Prediction method based on MAFD and LSTM and related equipment
By using multi-channel adaptive Fourier decomposition in long- and short-term memory networks to feature extraction of multi-channel time series signals, the problem that long- and short-term memory networks are difficult to capture channel correlation and gradient disappearance when processing multi-channel data is solved, and higher prediction accuracy is achieved.
Patent Information
- Application Number
- CN202411828193.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-12
- Publication Date
- 2025-05-13
AI Technical Summary
Long and short-term memory networks are difficult to effectively capture the correlation between channels when processing multi-channel time series signals, and may encounter gradient vanishing problems when processing long sequences, resulting in limited prediction accuracy.
Multi-channel adaptive Fourier decomposition is used to extract features of multi-channel time series signals, extract main frequency components and non-stationary features, and input these features to predict trends based on the prediction model trained by long and short-term memory networks.
Through multi-channel adaptive Fourier decomposition, it effectively captures the correlation between channels, provides high-quality feature input, improves the prediction accuracy of long and short-term memory networks, and thus improves the final prediction accuracy.
Smart Images
Figure CN119988923A_ABST
Abstract
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 MAFD and LSTM and related equipment. Background Art
[0002] In signal processing and predictive analysis, accurate time series prediction is crucial for many fields, such as financial markets, weather forecasting, industrial monitoring, etc. Long short-term memory networks are widely used in time series prediction due to their superiority in processing sequence data. However, although long short-term memory networks can process nonlinear data, they have difficulty in effectively capturing the correlation between channels when processing multi-channel data, and may encounter the gradient vanishing problem when processing long sequences, resulting in limited prediction accuracy. Summary of the invention
[0003] The present application provides a prediction method and related equipment based on MAFD and LSTM, 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 MAFD and LSTM, the method comprising:
[0005] Obtain multi-channel time series signals related to the target prediction task;
[0006] Multi-channel adaptive Fourier decomposition is used to extract features of the multi-channel time series signal and obtain the main frequency components and non-stationary features;
[0007] The main frequency components and non-stationary features are input as input data into the prediction model trained based on the long short-term memory network to obtain the model output data;
[0008] Determine prediction results based on the model output data.
[0009] In some embodiments, feature extraction is performed on the multi-channel time series signal using multi-channel adaptive Fourier decomposition, including:
[0010] Multi-channel adaptive Fourier decomposition is used to perform frequency decomposition on the time series signals of each channel to extract the main frequency components and non-stationary features in each channel.
[0011] In some embodiments, the training process of the prediction model includes:
[0012] Acquire a historical multi-channel time series signal set, where the historical time series signal includes multiple groups of preprocessed historical multi-channel time series signals;
[0013] Use multi-channel adaptive Fourier decomposition to extract historical main frequency components and non-stationary features from each group of historical multi-channel time series signals, and build a training data set based on the features extracted from each group of historical multi-channel time series signals;
[0014] The prediction model built based on the long short-term memory network is iteratively trained using the training data set until the preset end condition is met to obtain a trained prediction model.
[0015] In some embodiments, before acquiring the multi-channel time series signal related to the target prediction task, the method further includes:
[0016] Get the original multi-channel time series signal;
[0017] Preprocessing the time series signals of each channel of the original multi-channel time series signal, including data cleaning, standardization, filtering and noise reduction;
[0018] Performing data alignment on the time series signals of each channel based on the timestamps of the preprocessed time series signals of each channel;
[0019] A multi-channel time series signal is obtained based on the aligned time series signals of each channel.
[0020] In some embodiments, obtaining a multi-channel time series signal based on the aligned time series signals of each channel includes:
[0021] Based on the data enhancement technology, the signal-to-noise ratio of the time series signals of each channel after alignment is improved to obtain a multi-channel time series signal.
[0022] In some embodiments, determining a prediction result based on the model output data includes:
[0023] The prediction results are generated by integrating the model output data with the residuals of the multi-channel time series signals.
[0024] In some embodiments, the prediction result is generated by integrating the model output data with the residual of the multi-channel time series signal, including:
[0025] The model output data and the periodic components in the multi-channel time series signal are reconstructed into the time series of the original data, and the residual of the model output data is added to obtain the prediction result.
