Prediction method based on multi-channel time sequence and related equipment

Abstract

The invention provides a prediction method based on a multi-channel time sequence and related equipment. The method comprises the following steps: in response to a received prediction instruction, determining a prediction object specified by the prediction instruction, and obtaining multi-channel time sequence data related to the prediction object; extracting main frequency components and dynamic characteristics from the multi-channel time sequence data by using multi-channel adaptive Fourier decomposition; and constructing input data based on the main frequency components and the dynamic features, inputting the input data into a prediction model which corresponds to the prediction object and is obtained based on recurrent neural network training, and obtaining state information of the prediction object at multiple time points in the future based on output data of the prediction model. According to the method, the prediction precision of the model can be improved on the premise of not increasing the complexity and calculation cost of the model.

Description

Technical Field

[0001] The present application belongs to the technical field of signal processing and prediction analysis, and in particular, relates to a prediction method based on multi-channel time series and related equipment. Background Art

[0002] At present, the prediction methods based on time series have significant limitations when dealing with nonlinear and non-stationary data. In particular, traditional recurrent neural networks are difficult to effectively capture the correlation between complex multi-channel data. At the same time, the gradient vanishing problem is prone to occur when processing long sequence data, resulting in insufficient prediction accuracy. At present, there are related technologies to improve the prediction methods based on recurrent neural networks to improve the prediction accuracy. However, these improved methods often increase the complexity and computational cost of the model, and these improved methods still have great limitations in feature extraction and model adaptability. Summary of the invention

[0003] The present application provides a prediction method and related equipment based on multi-channel time series, aiming to solve the deficiencies or defects of the above-mentioned related technologies.

[0004] In a first aspect, the present application provides a prediction method based on a multi-channel time series, the method comprising:

[0005] In response to receiving a prediction instruction, determining a prediction object specified by the prediction instruction, and acquiring multi-channel time series data related to the prediction object;

[0006] Extracting main frequency components and dynamic features from the multi-channel time series data using multi-channel adaptive Fourier decomposition;

[0007] Input data is constructed based on the main frequency components and dynamic features, and the input data is input into the prediction model corresponding to the prediction object obtained by training the recurrent neural network. Based on the output data of the prediction model, the state information of the prediction object at multiple time points in the future is obtained.

[0008] In some embodiments, in response to receiving the prediction instruction, before determining the prediction object specified by the prediction instruction, the method further includes:

[0009] Build an initial prediction model based on a recurrent neural network;

[0010] Obtaining a training sample set related to the prediction object; the training sample set includes multiple training samples; each training sample includes a historical main frequency component and a historical dynamic feature extracted from a historical multi-channel time series data corresponding to the prediction object;

[0011] The training samples in the training sample set are called to iteratively train the initial prediction model until the preset end condition is met to obtain a trained prediction model.

[0012] In some embodiments, extracting main frequency components and dynamic features from the multi-channel time series data using multi-channel adaptive Fourier decomposition includes:

[0013] Multi-channel adaptive Fourier decomposition is used to perform frequency decomposition on the time series data of each channel to extract the main frequency components and dynamic features in each channel. In the process of frequency decomposition of the time series data of each channel, common basis functions are aligned between multiple channels to explore the correlation between channels.

[0014] In some embodiments, obtaining multi-channel time series data related to the prediction object includes:

[0015] Obtain the original time series data related to the forecast object;

[0016] The original time series data is preprocessed to obtain the multi-channel time series data; the preprocessing includes data cleaning and normalization processing.

[0017] In some embodiments, the prediction object is the meteorological data of the target city; the multi-channel time series data includes multi-channel data of the target city on a daily basis in the past multiple years, and the multi-channel data includes index data of multiple indicators including temperature, humidity, wind speed and precipitation; the main frequency component is information used to reflect periodic changes in the meteorological data, and the dynamic feature is information used to reflect short-term changes in the meteorological data; the state information of the prediction object at multiple time points in the future refers to the daily weather conditions of the target city within a preset period of time in the future.

