) after data reconstruction respectively to capture their different dynamic characteristics; the specific implementation steps are:

[0094] Step S7: Modeling of the high-frequency sequence (IMF_H).

[0095] Use the GARCH model to model the high-frequency sequence (IMF_H) after data reconstruction. The GARCH model can effectively capture the volatility clustering phenomenon in financial time series and is suitable for processing high-frequency noise data. The construction formula of this model is:

[0096] y t = μ + ε t

[0097] ε t = σ t z t

[0098]

[0099] where: y t represents the predicted value of the high-frequency part of the silver price; ε t represents the error term; represents the volatility; z t represents an independent and identically distributed random variable; α 0 、α 1 and β 1 represent the parameters of the model.

[0100] Step S8: Modeling of the low-frequency sequence (IMF_L).

[0101] To ensure the stationarity of the sequence, first perform differencing on the low-frequency sequence data (IMF_L) to eliminate non-stationarity and ensure the stability of model fitting; use the ARIMA model to model the training set of the low-frequency sequence (IMF_L) after data reconstruction. The ARIMA model can effectively predict the long-term trend and periodic fluctuations in the time series and is suitable for processing low-frequency trend data. The construction formula of its model is:

[0102] y t = c + φ 1 y t-1 + φ 2 y t-2 + L + φ p y t-p + θ 1 ε t-1 + L + θ q ε t-q + ε t

[0103] where: y t represents the predicted value of the low-frequency part of the silver price; φ 1 , φ 2 , L, φ p represent the autoregressive term coefficients; θ 1 , θ 2 , L, θ q represent the moving average term coefficients; ε t represents the error term; c represents the constant term.

[0104] Step S9: Modeling of the residual sequence.

[0105] Use the SVR (Support Vector Regression) model to model the residual sequence r n (t) after data reconstruction. The SVR model can effectively handle the complex relationship between non-linear data and the residual part. The non-linear SVR can be expressed as:

[0106] f(x) = z T φ(x) + b

[0107] where: φ(·): n → n h represents a non - linear function that maps points in the input space to a higher - dimensional output space; z ∈ in h represents the weight vector; b represents the bias.

[0108] The GARCH model is used to capture the conditional heteroskedasticity of high - frequency series. Its modeling process includes selecting an appropriate lag order and performing parameter estimation, and finally outputting volatility predictions.

[0109] The ARIMA model fits the trend of low - frequency series by performing autoregressive and moving average operations on the differenced series.

[0110] The SVR model is based on the radial basis kernel function (RBF), performs non - linear regression on the residual part, and optimizes the model by choosing the penalty coefficient and kernel parameters.

[0111] 5. Prediction reconstruction module.

[0112] This module reconstructs the prediction results of each subsequence after modeling and integrates them into the final silver price prediction value; the specific implementation steps are as follows:

[0113] Step S10: Reconstruct the prediction.

[0114] Determine the weighting coefficient ω through the cross - validation method (cross - validation) 1 , ω 2 , ω 3 , to ensure the best contribution of each part to the prediction result. Reconstruct the prediction results of all sub - models (GARCH, ARIMA, SVR) according to the corresponding frequency - band weights to generate the final silver price prediction result. This reconstruction process re - integrates the prediction results of different frequencies back into the original space of the price data; the reconstruction equation of the prediction result is:

[0115]

[0116] where: represents the prediction value of the high - frequency part of the GARCH model; represents the prediction value of the low - frequency part of the ARIMA model; represents the prediction value of the residual part of the SVR model; represents the final silver price prediction value; ω 1 , ω 2 , ω 3 represents the weighting coefficients of the high - frequency, low - frequency, and residual terms, which are determined through cross - validation according to the performance of each term in historical data.

[0117] Specific evaluation of each model:

[0118] Use indicators such as mean squared error (MSE) and mean absolute error (MAE) to evaluate the prediction results. By comparing with the real data, adjust the weight coefficients of each sub-model to further improve the prediction accuracy. The evaluation index equation is:

[0119]

[0120] Where: y(t i ) represents the real value; represents the predicted value; n represents the data volume of the test set.

[0121] The prediction results of the final sequence test set are as Figure 4 shown, and the model evaluation structure is shown in Table 2.