[0026] In a second aspect, the present application provides a prediction device based on MAFD and LSTM, the device comprising:
[0027] A signal acquisition module is used to acquire multi-channel time series signals related to the target prediction task;
[0028] A feature extraction module is used to extract features of the multi-channel time series signal using multi-channel adaptive Fourier decomposition to obtain main frequency components and non-stationary features;
[0029] The trend prediction module is used to input the main frequency components and non-stationary features as input data into the prediction model obtained by long short-term memory network training to obtain model output data;
[0030] The prediction result generation module is used to determine the prediction result based on the model output data.
[0031] 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 MAFD and LSTM provided in the above embodiment is implemented.
[0032] In a fourth aspect, the present application provides a computer device comprising: one or more processors; a memory; and 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, characterized in that when the processor executes the computer program, the prediction method based on MAFD and LSTM provided in the above-mentioned embodiment is implemented.
[0033] The present application can effectively combine the powerful signal decomposition capability of multi-channel adaptive Fourier decomposition with the prediction capability of long short-term memory network to process complex multi-channel time series signals, thereby improving the prediction accuracy. Specifically, a multi-channel time series signal related to the target prediction task is obtained, and then the multi-channel adaptive Fourier decomposition is used to extract features of the multi-channel time series signal to obtain the main frequency components and non-stationary features, and then the main frequency components and non-stationary features are input as input data into the prediction model obtained by training the long short-term memory network for trend prediction, and the model output result (i.e., the trend prediction result) is obtained, wherein the multi-channel adaptive Fourier decomposition can effectively capture the correlation between channels, and can provide high-quality features for the prediction model, which helps to improve the trend prediction accuracy, thereby improving the final prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] Figure 1 This is a flowchart of a prediction method based on MAFD and LSTM provided in one embodiment of the present application.
[0035] Figure 2 It is a functional module block diagram of a prediction device based on MAFD and LSTM provided in one embodiment of the present application.
[0036] Figure 3It is a specific structural block diagram of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION
[0037] 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.
[0038] In order to illustrate the technical solution described in this application, a specific embodiment is provided below for illustration.
[0039] See also Figure 1 , is a flowchart of a prediction method based on MAFD and LSTM provided in one embodiment of the present application. This embodiment mainly illustrates the method by applying it to a computer device.
[0040] The prediction method based on MAFD and LSTM provided in an embodiment of the present application includes the following steps:
[0041] S101. Acquire a multi-channel time series signal related to a target prediction task.
[0042] The multi-channel time series signal obtained in this step includes time series signals of multiple channels (types), and the specific content of the multi-channel time series signal is related to the task objective of the target prediction task.
[0043] In one embodiment, before acquiring the multi-channel time series signal related to the target prediction task, the method further includes:
[0044] a. Obtain the original multi-channel time series signal; the original multi-channel time series signal refers to the newly acquired unprocessed multi-channel time series signal.
[0045] b. Preprocess the time series signals of each channel of the original multi-channel time series signal, and the preprocessing includes data cleaning, standardization, filtering and noise reduction.
[0046] In order to ensure the quality and consistency of the signal used for subsequent processing, thereby improving the accuracy and stability of the subsequent signal decomposition and prediction model, a series of preprocessing is required for the original multi-channel time series signal. Preprocessing specifically includes data cleaning, standardization, filtering and noise reduction.
[0047] (1) Data cleaning can include missing data processing and outlier detection and processing.
[0048] Regarding missing data processing, in practical applications, the time series signals of each channel collected may have missing values, which may affect the effect of signal decomposition and prediction model. For missing data, interpolation methods (such as linear interpolation, spline interpolation) or filling methods (such as mean filling, front and back value filling) can be used to fill them. For severely missing data segments, you can consider discarding or marking them.
[0049] Regarding outlier detection and processing, outliers refer to extreme data points that are significantly different from normal signals. They may be caused by sensor failure or environmental interference. When processing, statistical methods (such as Z-score, box plot analysis) or machine learning methods (such as isolation forest algorithm) can be used to detect outliers in the signal. For detected outliers, they can be processed by smoothing or replacing them with the mean of neighboring points.
[0050] (2) Data normalization involves scaling the data to a specific range, such as [0, 1], using methods such as mean normalization, minimum-maximum normalization, and Z-score normalization, which helps the prediction model better generalize to new data.