[0018] In some embodiments, the prediction object is the stock price of multiple target stocks; the multi-channel time series data includes the stock price and trading volume of each target stock on each trading day in the past multiple years; the main frequency component is information used to reflect periodic changes in stock data, and the dynamic feature is information used to reflect short-term changes in stock data; the state information of the prediction object at multiple time points in the future refers to the stock price of each target stock on each trading day within a preset period of time in the future.

[0019] In some embodiments, the prediction object is the waveform of an electrocardiogram signal; the multi-channel time series data includes the electrocardiogram signals of each lead within a preset time period in the past; the main frequency component is information used to reflect the periodic changes of the electrocardiogram signal, and the dynamic feature is information used to reflect the short-term changes of the electrocardiogram signal; the state information of the prediction object at multiple time points in the future refers to the waveform of the electrocardiogram signal within a preset time period in the future.

[0020] In a second aspect, the present application provides a prediction device based on a multi-channel time series, the device comprising:

[0021] A prediction instruction response module, used for determining a prediction object specified by the prediction instruction in response to receiving the prediction instruction, and acquiring multi-channel time series data related to the prediction object;

[0022] An extraction module, for extracting main frequency components and dynamic features from the multi-channel time series data using multi-channel adaptive Fourier decomposition;

[0023] The prediction module is used to construct input data based on the main frequency components and dynamic features, input the input data into the prediction model obtained by training the recurrent neural network, and obtain the state information of the prediction object at multiple time points in the future based on the output data of the prediction model.

[0024] In a third aspect, the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the prediction method based on multi-channel time series provided in the above embodiment is implemented.

[0025] In a fourth aspect, the present application provides a computer device comprising: one or more processors; a memory; and one or more computer programs, wherein the processor and the memory are connected via a bus, wherein the one or more computer programs are stored in the memory and are configured to be executed by the one or more processors, and characterized in that when the processor executes the computer program, the prediction method based on multi-channel time series provided in the above-mentioned embodiment is implemented.

[0026] In this application, multi-channel adaptive Fourier decomposition and recurrent neural network are integrated in a unified framework, which can simplify the design and tuning process of complex time series prediction models without increasing model complexity and computational cost. Users can achieve efficient prediction without complex feature engineering, which improves the ease of use of the model. Among them, the main frequency components and dynamic features in the time series data are extracted from the multi-channel data through multi-channel adaptive Fourier decomposition, which improves the overall solution's ability to handle nonlinear and non-stationary data. The time series data can be preprocessed through multi-channel adaptive Fourier decomposition, which can reduce the demand for recurrent neural network dependence on long sequences, thereby alleviating the gradient vanishing problem and improving the performance of the model in long sequence prediction. The overall solution can flexibly handle high-dimensional, multi-channel data, ensuring that the model can still maintain high prediction accuracy and stability when facing data of different complexities. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is a flowchart of a prediction method based on multi-channel time series provided in one embodiment of the present application.

[0028] Figure 2It is a functional module block diagram of a prediction device based on multi-channel time series provided in one embodiment of the present application.

[0029] Figure 3 It is a specific structural block diagram of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0030] In order to make the purpose, technical solution and beneficial effects of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0031] In order to illustrate the technical solution described in this application, a specific embodiment is provided below for illustration.

[0032] See also Figure 1 , is a flowchart of a prediction method based on a multi-channel time series provided in one embodiment of the present application. This embodiment mainly illustrates the method by applying it to a computer device.

[0033] The prediction method based on multi-channel time series provided in an embodiment of the present application includes the following steps:

[0034] S101 . In response to receiving a prediction instruction, determining a prediction object specified by the prediction instruction, and acquiring multi-channel time series data related to the prediction object.