[0122] Table 2 Model evaluation results

[0123] Model RMSE MAD MAPE ARIMA(3,1,1) 115.698 97.78156 0.01617577 ARIMA(3,1,1)-GARCH(1,1) 114.1239 65.08649 0.01016945 SVR 307.8694 104.7642 0.01520168 EMD-SVR 96.1389 67.35279 0.01065845 CEEMDAN-SVR 90.87568 63.41755 0.01028434 CEEMDAN-ARIMA-SVR 65.74189 44.74764 0.007441622 CEEMDAN-ARIMA-GARCH-SVR 65.40319 44.86695 0.00746442

[0124] From the comparative analysis in Table 2, it can be observed that without considering the data decomposition technology, the traditional time series model performs better than the support vector regression (SVR) model in terms of root mean square error (RMSE); this indicates that the traditional model has more advantages in prediction accuracy. The reason is that although SVR has strong adaptability in dealing with non-linear relationships, its performance largely depends on the proper selection of two key hyperparameters, and it is often difficult to find the optimal hyperparameter combination; thus, this limits the full play of the potential of the SVR model. In contrast, due to its simplicity, the traditional time series model can often provide more stable prediction results in practical applications.

[0125] When the CEEMDAN algorithm is introduced to decompose the data, and combined with the ARIMA and GARCH models and SVR for modeling, the constructed CEEMDAN-ARIMA-GARCH-SVR model shows the smallest RMSE value among all the compared models; this result highlights the significant advantage of the method of the present invention in prediction accuracy. By decomposing the original complex data into different frequency components, that is, decomposing the data set into high-frequency, low-frequency, and residual sequences, and modeling these sequences separately, it can effectively simplify the processing of complex signals, thereby achieving more accurate predictions; and the method of the present invention improves the accuracy and reliability of predictions by decomposing and simplifying the data.

[0126] An embodiment of the present invention also provides an electronic device, including a memory and a processor.

[0127] The memory is used to store computer programs.

[0128] When the processor is used to execute the computer program stored in the memory, the steps of a futures price prediction method as described above are implemented.

[0129] An embodiment of the present invention further provides a computer-readable storage medium for storing a computer program, and when the computer program is executed by a processor, the steps of a futures price prediction method as described above are implemented.

[0130] The above embodiments merely represent several implementation manners of the present invention, and the description thereof is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the invention patent shall be subject to the appended claims.

Claims

1. A futures price forecasting method, characterized in that: The following steps are involved: Obtain historical closing price data of futures and form time series data; The time series data is decomposed by CEEMDAN decomposition to obtain the intrinsic mode component IMFs subsequences and residual sequences with different center frequencies; the t-test is used to decompose the intrinsic mode component IMFs subsequences with different center frequencies into high-frequency sequences and low-frequency sequences; The generalized autoregressive conditional heteroskedasticity model GARCH is used to model the high-frequency series to obtain the high-frequency fluctuation forecast value of the time series data; the autoregressive integrated moving average model ARIMA is used to model the low-frequency series to obtain the low-frequency trend forecast value of the time series data; the support vector regression model SVR is used to model the residual series to obtain the nonlinear residual forecast value of the time series data; Reconstruct the price forecast value of futures based on the high-frequency volatility forecast value, low-frequency trend forecast value and nonlinear residual forecast value of time series data.

2. A futures price forecasting method according to claim 1, characterized in that: The method of using the t-test to decompose the intrinsic mode component IMFs subsequences of different center frequencies into high-frequency sequences and low-frequency sequences includes: The t-statistic value of the intrinsic mode component IMFs subsequence with different center frequencies is obtained, and the formula is: in: represents the average number of intrinsic mode component IMFs subsequences with different center frequencies; n represents the number of intrinsic mode component IMFs subsequences with different center frequencies; s represents the sample standard deviation; The significance level threshold is set. When the t-statistic value deviates significantly from zero, the intrinsic mode component IMFs subsequence of this frequency is set as a high-frequency sequence, and the significance of the high-frequency sequence is enhanced according to the fluctuation percentage condition; when the t-statistic value is not significant, the intrinsic mode component IMFs subsequence of this frequency is set as a low-frequency sequence.