[0051] (3) Regarding filtering and noise reduction, the signal can be bandpass filtered to remove low-frequency noise and high-frequency interference in the signal according to the needs of the actual application scenario. This operation can be achieved by designing a suitable bandpass filter (such as Butterworth filter, Chebyshev filter) to retain the main frequency components in the signal and improve the signal quality. In order to remove random noise in the signal, methods such as wavelet transform (Wavelet Transform) or empirical mode decomposition (EMD) can be used to perform noise reduction on the signal to reduce the impact of noise on subsequent signal decomposition and prediction operations without damaging the main components of the signal.
[0052] c. Performing data alignment on the time series signals of each channel based on the timestamps of the preprocessed time series signals of each channel.
[0053] After preprocessing, it is also necessary to ensure that the timestamps of the time series signals of all channels are synchronized. You can first detect whether there is a time offset between the time series signals of different channels. If so, data alignment is required to ensure that the signals of all channels reflect the same event or phenomenon at the same time point.
[0054] In addition, you can also truncate or pad the multi-channel time series signals to a uniform length to ensure that the data length of all channels is consistent. If the signals of some channels are shorter than those of other channels, you can pad them by repeating the last value or linearly extrapolating to make the signal length of all channels the same.
[0055] d. Obtain a multi-channel time series signal based on the aligned time series signals of each channel.
[0056] In one embodiment, after the data alignment is completed, the time series signals of each channel obtained after the data alignment can be directly used as the multi-channel time series signal.
[0057] In another embodiment, a multi-channel time series signal is obtained based on the aligned time series signals of each channel, including: improving the signal-to-noise ratio of the aligned time series signals of each channel based on data enhancement technology to obtain the multi-channel time series signal.
[0058] After data alignment, the time series signals of each channel obtained after data alignment are further enhanced. The signal-to-noise ratio of the time series signals of each channel is improved through means such as smoothing and signal amplification, making the features in the signal more obvious, providing a better quality input signal for subsequent feature decomposition operations.
[0059] S102, using multi-channel adaptive Fourier decomposition to extract features from the multi-channel time series signal to obtain main frequency components and non-stationary features.
[0060] The multi-channel time series signal is subjected to feature extraction by using multi-channel adaptive Fourier decomposition, including: performing frequency decomposition on the time series signal of each channel by using multi-channel adaptive Fourier decomposition, and extracting the main frequency components and non-stationary features in each channel.
[0061] Multiple Adaptive Fourier Decomposition (MFAD) can effectively capture the local characteristics of nonlinear and non-stationary signals through adaptive basis functions, and at the same time, it can mine the correlation between channels by aligning common basis functions between multiple channels.
[0062] Specifically, multi-channel adaptive Fourier decomposition is to extend adaptive Fourier decomposition (AFD) to the multi-channel case. Adaptive Fourier decomposition can decompose the main trends and periodic fluctuations in time series data by adaptively selecting appropriate Fourier basis functions, and can gradually extract Fourier components by minimizing the reconstruction error until the error reaches a preset threshold.
[0063] The decomposition steps include:
[0064] a. Initialize the parameters of the Fourier basis function, such as period and amplitude;
[0065] b. Use the least squares method to fit the time series data and obtain the initial Fourier component;
[0066] c. Dynamically adjust parameters and minimize reconstruction error through iterative optimization;
[0067] d. Extract the main Fourier components as input for subsequent recurrent neural network modeling.
[0068] 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:
[0069]
[0070] 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:
[0071]
[0072] 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.
[0073] In multi-channel adaptive Fourier decomposition (MAFD), 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:
[0074]
[0075] 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 .
[0076] This embodiment uses multi-channel adaptive Fourier decomposition for feature extraction, which can accurately separate the main frequency components and non-stationary features in the signal, thereby providing high-quality feature input for the subsequent prediction model. This enables the prediction model to more accurately capture the dynamic changes of complex time series data and improve the overall prediction accuracy. In addition, multi-channel adaptive Fourier decomposition can adaptively select the decomposition basis function according to the characteristics of the signal, 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.
[0077] S103, inputting the main frequency components and non-stationary features as input data into a prediction model obtained by training a long short-term memory network to obtain model output data.
[0078] In this embodiment, the prediction model is a pre-trained model built based on a long short-term memory network, whose input data is the main frequency components and non-stationary features extracted by multi-channel adaptive Fourier decomposition, and the model output data is the predicted values of multiple future time points.