[0035] A prediction instruction refers to an instruction triggered by a user to instruct a computer device to perform a prediction task. The task content of the prediction task is indicated by the prediction instruction, and accordingly, the prediction object and the specific type of multi-channel time series data that needs to be obtained are determined by the task content of the prediction task. For example, if the task content of the prediction task is to predict meteorological data, then the prediction object can be the meteorological data of a certain city, and the multi-channel time series data that needs to be obtained can be the indicator data of multiple meteorological indicators such as daily temperature, humidity, wind speed and precipitation in the city in the past several years; for another example, if the task content of the prediction task is electrocardiogram waveform prediction, then the prediction object can be the waveform of the electrocardiogram signal, and the multi-channel time series data that needs to be obtained includes the electrocardiogram signals of each lead within the past preset time period.

[0036] In one embodiment, obtaining multi-channel time series data related to the prediction object includes: obtaining original time series data related to the prediction object; preprocessing the original time series data to obtain the multi-channel time series data; the preprocessing includes data cleaning and normalization.

[0037] Data cleaning can include missing data processing and outlier detection and processing. Normalization can be done by scaling the data to a specific range, such as [0,1], using mean normalization, minimum-maximum normalization, or Z-score normalization, so that the prediction model can be better generalized to new data.

[0038] This embodiment can ensure the quality and consistency of the time series data used for subsequent processing by performing a series of preprocessing on the original multi-channel time series signals, so that subsequent feature extraction operations can extract higher quality features, and further allow subsequent prediction operations to obtain more accurate prediction results.

[0039] S102, extracting main frequency components and dynamic features from the multi-channel time series data using multi-channel adaptive Fourier decomposition.

[0040] The main frequency components and dynamic features are extracted from the multi-channel time series data using multi-channel adaptive Fourier decomposition, including: using multi-channel adaptive Fourier decomposition to perform frequency decomposition on the time series data of each channel, and extracting the main frequency components and dynamic features in each channel; wherein, in the process of performing frequency decomposition on the time series data of each channel, common basis functions are aligned between multiple channels to explore the correlation between channels.

[0041] In this embodiment, Multiple Adaptive Fourier Decomposition (MFAD) uses adaptive basis functions to effectively capture the local characteristics of nonlinear and non-stationary signals, and in the frequency decomposition process, it mines the correlation between channels by aligning common basis functions between multiple channels. Based on the mined correlation between channels, the final prediction accuracy of the prediction model can be improved.

[0042] Multi-channel adaptive Fourier decomposition extends adaptive Fourier decomposition (AFD) to multi-channels. The adaptive Fourier decomposition itself will adaptively select the appropriate Fourier basis function, and then decompose the main frequency components and dynamic characteristics from the relevant time series data, and gradually extract the Fourier components by minimizing the reconstruction error until the error reaches a preset threshold.

[0043] The specific decomposition steps are as follows:

[0044] a. Initialize the parameters of the Fourier basis function, such as period and amplitude;

[0045] b. Use the least squares method to fit the time series data and obtain the initial Fourier component;

[0046] c. Dynamically adjust parameters and minimize reconstruction error through iterative optimization;

[0047] d. Extract the main Fourier components as input for subsequent recurrent neural network modeling.

[0048] Adaptive Fourier Decomposition Using Matching Pursuit Method to Identify Adaptive Basis Function Systems Each function B n It is called the modified Blaschke product. The modified Blaschke product of each basis function is defined as follows:

[0049]

[0050] Among them, a n is the complex coefficient in the unit disk D = {z∈C:|z|<1, and is an adaptive parameter chosen to maximize energy convergence at each decomposition level. For any sequence of a in the unit disk D n are orthogonal. The processed signal G(e jt ) can be expressed as the series expansion form of the adaptive basis function:

[0051]

[0052] Among them, A n For nth th These coefficients are obtained by combining the signal G and the basis function B n The inner product of <G,B n >Calculated.

[0053] In multi-channel adaptive Fourier decomposition, all channels use the same number of single components and the same basis functions to maintain the alignment of common oscillations among channels. The total energy to be maximized in multi-channel adaptive Fourier decomposition is:

[0054]

[0055] Where C is the total number of channels, G c,n represents the processed signal of the cth channel in the nth decomposition level. This formula is used to determine the optimal parameter a that maximizes the energy of all channels n .