3. A futures price forecasting method according to claim 1, characterized in that: The generalized autoregressive conditional heteroskedasticity model GARCH is used to model the high frequency series, including: After obtaining the high-frequency sequence, the generalized autoregressive conditional heteroskedasticity model GARCH is used to model the high-frequency sequence. The generalized autoregressive conditional heteroskedasticity model GARCH extracts the error term variance and volatility of the high-frequency sequence in the past, and obtains the volatility of the high-frequency sequence at the current moment, so as to capture the volatility clustering phenomenon in the futures time series data. The modeling equation of the generalized autoregressive conditional heteroskedasticity model GARCH is: y t =μ+e t e t =s t z t Where: y t Represents the high-frequency forecast value of futures prices; ε t represents the error term; represents volatility; z t represents independent and identically distributed random variables; α0, α1 and β1 represent the parameters of the GARCH model; μ represents a constant term.

4. A futures price forecasting method according to claim 1, characterized in that: The autoregressive integrated moving average model ARIMA is used to model the low-frequency series, including: After obtaining the low-frequency sequence, the autoregressive integrated moving average model ARIMA is used to model the stationary shuffle sequence; First, the difference term in the autoregressive integrated moving average model ARIMA is used to convert the non-stationary low-frequency sequence into a stationary low-frequency sequence; then the autoregressive AR in the autoregressive integrated moving average model ARIMA is used to extract the long-term trend between the stationary low-frequency sequence at the current moment and the stationary low-frequency sequence at the past moment; the moving average MA in the autoregressive integrated moving average model ARIMA is used to extract the periodic fluctuation between the stationary low-frequency sequence at the current moment and the stationary low-frequency sequence at the past moment; The modeling equation of the autoregressive integrated moving average model ARIMA is: y t =c+φ1y t-1 +φ2y t-2 +L+φ p y t-p +θ1ε t-1 +L+θ q e t-q +e t Where: y t Represents the low-frequency part of the futures price forecast value; φ1, φ2, L, φ p represents the autoregressive coefficient; θ1, θ2, L, θ q represents the moving average coefficient; ε t represents the error term; c represents the constant term.

5. A futures price forecasting method according to claim 1, characterized in that: The support vector regression model SVR is used to model the residual sequence, including: After obtaining the residual sequence, the support vector regression model SVR extracts the nonlinear residual between the nonlinear sequence data and the residual part in the residual sequence; The modeling equation of the support vector regression model SVR is: f(x)=z T φ(x)+b in: represents a nonlinear function that maps points in the input space to a higher-dimensional output space; represents the weight vector; b represents the bias.

6. A futures price forecasting method according to claim 1, characterized in that: The reconstructed futures price forecast value includes: The weighting coefficients ω1, ω2, ω3 of the high-frequency fluctuation prediction value, the low-frequency trend prediction value and the nonlinear residual prediction value are determined by cross-validation method; According to the weighted coefficients of the high-frequency volatility forecast value, the low-frequency trend forecast value and the nonlinear residual forecast value, the futures price is reconstructed according to the corresponding frequency band weights. The reconstruction equation is: in: Represents the high-frequency fluctuation forecast value of the GARCH model; Represents the low-frequency trend forecast value of the ARIMA model; Represents the nonlinear residual prediction value of the SVR model; Represents the final futures price forecast value.

7. A futures price prediction system, characterized in that: include: The data acquisition module is used to obtain the historical closing price data of futures and form time series data; The signal decomposition module is used to decompose the time series data using CEEMDAN decomposition to obtain the intrinsic mode component IMFs subsequences and residual sequences of different center frequencies; the t-test is used to decompose the intrinsic mode component IMFs subsequences of different center frequencies into high-frequency sequences and low-frequency sequences; The multi-model modeling module is used to model high-frequency sequences using the generalized autoregressive conditional heteroskedasticity model GARCH to obtain high-frequency fluctuation forecasts for time series data; to model low-frequency sequences using the autoregressive integrated moving average model ARIMA to obtain low-frequency trend forecasts for time series data; and to model residual sequences using the support vector regression model SVR to obtain nonlinear residual forecasts for time series data. The prediction reconstruction module is used to reconstruct the price forecast value of futures based on the high-frequency fluctuation forecast value, low-frequency trend forecast value and nonlinear residual forecast value of time series data.

8. An electronic device, characterized in that: include: Memory and processor; The memory is used to store computer programs; The processor is used to implement the steps of a futures price forecasting method as described in any one of claims 1 to 6 when executing the computer program stored in the memory.

9. A computer-readable storage medium, characterized in that: Used to store a computer program, which, when executed by a processor, implements the steps of a futures price forecasting method as described in any one of claims 1 to 6.

Images