[0079] The long short-term memory network is a special recursive neural network that can effectively capture the long-term and short-term dependencies in time series signals. The basic structure of the long short-term memory network includes a forget gate, an input gate, and an output gate. The basic formula includes:
[0080] f t =σ(W f ·[h t-1 ,x t ]+b f );
[0081] i t =σ(W i ·[h t-1 ,x t ]+b i );
[0082]
[0083] o t =σ(W o ·[h t-1 ,x t ]+b o );
[0084] h t =o t tanh(C t );
[0085] Among them, f t For the forget gate, it is the input gate, o t is the output gate, C t is the cell state, h t In hidden state.
[0086] In one embodiment, the process of training a prediction model constructed based on a long short-term memory network may include: obtaining a set of historical multi-channel time series signals, the historical time series signals including multiple groups of preprocessed historical multi-channel time series signals; using multi-channel adaptive Fourier decomposition to extract historical main frequency components and non-stationary features from each group of historical multi-channel time series signals, and constructing a training data set based on the features extracted from each group of historical multi-channel time series signals; using the training data set to iteratively train the prediction model constructed based on the long short-term memory network until a preset end condition is met to obtain a trained prediction model.
[0087] When training the prediction model, this embodiment uses the Fourier components extracted by multi-channel adaptive Fourier decomposition as the input of the prediction model to train the prediction model, so that the prediction model can capture the long-term and short-term dependencies in the input data. During the training process, the mean square error (MSE) can be used as the loss function, and the Adam optimizer optimizes the model parameters.
[0088] Both multi-channel adaptive Fourier decomposition and long short-term memory network can be processed in parallel on multi-channel data, which can make full use of the computing power of modern multi-core computing platforms, thereby further accelerating the training and prediction process of the model. This also makes the prediction method provided by this application very suitable for large-scale real-time data processing applications. In addition, the effective combination of multi-channel adaptive Fourier decomposition and long short-term memory network has powerful signal processing capabilities and efficient prediction performance, has broad application prospects, and can be applied to multiple fields, such as financial market prediction, weather forecast, medical data analysis, industrial equipment monitoring, etc. In addition, its versatility and efficiency enable it to provide excellent prediction results when facing different types of time series data.
[0089] S104: Determine a prediction result based on the model output data.
[0090] In one embodiment, the model output result can be directly used as the prediction result of the target prediction task.
[0091] In another embodiment, determining the prediction result based on the model output data may include: integrating the residual of the model output data and the input data to generate the prediction result.
[0092] Time series signals often contain non-stationary features, such as trends, seasonality, and random fluctuations. This embodiment integrates the predicted data output by the prediction model with the residuals of the original data, and then uses the integrated result as the prediction result. By integrating the residuals, the prediction model can better capture the non-stationary features, especially when the long short-term memory network is difficult to fully capture, thereby improving the accuracy of the final prediction result.
[0093] Specifically, generating a prediction result by integrating the residuals of the model output data and the input data may include: reconstructing the model output data and the periodic components in the input data into a time series of the original data, and adding the residuals of the model output data to obtain the prediction result.
[0094] The integration operation may specifically include reconstructing the model output data and the periodic components in the input data into a time series of the original data, and adding the residual of the model output data to obtain a prediction result.
[0095] The prediction method provided in the above embodiment will be described in detail below using a plurality of different target prediction tasks as examples.
[0096] In one embodiment, the target prediction task is time series prediction for financial markets, such as stock price prediction.
[0097] In this embodiment, a training data set based on historical stock market data can be first constructed, and the training data set contains the stock price information (opening price, closing price, highest price, lowest price) and trading volume information of multiple companies every day in the past years; then, based on the training samples in the training data set (each training sample includes the stock price information and trading volume of multiple companies every day in the past period of time), a prediction model based on a long short-term memory network is trained, and the prediction model will predict the future stock price trend of each company by learning the long-term dependency and short-term fluctuation of the time series signal. Among them, the input data of the prediction model is the main frequency components and characteristic components of the stock price information and trading volume extracted by multi-channel adaptive Fourier decomposition, and the output data is the stock price information of each company every day in the next week.
[0098] After completing the model training, the prediction model can be used for inference. Among them, the multi-channel time series signal of the target prediction task that needs to be obtained is the stock price information (opening price, closing price, highest price, lowest price) and trading volume of multiple specified companies (i.e., companies whose stock price trends need to be predicted) every day in the past period of time (such as one or more years); then use multi-channel adaptive Fourier decomposition to decompose the stock price and trading volume of each company, and extract the main frequency components and unsteady characteristics of each company. Finally, the main frequency components and unsteady characteristics of each company are input into the prediction model to obtain the stock price information of each company every day in the next week output by the prediction model.