[0056] This embodiment uses multi-channel adaptive Fourier decomposition for feature extraction, which can accurately separate the main frequency components and dynamic features in the signal, thereby providing high-quality features for the prediction model built based on the recurrent neural network, so that the prediction model can more accurately capture the dynamic changes of complex time series data, and then output more accurate prediction results. Furthermore, multi-channel adaptive Fourier decomposition can adaptively select the decomposition basis function according to the characteristics of the relevant time series data, avoiding the limitations of traditional fixed basis functions in processing nonlinear and non-stationary signals. This adaptive ability can make the final extracted features more representative, which will further improve the prediction performance of the prediction model.

[0057] S103, constructing input data based on the main frequency components and dynamic features, and inputting the input data into a prediction model corresponding to the prediction object obtained by training a recurrent neural network, and obtaining state information of the prediction object at multiple future time points based on the output data of the prediction model.

[0058] Among them, the main frequency components and dynamic features can be spliced ​​into a comprehensive feature as the input data of the prediction model.

[0059] The prediction model is pre-trained, and its training process may include:

[0060] (1) Building an initial prediction model based on a recurrent neural network;

[0061] (2) obtaining a training sample set related to the prediction object; the training sample set includes a plurality of training samples; each training sample includes a historical main frequency component and a historical dynamic feature extracted from a historical multi-channel time series data corresponding to the prediction object;

[0062] (3) Calling the training samples in the training sample set to iteratively train the initial prediction model until the preset end condition is met to obtain a trained prediction model.

[0063] When training the prediction model, this embodiment uses the main frequency components and dynamic features extracted by multi-channel adaptive Fourier decomposition as the input of the prediction model to train the prediction model. The prediction model can capture short-term and long-term dependencies from the input data through a loop structure, thereby realizing prediction. In the training process, back propagation algorithms (such as BPTT, Backpropagation Through Time, etc.) and gradient descent algorithms can be used to ensure the convergence and stability of the prediction model. At the same time, the hyperparameters of the model can be optimized by methods such as cross-validation to avoid overfitting.

[0064] Furthermore, in order to cope with the dynamic changes of data, this embodiment also provides a mechanism for dynamically updating the prediction model, that is, after the prediction model is trained, the prediction model is allowed to make predictions based on new multi-channel time series data. This mechanism allows the multi-channel adaptive Fourier decomposition and prediction model to be continuously updated and self-adjusted to ensure the prediction accuracy of the model in different time periods.

[0065] The prediction method provided in the above embodiment will be described in detail below using a plurality of different target prediction tasks as examples.

[0066] In one embodiment, the task content of the forecast task indicated by the forecast instruction is to forecast meteorological data, and the forecast object is the meteorological data of the target city; the target city is a city specified by the user. Accordingly, the acquired multi-channel time series data includes multi-channel data of the target city for each day in the past multiple years, and the multi-channel data includes index data of multiple indexes in temperature, humidity, wind speed and precipitation; the main frequency components extracted by multi-channel adaptive Fourier decomposition are information used to reflect the periodic changes in the meteorological data, and the dynamic features are information used to reflect the short-term changes in the meteorological data; the state information of the forecast object at multiple time points in the future specifically refers to the daily weather conditions of the target city within a preset time period in the future.

[0067] In this embodiment, a training data set generated based on daily meteorological data (including data from multiple channels such as temperature, humidity, wind speed, precipitation, etc.) of a city in multiple years can be first constructed, and a prediction model based on a recurrent neural network can be trained based on the training data set. The prediction model will capture the short-term and long-term dependencies implied by the time series data in the training samples through its cyclic structure, and then predict the weather conditions for a period of time in the future. Among them, the input data of the prediction model is the main frequency components and dynamic features extracted from various meteorological indicators (such as temperature, humidity, wind speed, precipitation) using multi-channel adaptive Fourier decomposition. The main frequency components in this embodiment refer to periodic changes in meteorological data, such as seasonal temperature fluctuations, periodic precipitation, etc., which can reflect long-term regular information, and dynamic features refer to short-term changes in meteorological data, such as information that characterizes sudden rainfall, temperature drops, etc.