[0099] Among them, the features extracted by multi-channel adaptive Fourier decomposition can better reflect the cyclical fluctuations of the stock market, and the prediction model constructed based on the long short-term memory network can achieve higher prediction accuracy based on these features.
[0100] In another embodiment, the target prediction task is medical signal prediction, such as electrocardiogram signal prediction.
[0101] In this embodiment, the electrocardiogram (ECG) signal data (including ECG signals of multiple leads) in the MIT-BIH arrhythmia database can be selected as a training data set, and then the prediction model is trained based on the training data set. The prediction model will predict the future waveform of the ECG signal by learning the long-term dependencies and short-term fluctuations of the time series signal. Among them, the input data of the prediction model is the main components extracted from the ECG signal, such as P wave, QRS complex and T wave, and the output data is the waveform of the ECG signal for a period of time in the future.
[0102] The prediction method provided in this embodiment performs well in identifying arrhythmias and predicting future ECG signal waveforms, and has higher accuracy and sensitivity. Among them, the multi-channel decomposition capability of the multi-channel adaptive Fourier decomposition can effectively utilize the correlation between the leads in the ECG signal, which can effectively improve the accuracy of the final prediction result.
[0103] In yet another embodiment, the target prediction task is weather data prediction.
[0104] In this embodiment, a training data set generated based on daily meteorological data of a city for many years (including data of multiple channels such as temperature, humidity, wind speed, precipitation, etc.) can be constructed first, and a prediction model can be trained based on the training data set. The prediction model will predict the weather conditions for a period of time in the future 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 non-stationary 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 indicators, such as seasonal temperature fluctuations, periodic precipitation, etc., which can reflect long-term regular information. Non-stationary features refer to short-term changes in meteorological indicators, such as information that characterizes sudden rainfall, sudden temperature drops, etc.
[0105] The prediction method provided in this embodiment can accurately predict future meteorological trends, especially under the joint action of multi-channel data (such as temperature, humidity and precipitation), the prediction accuracy can be significantly improved. In addition, the features extracted by multi-channel adaptive Fourier decomposition better capture the periodic changes and random changes in meteorological data, and combining it with the powerful time series modeling ability of long short-term memory network can make the final prediction result more reliable.
[0106] The present application also provides a prediction device based on MAFD and LSTM. In some embodiments, see Figure 2 , the device comprises:
[0107] A signal acquisition module 10 is used to acquire a multi-channel time series signal related to a target prediction task;
[0108] A feature extraction module 20 is used to extract features from the multi-channel time series signal using multi-channel adaptive Fourier decomposition to obtain main frequency components and non-stationary features;
[0109] The trend prediction module 30 is used to input the main frequency components and non-stationary features as input data into the prediction model obtained by long short-term memory network training to obtain model output data;
[0110] The prediction result generating module 40 is used to determine the prediction result based on the model output data.
[0111] In one embodiment, the feature extraction module 20 includes:
[0112] The frequency decomposition unit is used to perform frequency decomposition on the time series signal of each channel using multi-channel adaptive Fourier decomposition, and extract the main frequency components and non-stationary features in each channel.
[0113] In one embodiment, the device further comprises a model training module. The model training module comprises:
[0114] A historical data acquisition unit, used for acquiring a historical multi-channel time series signal set, wherein the historical time series signal includes a plurality of groups of pre-processed historical multi-channel time series signals;
[0115] A feature extraction unit is used to extract historical main frequency components and non-stationary features from each group of historical multi-channel time series signals using multi-channel adaptive Fourier decomposition, and to construct a training data set based on the features extracted from each group of historical multi-channel time series signals;
[0116] The training unit is used to iteratively train the prediction model built based on the long short-term memory network using the training data set until a preset end condition is met to obtain a trained prediction model.
[0117] In one embodiment, the device further includes a signal processing module. The signal processing module includes:
[0118] A raw data acquisition unit, used for acquiring raw multi-channel time series signals;
[0119] A preprocessing unit, used for preprocessing the time series signals of each channel of the original multi-channel time series signal, the preprocessing including data cleaning, standardization, filtering and noise reduction;
[0120] A data alignment unit, used for performing data alignment on the time series signals of each channel based on the time stamps of the preprocessed time series signals of each channel;
[0121] The original signal generating unit is used to obtain a multi-channel time series signal based on the aligned time series signals of each channel.