[0068] In the prediction method provided in this embodiment, the frequency features extracted by multi-channel adaptive Fourier decomposition combined with the sequence modeling capability of recurrent neural networks can enable the prediction model to more accurately capture the complex dynamics in weather patterns, and thus accurately predict future meteorological trends. In particular, under the joint action of multi-channel data (such as temperature, humidity, and precipitation), the prediction accuracy can be significantly improved.

[0069] In another embodiment, the task content of the prediction task indicated by the prediction instruction is stock price prediction. Accordingly, the prediction object is specifically the stock price of multiple target stocks; the acquired multi-channel time series data includes the stock price and trading volume of each target stock on each trading day in the past multiple years; the main frequency components extracted by multi-channel adaptive Fourier decomposition are used to reflect the information of periodic changes in stock data, and the dynamic features are used to reflect the information of short-term changes in stock data; the state information of the prediction object at multiple time points in the future refers to the stock price of each target stock on each trading day within a preset time period in the future.

[0070] In this embodiment, a training data set is constructed in advance based on historical stock market data, and the training data set contains the stock prices (including opening price, closing price, highest price and lowest price, etc.) and trading volume of multiple companies on each stock trading day in the past multiple years; then, based on the training samples in the training data set (each training sample includes the stock prices and trading volumes of multiple companies on each day in the past period of time), a prediction model constructed based on a recurrent neural network is trained, and the prediction model will predict the future stock prices of each company by learning the long-term dependencies and short-term fluctuations of time series signals. Among them, the input data of the prediction model is the main frequency components and dynamic features of the stock price and trading volume extracted by multi-channel adaptive Fourier decomposition, and the output data is the stock price information of each company on each stock trading day in the next week.

[0071] This embodiment can extract high-quality features that can better reflect the cyclical fluctuations of the stock market from stock market data through multi-channel adaptive Fourier decomposition. The prediction model built based on the recurrent neural network can better capture the dynamic changes of stock prices based on these features and achieve higher prediction accuracy.

[0072] In another embodiment, the task content of the prediction task indicated by the prediction instruction is electrocardiogram waveform prediction, the prediction object is the waveform of the electrocardiogram signal, the multi-channel time series data to be acquired includes the electrocardiogram signals of each lead within a preset time period in the past, the main frequency components extracted by multi-channel adaptive Fourier decomposition are information used to reflect the periodic changes of the electrocardiogram signal (such as P wave, QRS complex and T wave), and the dynamic characteristics are information used to reflect the short-term changes of the electrocardiogram signal; the state information of the prediction object at multiple time points in the future refers to the waveform of the electrocardiogram signal within a preset time period in the future.

[0073] This embodiment uses the multi-channel decomposition capability of the multi-channel adaptive Fourier decomposition to dig out the correlation between the ECG signals of each lead, thereby making effective use of it, and ultimately improving the prediction accuracy of the future ECG signal waveform.

[0074] The present application also provides a prediction device based on multi-channel time series. In some embodiments, see Figure 2 , the device comprises:

[0075] A prediction instruction response module 10 is used to determine a prediction object specified by the prediction instruction in response to receiving the prediction instruction, and obtain multi-channel time series data related to the prediction object;

[0076] An extraction module 20, for extracting main frequency components and dynamic features from the multi-channel time series data using multi-channel adaptive Fourier decomposition;

[0077] The prediction module 30 is used to construct input data based on the main frequency components and dynamic features, input the input data into a prediction model obtained by training a recurrent neural network, and obtain the state information of the prediction object at multiple time points in the future based on the output data of the prediction model.