[0122] In one embodiment, when the original signal generating unit obtains a multi-channel time series signal based on the aligned time series signals of each channel, it is used to improve the signal-to-noise ratio of the aligned time series signals of each channel based on data enhancement technology to obtain a multi-channel time series signal.
[0123] In one embodiment, the prediction result generation module 40 includes:
[0124] The integration unit is used to integrate the model output data with the residual of the multi-channel time series signal to generate a prediction result.
[0125] In one embodiment, when the integration unit integrates the model output data and the residual of the multi-channel time series signal, it is used to reconstruct the model output data and the periodic components in the multi-channel time series signal into a time series of the original data, and add the residual of the model output data to obtain a prediction result.
[0126] The prediction device based on MAFD and LSTM provided in one embodiment of the present application and the prediction method based on MAFD and LSTM 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.
[0127] An embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the prediction method based on MAFD and LSTM as provided in an embodiment of the present application.
[0128] Figure 3 A specific structural block diagram of a computer device provided in an embodiment of the present application is shown, and 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, the prediction method based on MAFD and LSTM as provided in an embodiment of the present application is implemented.
[0129] 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.
[0130] 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).
[0131] The technical features of the above embodiments may be arbitrarily combined. 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.
[0132] 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 MAFD and LSTM, characterized in that: The method comprises: Obtain multi-channel time series signals related to the target prediction task; Using multi-channel adaptive Fourier decomposition to extract features of the multi-channel time series signal to obtain main frequency components and non-stationary features; Inputting the main frequency components and non-stationary features as input data into a prediction model obtained by training a long short-term memory network to obtain model output data; A prediction result is determined based on the model output data.
2. The method according to claim 1, characterized in that Using multi-channel adaptive Fourier decomposition to extract features from the multi-channel time series signal includes: Multi-channel adaptive Fourier decomposition is used to perform frequency decomposition on the time series signals of each channel to extract the main frequency components and non-stationary features in each channel.
3. The method according to claim 1, characterized in that The training process of the prediction model includes: Acquire a historical multi-channel time series signal set, wherein the historical time series signal includes a plurality of groups of pre-processed historical multi-channel time series signals; Use multi-channel adaptive Fourier decomposition to extract historical main frequency components and non-stationary features from each group of historical multi-channel time series signals, and build a training data set based on the features extracted from each group of historical multi-channel time series signals; The prediction model constructed based on the long short-term memory network is iteratively trained using the training data set until a preset end condition is met, thereby obtaining a trained prediction model.
4. The method according to claim 1, characterized in that Before acquiring the multi-channel time series signal related to the target prediction task, the method further includes: Get the original multi-channel time series signal; Preprocessing the time series signals of each channel of the original multi-channel time series signal, wherein the preprocessing includes data cleaning, standardization, filtering and noise reduction; Performing data alignment on the time series signals of each channel based on the timestamps of the preprocessed time series signals of each channel; The multi-channel time series signal is obtained based on the aligned time series signals of each channel.
5. The method according to claim 4, characterized in that The multi-channel time series signal is obtained based on the aligned time series signals of each channel, including: The signal-to-noise ratio of the aligned time series signals of each channel is improved based on the data enhancement technology to obtain the multi-channel time series signal.
6. The method according to claim 1, characterized in that Determining a prediction result based on the model output data includes: The prediction result is generated by integrating the model output data with the residual of the multi-channel time series signal.
7. The method according to claim 1, characterized in that Integrating the model output data with the residual of the multi-channel time series signal to generate a prediction result includes: The model output data and the periodic components in the multi-channel time series signal are reconstructed into a time series of original data, and the residual of the model output data is added to obtain the prediction result.
8. A prediction device based on MAFD and LSTM, characterized in that: The device comprises: A signal acquisition module is used to acquire multi-channel time series signals related to the target prediction task; A feature extraction module, used for extracting features from the multi-channel time series signal using multi-channel adaptive Fourier decomposition to obtain main frequency components and non-stationary features; A trend prediction module, used for inputting the main frequency components and non-stationary features as input data into a prediction model obtained by training a long short-term memory network to obtain model output data; The prediction result generation module is used to determine the prediction result based on the model output data.
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 MAFD and LSTM as described in 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 MAFD and LSTM as described in any one of claims 1 to 7 when executing the computer program.