[0078] In some embodiments, the device further includes a prediction model training module. The prediction model training module includes:

[0079] A model building unit, used to build an initial prediction model based on a recurrent neural network;

[0080] A sample acquisition unit is used to acquire a training sample set related to the prediction object; the training sample set includes multiple training samples; each training sample includes a historical main frequency component and a historical dynamic feature extracted from a historical multi-channel time series data corresponding to the prediction object;

[0081] The iterative training unit is used to call the training samples in the training sample set to iteratively train the initial prediction model until a preset end condition is met to obtain a trained prediction model.

[0082] In some embodiments, the extraction module includes:

[0083] The frequency decomposition unit is used to perform frequency decomposition on the time series data of each channel using multi-channel adaptive Fourier decomposition, and extract the main frequency components and dynamic features in each channel; wherein, in the process of performing frequency decomposition on the time series data of each channel, common basis functions are aligned between multiple channels to explore the correlation between channels.

[0084] In some embodiments, the predicted instruction response module 10 includes:

[0085] The original data acquisition unit is used to acquire the original time series data related to the prediction object;

[0086] The preprocessing unit is used to preprocess the original time series data to obtain the multi-channel time series data; the preprocessing includes data cleaning and normalization processing.

[0087] In some embodiments, the prediction object is the meteorological data of the target city; the multi-channel time series data includes multi-channel data of the target city on a daily basis in the past multiple years, and the multi-channel data includes index data of multiple indicators including temperature, humidity, wind speed and precipitation; the main frequency component is information used to reflect periodic changes in the meteorological data, and the dynamic feature is information used to reflect short-term changes in the meteorological data; the state information of the prediction object at multiple time points in the future refers to the daily weather conditions of the target city within a preset period of time in the future.

[0088] In some embodiments, the prediction object is the stock price of multiple target stocks; the multi-channel time series data includes the stock price and trading volume of each target stock on each trading day in the past multiple years;

[0089] The main frequency component is used to reflect the periodic changes in stock data, and the dynamic characteristics are used to reflect the short-term changes in stock data; the state information of the prediction object at multiple time points in the future refers to the stock price of each target stock on each trading day within a preset period of time in the future.

[0090] In some embodiments, the prediction object is the waveform of an electrocardiogram signal; the multi-channel time series data includes the electrocardiogram signals of each lead within a preset time period in the past; the main frequency component is information used to reflect the periodic changes of the electrocardiogram signal, and the dynamic feature is information used to reflect the short-term changes of the electrocardiogram signal; the state information of the prediction object at multiple time points in the future refers to the waveform of the electrocardiogram signal within a preset time period in the future.

[0091] The prediction device based on multi-channel time series provided in one embodiment of the present application and the prediction method based on multi-channel time series provided in the present application belong to the same concept. The specific implementation process is detailed in the full text of the specification and will not be repeated here.

[0092] An embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the prediction method based on multi-channel time series provided in an embodiment of the present application is implemented.

[0093] Figure 3A specific structural block diagram of a computer device provided in an embodiment of the present application is shown. The computer device 100 includes: one or more processors 101, a memory 102, and one or more computer programs, wherein the processor 101 and the memory 102 are connected via a bus, the one or more computer programs are stored in the memory 102, and are configured to be executed by the one or more processors 101, and when the processor 101 executes the computer program, a prediction method based on a multi-channel time series as provided in an embodiment of the present application is implemented.

[0094] It should be understood that each step in each embodiment of the present application is not necessarily performed in sequence according to the order indicated by the step number. Unless there is clear explanation in this article, the execution of these steps does not have strict order restriction, and these steps can be performed in other orders. Moreover, at least a part of the steps in each embodiment can include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these sub-steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a part of other steps or sub-steps or stages of other steps.

[0095] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink), DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0096] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0097] The above embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for those of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A prediction method based on multi-channel time series, characterized in that: The method comprises: In response to receiving a prediction instruction, determining a prediction object specified by the prediction instruction, and acquiring multi-channel time series data related to the prediction object; extracting main frequency components and dynamic features from the multi-channel time series data using multi-channel adaptive Fourier decomposition; Input data is constructed based on the main frequency components and the dynamic features, and the input data is input into a prediction model corresponding to the prediction object obtained by training a recurrent neural network, and state information of the prediction object at multiple future time points is obtained based on the output data of the prediction model.

2. The method according to claim 1, characterized in that In response to receiving the prediction instruction, before determining the prediction object specified by the prediction instruction, the method further includes: Build an initial prediction model based on a recurrent neural network; Acquire a training sample set related to the prediction object; the training sample set includes a plurality of training samples; each training sample includes a historical main frequency component and a historical dynamic feature extracted from a historical multi-channel time series data corresponding to the prediction object; The training samples in the training sample set are called to iteratively train the initial prediction model until a preset end condition is met to obtain a trained prediction model.

3. The method according to claim 1, characterized in that The main frequency components and dynamic features are extracted from the multi-channel time series data using multi-channel adaptive Fourier decomposition, including: Multi-channel adaptive Fourier decomposition is used to perform frequency decomposition on the time series data of each channel to extract the main frequency components and dynamic features in each channel. In the process of frequency decomposition of the time series data of each channel, common basis functions are aligned between multiple channels to explore the correlation between channels.

4. The method according to claim 1, characterized in that Acquire multi-channel time series data related to the prediction object, including: Acquire original time series data related to the prediction object; The original time series data is preprocessed to obtain the multi-channel time series data; the preprocessing includes data cleaning and normalization processing.

5. The method according to claim 1, characterized in that The forecast object is the meteorological data of the target city; The multi-channel time series data includes multi-channel data of each day of the target city in the past multiple years, and the multi-channel data includes index data of multiple indexes including temperature, humidity, wind speed and precipitation; The main frequency component is information used to reflect periodic changes in meteorological data, and the dynamic feature is information used to reflect short-term changes in meteorological data; The state information of the predicted object at multiple time points in the future refers to the daily weather conditions of the target city within a preset time period in the future.

6. The method according to claim 1, characterized in that The prediction object is the stock prices of multiple target stocks; The multi-channel time series data includes the stock price and trading volume of each target stock on each trading day in the past multiple years; The main frequency component is information used to reflect periodic changes in stock data, and the dynamic feature is information used to reflect short-term changes in stock data; The state information of the prediction object at multiple time points in the future refers to the stock price of each target stock on each trading day within a preset time period in the future.

7. The method according to claim 1, characterized in that The prediction object is the waveform of the electrocardiogram signal; The multi-channel time series data includes the electrocardiogram signals of each lead within a preset time period in the past; The main frequency component is information for reflecting the periodic changes of the electrocardiogram signal, and the dynamic feature is information for reflecting the short-term changes of the electrocardiogram signal; The state information of the predicted object at multiple time points in the future refers to the waveform of the electrocardiogram signal within a preset time period in the future.

8. A prediction device based on multi-channel time series, characterized in that: The device comprises: A prediction instruction response module, configured to, in response to receiving a prediction instruction, determine a prediction object specified by the prediction instruction, and acquire multi-channel time series data related to the prediction object; An extraction module, for extracting main frequency components and dynamic features from the multi-channel time series data using multi-channel adaptive Fourier decomposition; A prediction module is used to construct input data based on the main frequency components and the dynamic features, input the input data into a prediction model obtained by training a recurrent neural network, and obtain the state information of the predicted object at multiple time points in the future based on the output data of the prediction model.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the prediction method based on multi-channel time series according to any one of claims 1 to 7 are implemented.

10. A computer device, characterized in that: include: one or more processors; Memory; as well as One or more computer programs, the processor and the memory are connected via a bus, wherein the one or more computer programs are stored in the memory and are configured to be executed by the one or more processors, wherein the processor implements the steps of the prediction method based on multi-channel time series as described in any one of claims 1 to 7 when executing the computer program.

Images